Advances in Decision Sciences, Image Processing, Security and Computer Vision: International Conference on Emerging Trends in Engineering (ICETE), Vol. 2 [1st ed. 2020] 978-3-030-24317-3, 978-3-030-24318-0

Table of contents :
Front Matter ....Pages i-xxxiii
Random Walk with Clustering for Image Segmentation ( Mohebbanaaz, Madiraju Sirisha, K. Joseph Rajiv)....Pages 1-9
Design and Simulation of Microstrip Branch Line Coupler and Monopulse Comparator for Airborne Radar Applications (Annapureddy Venkata Reddy, V. G. Borkar)....Pages 10-18
RF Imagery from SAR Data Using Chirp Scaling Algorithm (P. P. Sastri, Tapas Kumar Pal, B. S. V. Prasad)....Pages 19-28
Design and Analysis of Compact Wideband Elliptical Patch Antenna (Subba Rao Chalasani, Swathi Lakshmi Boppana, Phani Rama Krishna Kuruganti)....Pages 29-35
LOS Rate Estimation Using Extended Kalman Filter (R. Kranthi Kumar, R. Sandhya, R. Laxman, A. Chandrakanth)....Pages 36-43
Design Concepts of S/X Band Feed System for 1.2 m Prime Focus Antenna (G. Rahul)....Pages 44-51
CPW Fed Antenna Inspired by a Broad Side Coupled Hexagonal SRR for X-Band Applications (A. Swetha, M. Vanidivyatha)....Pages 52-60
Pulse Compression Waveforms: Applications to Radar Systems (E. V. Suryanarayana, P. Siddaiah, Tara Dutt Bhatt)....Pages 61-68
Reduced Complexity Hybrid PAPR Reduction Schemes for Future Broadcasting Systems (Thota Sravanti, Yedukondalu Kamatham, Chandra Sekhar Paidimarry)....Pages 69-76
Speech Enhancement Through an Extended Sub-band Adaptive Filter for Nonstationary Noise Environments (G. Amjad Khan, K. E. Sreenivasa Murthy)....Pages 77-88
Design of Dayadi 1-bit CMOS Full Adder Based on Power Reduction Techniques (Pabba Sowmya, D. Lakshmaiah, J. Manga, Gunturu Sai Shankar, Desham Sai Prasad)....Pages 89-96
Size Deduced Printed Antenna for W-LAN Applications (Patturi Ravali, Samiran Chatterjee, K. Radhika Reddy, Narmala Raju, Adhimulam Rohith Kumar, Gorla Haritha et al.)....Pages 97-103
Design of 31.2/40.1667 GHz Dual Band Antenna for Future mmwave 5G Femtocell Access Point Applications (V. Harini, M. V. S. Sairam, R. Madhu)....Pages 104-111
Implementation of Master – Slave Communication for Smart Meter Using 6LOWPAN (Navya Namratha Doppala, Ameet Chavan)....Pages 112-118
Dynamic Neural Networks with Semi Empirical Model for Mobile Radio Path Loss Estimation (Bhuvaneshwari Achayalingam, Hemalatha Rallapalli, Satya Savithri Tirumala)....Pages 119-129
Throughput and Spectrum Sensing Trade-Off by Incorporating Self-interference Suppression for Full Duplex Cognitive Radio (Srilatha Madhunala, Hemalatha Rallapalli)....Pages 130-138
IOT Based Monitor and Control of Office Area Using ZYBO (Priyanka Paka, Rajani Akula)....Pages 139-146
Robust Speaker Recognition Systems with Adaptive Filter Algorithms in Real Time Under Noisy Conditions (Hema Kumar Pentapati, Madhu Tenneti)....Pages 147-154
Cognitive Radio: A Conceptual Future Radio and Spectrum Sensing Techniques - A Survey (N. Rajanish, R. M. Banakar)....Pages 155-165
Non-max Suppression for Real-Time Human Localization in Long Wavelength Infrared Region (Anuroop Mrutyunjay, Pragnya Kondrakunta, Hemalatha Rallapalli)....Pages 166-174
RF Energy Harvesting for Spectrum Management in Cognitive Radio Networks ( Suneetha, Harini, Yashaswini, Hari Haran)....Pages 175-184
Design and Development of a Broadband Planar Dipole Antenna (S. D. Ahirwar, D. Ramakrishna, V. M. Pandharipande)....Pages 185-193
Actuation System Simulation and Validation in Hardware in Loop Simulation (HILS) for an Aerospace Vehicle (M. V. K. S. Prasad, Sreehari Rao Patri, Jagannath Nayak)....Pages 194-200
A Real Time Low Cost Water Quality Progress Recording System Using Arduino Uno Board (Mohammad Mohiddin, Kedarnath Bodapally, Sravani Siramdas, Shainaz, Sai Kumar Sriramula)....Pages 201-209
Evaluation of Scheduling Algorithms on Linux OS (Dhruva R. Rinku, M. Asha Rani)....Pages 210-217
Protection Against IED (Improvised Explosive Device) a Dreaded and Fearful Weapon of Terrorist – Problems, Solutions and Challenges (G. Kumaraswamy Rao, M. Madhavi Latha)....Pages 218-225
Design of Waveform for Airborne Radar in Sea Clutter Environment (Balu Ramavathu, A. Bharathi)....Pages 226-234
Design and Development of High Performance Ceramic Packaging for Low Noise and High Gain GaAs MMIC (Ravi Gugulothu, Sangam V. Bhalke, Anant A. Naik, Lalkishore K, Ramakrishna Dasari)....Pages 235-242
Towards 5G: A Survey on Waveform Contenders (G. Shyam Kishore, Hemalatha Rallapalli)....Pages 243-250
Leakage Current Reduction Techniques Using Current Mode Logic Circuits (K. A. Jyotsna, P. Satish Kumar, B. K. Madhavi)....Pages 251-258
An Airborne Miniature Circularly Polarized Active Antenna for GPS and GLONASS Applications (Kumbha Sambasiva Rao, D. R. Jahagirdar, K. Durga Bhavani)....Pages 259-265
Efficient Obstacle Detection and Guidance System for the Blind (Haptic Shoe) (Rajani Akula, Bhagavatula Ramya Sai, Kokku Jaswitha, Molugu Sanjay Kumar, Veeramreddy Yamini)....Pages 266-271
VLSI Design and Synthesis of Reduced Power and High Speed ALU Using Reversible Gates and Vedic Multiplier (M. Dasharatha, B. Rajendra Naik, N. S. S. Reddy, Shoban Mude)....Pages 272-280
Performance Analysis of VLSI Circuits in 45 nm Technology (Sudhakar Alluri, B. Rajendra Naik, N. S. S. Reddy, M. Venkata Ramanaiah)....Pages 281-289
Prediction of Differential GPS Corrections Using AR and ARMA Models (K. Madhu Krishna, P. Naveen Kumar, R. Naraiah)....Pages 290-298
Positioning Parameters for Stand-Alone and Hybrid Modes of the Indian NavIC System: Preliminary Analysis (Devadas Kuna, Naveen Kumar Perumalla, R. Anil Kumar)....Pages 299-306
An Annular Ring Antenna with Slotted Ground Plane for Dual Band Wireless Applications (Tangalla Manoj Kumar, Namburi Randy Jonathan, Paritosh Peshwe, Srinivas Doddipalli, Ashwin Kothari)....Pages 307-313
Local Ionospheric TEC Maps for DGPS Applications (E. Upendranath Goud, K. C. T. Swamy)....Pages 314-319
Comparison of VTEC Due to IRI-2016 Model and IRNSS over Low Latitude Region (D. Kavitha, P. Naveen Kumar, K. Praveena)....Pages 320-326
An Experimental System Level Performance Analysis of Embedded Systems for GSM Application (M. Rajendra Prasad, D. Krishna Reddy)....Pages 327-334
Calculation of Voltages and Power Losses with Multiple DG Units with Change of Load (S. Vijender Reddy, M. Manjula)....Pages 335-344
Node MCU Based Power Monitoring and Smart Billing System Using IoT (K. Lova Raju, K. Yaswanth Pavankalyan, Sk. Md. Khasim, A. Naveen)....Pages 345-354
Real Time Neuro-Hysteresis Controller Implementation in Shunt Active Power Filter (Pratap Sekhar Puhan, S. D. Sandeep)....Pages 355-363
A Novel Configuration of Multi-stage Outrunner Electromagnetic Launching for Aircraft Catapult System (Sandhya Thotakura, Kondamudi Srichandan, P. Mallikarjuna Rao)....Pages 364-372
Analysis of CMV Reduction Methods for Three-Level NPC Inverter Fed Induction Motor (R. Linga Swamy, R. Somanatham)....Pages 373-382
Blaze Averter – A Cooking Sustenant for Visually Challenged People (Sai Venkata Alekhya Koppula, Suma Bindu Inumarthy, Devi Radha Sri Krishnaveni Korla, Pavani Mukkamala, Padma Vasavi Kalluru)....Pages 383-389
Symmetrical and Asymmetrical Fault Response of DC Link in AC-DC Interface HVDC System (Maanasa Devi Atluri, N. V. L. H. Madhuri Ramineedi, Revathi Devarajula)....Pages 390-397
Condition Assessment of Composite Insulator Removed from Service (B. Sravanthi, K. A. Aravind, Pradeep Nirgude, M. Manjula, V. Kamaraju)....Pages 398-405
Power Quality Improvement of Weak Hybrid PEMFC and SCIG Grid Using UPQC (G. Mallesham, C. H. Siva Kumar)....Pages 406-413
Comparison of State Estimation Process on Transmission and Distribution Systems (M. S. N. G. Sarada Devi, G. Yesuratnam)....Pages 414-423
E-Mobility in India: Plug-in HEVs and EVs (G. Sree Lakshmi, Vimala Devi, G. Divya, G. Sravani)....Pages 424-434
Unified Simulation Test Facility for Aerospace Vehicles (K. Rama Rao, B. Mangu, Manchem Rama Prasanna)....Pages 435-442
Machine Learning Based Prediction Model for Monitoring Power System Security (N. Srilatha, B. Priyanka Rathod, G. Yesuratnam)....Pages 443-450
In-depth Automation and Structural Analysis to Enhance the Power Distribution Reliability of an Urban Neighbourhood (Ranjith Kumar Mittapelli, E. Tharun Sai, Navya Pragathi)....Pages 451-461
Implementation of Ant-Lion Optimization Algorithm in Energy Management Problem and Comparison (P. S. Preetha, Ashok Kusagur)....Pages 462-469
Modulated Frequency Triangular Carrier Based Space Vector PWM Technique for Spreading Induction Motor Acoustic Noise Spectrum (Sadanandam Perumandla, Poonam Upadhyay, A. Jayalaxmi, Jaya Prakash Nasam)....Pages 470-480
Hardware In Loop Simulation of Advanced Aerospace Vehicle (A. Shiva Krishna Prasad, N. Susheela)....Pages 481-489
Analysis of Power Transmission System Using Quadrature Booster (Nagasrinivasulu Mallepogu, Suresh Babu Daram, Kamal Kumar Usurupati, Jayachandra S., Sairam Seshapalli, S. Venkataramu P.)....Pages 490-497
Backtracking Search Optimization Algorithm Based MPPT Technique for Solar PV System (Mounika Sriram, K. Ravindra)....Pages 498-506
A New Topology of Interline Unified Power-Quality Conditioner for Multi Feeder System (Surya Prakash Thota, Satish Kumar Peddapelli)....Pages 507-519
Adaptive NMPC Controller for Two-Link Planar Manipulator with Parameter Estimator (B. Raja Gopal Redy, N. Karuppiah, Dubbaka Vinay Kumar Goud, Samudrala Prashant Ganesh)....Pages 520-527
Solving the Complexity of Geometry Friends by Using Artificial Intelligence (D. Marlene Grace Verghese, Suresh Bandi, G. Jacob Jayaraj)....Pages 528-533
An Efficient Classification Technique for Text Mining and Applications (Vb. Narasimha, Sujatha)....Pages 534-544
Clinical Big Data Predictive Analytics Transforming Healthcare: - An Integrated Framework for Promise Towards Value Based Healthcare (Tawseef Ahmad Naqishbandi, N. Ayyanathan)....Pages 545-561
A Study of Different Techniques in Educational Data Mining (Nadia Anjum, Srinivasu Badugu)....Pages 562-571
A Study on Overlapping Community Detection for Multimedia Social Network (Sabah Fatima, Srivinasu Badugu)....Pages 572-578
A Review on Different Question Answering System Approaches (Tahseen Sultana, Srinivasu Badugu)....Pages 579-586
A Study of Malicious URL Detection Using Machine Learning and Heuristic Approaches (Aliya Begum, Srinivasu Badugu)....Pages 587-597
A Review on Network Intrusion Detection System Using Machine Learning (T. Rupa Devi, Srinivasu Badugu)....Pages 598-607
Review on Facial Expression Recognition System Using Machine Learning Techniques (Amreen Fathima, K. Vaidehi)....Pages 608-618
Document Clustering Using Different Unsupervised Learning Approaches: A Survey (Munazza Afreen, Srinivasu Badugu)....Pages 619-629
A Study of Liver Disease Classification Using Data Mining and Machine Learning Algorithms (Hajera Subhani, Srinivasu Badugu)....Pages 630-640
A Study of Routing Protocols for Energy Conservation in Manets (Aqsa Parveen, Y. V. S. Sai Pragathi)....Pages 641-647
Credit Risk Valuation Using an Efficient Machine Learning Algorithm (Ramya Sri Kovvuri, Ramesh Cheripelli)....Pages 648-657
Handwritten Mathematical Symbol Recognition Using Machine Learning Techniques: Review (Syeda Aliya Firdaus, K. Vaidehi)....Pages 658-671
Automatic Water Level Detection Using IoT (Ch. V. S. S. Mahalakshmi, B. Mridula, D. Shravani)....Pages 672-677
Optimal Scheduling of Tasks in Cloud Computing Using Hybrid Firefly-Genetic Algorithm (Aravind Rajagopalan, Devesh R. Modale, Radha Senthilkumar)....Pages 678-687
A Novel Approach for Rice Yield Prediction in Andhra Pradesh (Nagesh Vadaparthi, G. Surya Tejaswini, N. B. S. Pallavi)....Pages 688-692
Representation Techniques that Best Followed for Semantic Web - Web Mining (K. Vaishali, Sriramula Nagaprasad)....Pages 693-699
Isolated Health Surveillance System Through IoT Using Raspberry Pi (Sumayya Afreen, Asma Begum, G. Saraswathi, Ayesha Nuzha)....Pages 700-706
Design and Implementation of RPL in Internet of Things (M. V. R. Jyothisree, S. Sreekanth)....Pages 707-718
Detection and Tracking of Text from Video Using MSER and SIFT (M. Manasa Devi, M. Seetha, S. Viswanada Raju, D. Srinivasa Rao)....Pages 719-727
Blockchain Enabled Smart Learning Environment Framework (G. R. Anil, Salman Abdul Moiz)....Pages 728-740
Global Snapshot of a Large Wireless Sensor Network (Surabhi Sharma, T. P. Sharma, Kavitha Kadarala)....Pages 741-752
Back Matter ....Pages 753-755

Citation preview

Learning and Analytics in Intelligent Systems 4

Suresh Chandra Satapathy · K. Srujan Raju · K. Shyamala · D. Rama Krishna · Margarita N. Favorskaya Editors

Advances in Decision Sciences, Image Processing, Security and Computer Vision International Conference on Emerging Trends in Engineering (ICETE), Vol. 2

Learning and Analytics in Intelligent Systems Volume 4

Series Editors George A. Tsihrintzis, University of Piraeus, Piraeus, Greece Maria Virvou, University of Piraeus, Piraeus, Greece Lakhmi C. Jain, Faculty of Engineering and Information Technology, Centre for Artificial Intelligence, University of Technology Sydney, NSW, Australia; University of Canberra, Canberra, ACT, Australia; KES International, Shoreham-by-Sea, UK; Liverpool Hope University, Liverpool, UK

The main aim of the series is to make available a publication of books in hard copy form and soft copy form on all aspects of learning, analytics and advanced intelligent systems and related technologies. The mentioned disciplines are strongly related and complement one another significantly. Thus, the series encourages cross-fertilization highlighting research and knowledge of common interest. The series allows a unified/integrated approach to themes and topics in these scientific disciplines which will result in significant cross-fertilization and research dissemination. To maximize dissemination of research results and knowledge in these disciplines, the series publishes edited books, monographs, handbooks, textbooks and conference proceedings.

More information about this series at http://www.springer.com/series/16172

Suresh Chandra Satapathy K. Srujan Raju K. Shyamala D. Rama Krishna Margarita N. Favorskaya •

•

•

•

Editors

Advances in Decision Sciences, Image Processing, Security and Computer Vision International Conference on Emerging Trends in Engineering (ICETE), Vol. 2

123

Editors Suresh Chandra Satapathy School of Computer Engineering Kalinga Institute of Industrial Technology (KIIT) Deemed to be University Bhubaneswar, Odisha, India K. Shyamala Department of CSE Osmania University, University College of Engineering Hyderabad, Telangana, India

K. Srujan Raju Department of CSE CMR Technical Campus Hyderabad, Telangana, India D. Rama Krishna Department of ECE Osmania University, University College of Engineering Hyderabad, Telangana, India

Margarita N. Favorskaya Institute of Informatics and Telecommunications Reshetnev Siberian State University of Science and Technology Krasnoyarsk, Russia

ISSN 2662-3447 ISSN 2662-3455 (electronic) Learning and Analytics in Intelligent Systems ISBN 978-3-030-24317-3 ISBN 978-3-030-24318-0 (eBook) https://doi.org/10.1007/978-3-030-24318-0 © Springer Nature Switzerland AG 2020 This work is subject to copyright. All rights are reserved by the Publisher, whether the whole or part of the material is concerned, specifically the rights of translation, reprinting, reuse of illustrations, recitation, broadcasting, reproduction on microfilms or in any other physical way, and transmission or information storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar methodology now known or hereafter developed. The use of general descriptive names, registered names, trademarks, service marks, etc. in this publication does not imply, even in the absence of a specific statement, that such names are exempt from the relevant protective laws and regulations and therefore free for general use. The publisher, the authors and the editors are safe to assume that the advice and information in this book are believed to be true and accurate at the date of publication. Neither the publisher nor the authors or the editors give a warranty, expressed or implied, with respect to the material contained herein or for any errors or omissions that may have been made. The publisher remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. This Springer imprint is published by the registered company Springer Nature Switzerland AG The registered company address is: Gewerbestrasse 11, 6330 Cham, Switzerland

Dedicated to Our Alma Mater & Eminent Professors who taught us for their inspiring vision, unwavering conviction and tireless efforts that have resulted in nurturing hundreds of eminent global citizens and effective human beings. “Once an Osmanian Always an Osmanian”

University College of Engineering, Osmania University, Hyderabad, India

University College of Engineering (UCE) has the distinction of being the oldest and the biggest among the engineering colleges of the State of Telangana, India. It was established in the year 1929, eleven years after the formation of Osmania University. The college was the sixth engineering college to be established in the whole of British India. The college moved to its present permanent building in the year 1947. Today, it is the biggest among the campus colleges of Osmania University. The golden jubilee of the college was celebrated in 1979, the diamond jubilee in 1989 and the platinum jubilee in 2004. The college was made autonomous in 1994. University Grants Commission of India conferred autonomy status to the college for a period of 6 years (2016–2017 to 2021–2022). The college offers four-year engineering degree courses leading to the award of Bachelor of Engineering (B.E.) in biomedical engineering, civil engineering, computer science and engineering, electrical and electronics engineering, electronics and communications engineering and mechanical engineering. The college also offers graduate programs and Ph.D. in the various branches of engineering. As of today, there is a yearly intake of 320 undergraduate students (full-time) and 290 postgraduate students (full-time and part-time). There are 143 teaching staff members, including 40 professors. The UG programs offered have been accredited by the National Board of Accreditation, New Delhi. Osmania University is accredited by NAAC with “A+” Grade. UCE, OU, is the first engineering college to get ISO 9001 Certification in Telangana State. University College of Engineering was awarded the Best Engineering College by Indian Society for Technical Education (Telangana) in the year 2010. UCE, OU, was adjudged as the Best Engineering College in the country for the academic year 2003–2004 by Indian Society for Technical Education, New Delhi, and by Star News for the years 2010–2011 and 2011–2012. The college has successfully completed the Technical Education Quality Improvement Programme (TEQIP-I) under the World Bank financial assistance of Rs. 15.48 crores during the period 2003–2008. The outcome of the project has resulted in: (i) increase in pass percentage of UG/PG students, (ii) enhancement of research publications of staff by threefolds, (iii) introduction of six PG programs in vii

viii

University College of Engineering, Osmania University, Hyderabad, India

niche areas, (iv) introduction of credit-based system and (v) substantial increase in internal revenue generation. The college has successfully completed Phase II of TEQIP program with a financial assistance of Rs. 12.5 crores and additional grant of 5 crores under the best-performing institution category. Recently, the college has been approved as a minor center under QIP for full-time Ph.D. programs. The college has been selected for TEQIP Phase III twinning program with a financial assistance Rs. 7 crores. The college has been granted “Visvesvaraya Ph.D. Scheme for Electronics and IT” for full-time Ph.D. program. The GIAN program of MHRD has sanctioned 7 programs in specialized research area to the college. The college has been ranked 80 in NIRF Engineering College Ranking Survey by MHRD Survey, New Delhi, India, for the year 2017–2018.

Alumni Association University College of Engineering, Osmania University, Hyderabad, India

Once, University College of Engineering was declared autonomous, the idea of having Alumni Association, whose membership would include all past students of the college and present or past faculty members of the college, gained momentum. Under the dynamic leadership of then Principal, Prof. D. C. Reddy, the first get-together was held on Saturday the July 3, 1996. After the revival of the Association in 2015, the two subsequent Executive Committees under the leadership of Er. Rammohan Rao, Er. P. Ram Reddy and Dr. D. Vijay Kumar with the support of patrons, fellow members, learned faculty and students have been set out to fulfill the objectives of Alumni Association. The Association is a not-for-profit organization and works with the staff and students of University College of Engineering, and the objectives are: • Provide a platform for the alumni to connect with each other for the exchange of information and ideas and communicate their accomplishments, interests and concerns. • Foster alumni pride and enhance the reputation of the university and OUCE in particular. • Enrich the emotional bondage among the students, alumni and faculty. • Extend maximum help to the college in the placement and internship of students in reputed organizations. • Recognize alumni for their significant contributions to education. • Propose and execute special projects: buildings, technical projects, seminars, conferences, etc. • Support poor/economically backward students financially by floating scholarships. • Institute awards for meritorious performance for students. • Institute awards for the alumni for their contribution to the college and the society. • Inspire and invoke the spirit of innovation among the students leading to finding technical solutions to the problems of the society leading to patents to students and the college.

ix

x

Alumni Association University College of Engineering

In the past four years, the Executive Body set out to execute the above objectives by taking up many initiatives like conducting global alumni meets, alumni talks, funding student innovation, patent and research, facilitating student internships, industry interactions and career development programs, support for student clubs and other activities, facilitating in setting up the technology business incubator, etc. To further the objectives of the Association to support the faculty and research scholars, the Association has organized the First International Conference on Emerging Trends in Engineering under its aegis.

Foreword

Alumni Association is constantly setting up new benchmarks with every passing year with a lot of good work done toward furthering the Association goals and giving back to the alma mater. One of the key focus areas has been to bridge the industry academia gap, thereby promoting cutting-edge skill development and research, thereby enabling the university to be a hub of innovation. This publication is an outcome of the First International Conference on Emerging Trends in Engineering (ICETE). As part of the initiatives of Alumni Association, the conference was organized to enhance the information exchange of theoretical research/practical advancements at national and international levels in key fields of engineering and to promote professional interaction among students, scholars, researchers, educators, professionals from industries and other groups to share latest findings in their respective fields toward sustainable developments. The entire organizing team has worked hard over the last few months in putting together the complete structure for the event and coordinating with all the eminent speakers across the globe to ensure that the 2-day conference brings together the best minds in the industry to come together and share their valuable insights with the entire fraternity. We are honored to have eminent speakers grace the conference this year. We received 619 papers from about more than 100 institutions/organizations in 14 countries. The papers have gone through a rigorous evaluation process, and the best papers have been selected for presenting on the days of the conference. Only, the presented and approved papers have come for publishing. I want to thank the Technical Program Committee for bringing together research scholars from diverse background and working tirelessly in picking the final list and bringing out this publication. With a rich history of over 100 years producing world-class students and alumni who have made a mark all over the world, we aim to continue the tradition by hosting such world-class conferences and live up to the expectations of our alma mater. April 2019

D. Vijay Kumar

xi

Preface

This book constitutes the thoroughly refereed post-conference proceedings of the First International Conference on Emerging Trends in Engineering (ICETE), held at University College of Engineering and organized by Alumni Association, University College of Engineering, Osmania University, Hyderabad, India, on March 22–23, 2019. The aim of this conference is to enhance the information exchange of theoretical research/practical advancements at national and international levels in the fields of biomedical engineering, civil engineering, computer science engineering, electrical engineering, electronics and communication engineering, mechanical engineering and mining engineering. This encourages and promotes professional interaction among students, scholars, researchers, educators, professionals from industries and other groups to share latest findings in their respective fields toward sustainable developments. The refereed conference proceedings of the ICETE are published in three volumes covering seven streams, i.e. biomedical, civil, computer science, electrical and electronics, electronics and communication, mechanical and mining engineering. Out of 619 paper submissions from about fourteen countries in seven streams of engineering, only 214 papers are being published after reviewing thoroughly; this volume 2 under the theme “Advances in Decision Sciences, Image Processing, Security and Computer Vision - International Conference on Emerging Trends in Engineering (ICETE)” comprises of the comprehensive state-of-the-art technical contributions in the areas of electronics and communication engineering and electrical and electronics engineering. Major topics of these research papers include latest findings in the respective fields toward sustainable developments including signal processing and communications, GNSS and VLSI, microwave and antennas, signal, speech and image processing, power systems and power electronics. Margarita N. Favorskaya Suresh Chandra Satapathy K. Shyamala D. Rama Krishna K. Srujan Raju xiii

Acknowledgements

We thank all the authors for their contributions and timely response. We also thank all the reviewers who read the papers and made valuable suggestions for improvement. We would like to thank Prof. S. Ramachandram, Vice-Chancellor, Osmania University, and Prof. M. Kumar, Principal, University College of Engineering, for having faith in us. Dr. D. Rama Krishna and Prof. K, Shyamala of UCE, OU, for leading from the front; the TPC team, for pulling off a brilliant job; Heads of all departments and all learned faculty, for all the support. Also, last but not the least, we convey our thanks to all the research scholars without whose relentless slogging this conference and publication would not have seen light. We thank our sponsors Power Grid Corporation of India Ltd., Defence Research and Development Organization (DRDO), CCL Products (India) Ltd., The Singareni Collieries Company Ltd., TEQIP-III and all other financial contributors. We extend our thanks to all the Executive Body members of Alumni Association for their support and Sri. R. V. Rammohan Rao for the support when needed. Finally, we thank the Springer team comprising Prof. Suresh Chandra Satapathy, Prof. K. Srujan Raju and Dr. M. Ramakrishna Murthy for guiding and helping us throughout. April 2019

D. Vijay Kumar

xv

ICETE Organizing Committee

Chief Patron S. Ramachandram (Vice-chancellor)

Osmania University, Hyderabad, India

Patrons Kumar Molugaram (Principal) P. Laxminarayana (Dean) D. C. Reddy (Former Vice-chancellor) D. N. Reddy (Former Vice-chancellor) R. V. Rammohan Rao (Past President)

University College of Engineering (A), Osmania University, Hyderabad, India Faculty of Engineering, Osmania University, Hyderabad, India Osmania University, Hyderabad, India Jawaharlal Nehru Technological University, Hyderabad, India Alumni Association, University College of Engineering (A), Osmania University, Hyderabad, India

Chairpersons P. Ram Reddy (President) P. V. N. Prasad

Alumni Association, University College of Engineering (A), Osmania University, Hyderabad, India Department of Electrical Engineering, University College of Engineering (A), Osmania University, Hyderabad, India

xvii

xviii

ICETE Organizing Committee

Conveners D. Vijay Kumar (General Secretary) D. Rama Krishna

Alumni Association, University College of Engineering (A), Osmania University, Hyderabad, India Department of Electronics and Communication Engineering, University College of Engineering (A), Osmania University, Hyderabad, India

Publication Committee Suresh Chandra Satapathy (Chair)

Kumar Molugaram (Co-chair, Principal) K. Srujan Raju (Co-chair) Sriram Venkatesh

K. Shyamala

D. Vijay Kumar (General Secretary) D. Rama Krishna

School of Computer Engineering, Kalinga Institute of Industrial Technology (KIIT), Deemed to be University, Bhubaneswar, Odisha University College of Engineering, Osmania University, Hyderabad, Telangana Department of CSE, CMR Technical Campus, Hyderabad, Telangana Department of Mechanical Engineering, University College of Engineering, Osmania University, Hyderabad Department of Computer Science and Engineering, University College of Engineering, Osmania University, Hyderabad Alumni Association, University College of Engineering (A), Osmania University, Hyderabad, India Department of Electronics and Communication Engineering, University College of Engineering, Osmania University, Hyderabad

International Advisory Committee J. N. Reddy Ramulu Mamidala Chandra Kambhamettu P. Nageswara Rao Suman Das Tariq Muneer Rao S. Govindaraju Nitin K. Tripathi Srinivasulu Ale

Texas A&M University, USA University of Washington, USA University of Delaware, USA University of Northern Iowa, USA Georgia Institute of Technology, USA Edinburgh Napier University, Edinburgh, UK Purdue University, Indiana, USA Asian Institute of Technology, Bangkok The University of Texas, Texas, USA

ICETE Organizing Committee

Prasad Enjeti Akshay K. Rathore Sheldon Williamson Malcolm McCulloch Bimal K. Bose Manos Varvarigos Vijay Vittal Sudhakar M. Reddy Vishnu Pendyala Shantanu Narayen

xix

Texas A&M University, Texas, USA Concordia University, Canada University of Ontario, Canada University of Oxford, UK University of Tennessee, USA Monash University, Australia Arizona University, USA University of Iowa, USA CISCO Systems, USA CEO, Adobe Systems, USA

National Advisory Committee A. Venugopal Reddy (Vice-chancellor) V. M. Pandharipande (Former Vice-chancellor) Rameshwar Rao (Former Vice-chancellor) A. K. Tiwari (Director) K. V. S. Hari Sameen Fatima (Former Principal) N. K. Kishore D. Nagesh Kumar B. G. Fernandes Dinesh Bhatia G. Rameshwar R (Chairman) A. K. Singh (OS and Director) J. V. R. Sagar (Director) John D’Souza Arvind Tiwari B. H. V. S. Narayana Murthy (OS and Director) R. Soundara Rajan (Senior Project Manager) Bathini Srinivas

JNTUH, Hyderabad BAMU, Aurangabad JNTUH, Hyderabad CARE Foundation, University of Hyderabad, Hyderabad IISc, Bangalore UCE, OU IIT Kharagpur IISc, Bangalore IIT Bombay NEHU, Shillong, India Institute of Engineering, Hyderabad Chapter DLRL ANURAG NITK Surathkal GE-GRC, JFWTC RCI BDL MathWorks, India

xx

ICETE Organizing Committee

Finance Committee Sriram Venkatesh A. Krishnaiah P. Ramesh Babu V. Bhikshma G. Mallesham M. Malini B. Rajendra Naik V. Uma Maheshwar P. Naveen Kumar D. N. Prasad (Advisor (Coal)) M. Shyam Prasad Reddy (General Secretary) T. Venkatesam (Superintendent Engineer (Retd.)) M. S. Venkatramayya Satish Naik R. Thomas Syed Basharath Ali P. Narotham Reddy

ME, UCE, OU ME, UCE, OU ME, UCE, OU CE, UCE, OU EE, UCE, OU BME, UCE, OU ECE, UCE, OU ME, UCE, OU ECE, UCE, OU SCCL TREA AA UCE, OU

Mining AA UCE, AA UCE, AA UCE, AA UCE,

OU OU OU OU

Organizing Committee E. Vidya Sagar (Vice-principal) K. Shyamala P. Chandra Sekhar M. Gopal Naik P. Usha Sri M. Venkateswara Rao M. V. Ramana Rao G. Yesuratnam P. Raja Sekhar B. Mangu M. Chandrashekhar Reddy Narsimhulu Sanke M. A. Hameed B. Sujatha L. Nirmala Devi N. Susheela S. Prasanna R. Rajender G. Narender P. Koti Lakshmi

UCE, OU CSE, UCE, OU ECE, OU CE, UCE, OU ME, UCE, OU BME, UCE, OU EED, UCE, OU EED, UCE, OU CE, UCE, OU EED, UCE, OU ME, UCE, OU ME, UCE, OU CSE, UCE, OU CSE, UCE, OU ECE, UCE, OU EED, UCE, OU CE, UCE, OU CE, UCE, OU ME, UCE, OU ECE, UCE, OU

ICETE Organizing Committee

M. Srinivas B. Sirisha B. Ramana Naik C. V. Raghava (Chairman) P. Lakshman Rao (President) P. Kishore (Secretary) K. J. Amarnath J. V. Dattatreyulu Raikoti Anand Srinivas K. Praveen Dorna K. Chakradhar Pradeep Kumar Nimma A. Thara Singh Prasanth Kumar Manchikatla

xxi

BME, UCE, OU EE, UCE, OU Alumnus CVR College of Engineering, Hyderabad OUECE Association OUECE Association Mining B.E (Mining) Alumnus AA, UCE, OU AA, UCE, OU AA, UCE, OU Alumnus Alumnus

Technical Committee K. Shyamala P. V. Sudha M. Manjula B. Mangu P. Satish Kumar J. Upendar M. Malini D. Suman K. L. Radhika K. Shashikanth L. Siva Rama Krishna E. Madhusudan Raju R. Hemalatha M. Shyamsunder

CSE, UCE, OU CSE, UCE, OU EED, UCE, OU EED, UCE, OU EED, UCE, OU EED, UCE, OU BME, UCE, OU BME, UCE, OU CE, UCE, OU CE, UCE, OU ME, UCE, OU ME, UCE, OU ECE, UCE, OU ECE, UCE, OU

Supported and Strengthened By J. Suman (Research Scholar) D. Sai Kumar (Research Scholar) P. Raveendra Babu (Research Scholar) G. Shyam Kishore (Research Scholar) Jaya Prakash (Research Scholar)

CSE CSE ECE ECE EEE

Contents

Random Walk with Clustering for Image Segmentation . . . . . . . . . . . . Mohebbanaaz, Madiraju Sirisha, and K. Joseph Rajiv Design and Simulation of Microstrip Branch Line Coupler and Monopulse Comparator for Airborne Radar Applications . . . . . . . Annapureddy Venkata Reddy and V. G. Borkar

1

10

RF Imagery from SAR Data Using Chirp Scaling Algorithm . . . . . . . . P. P. Sastri, Tapas Kumar Pal, and B. S. V. Prasad

19

Design and Analysis of Compact Wideband Elliptical Patch Antenna . . . Subba Rao Chalasani, Swathi Lakshmi Boppana, and Phani Rama Krishna Kuruganti

29

LOS Rate Estimation Using Extended Kalman Filter . . . . . . . . . . . . . . R. Kranthi Kumar, R. Sandhya, R. Laxman, and A. Chandrakanth

36

Design Concepts of S/X Band Feed System for 1.2 m Prime Focus Antenna . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . G. Rahul

44

CPW Fed Antenna Inspired by a Broad Side Coupled Hexagonal SRR for X-Band Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . A. Swetha and M. Vanidivyatha

52

Pulse Compression Waveforms: Applications to Radar Systems . . . . . . E. V. Suryanarayana, P. Siddaiah, and Tara Dutt Bhatt

61

Reduced Complexity Hybrid PAPR Reduction Schemes for Future Broadcasting Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Thota Sravanti, Yedukondalu Kamatham, and Chandra Sekhar Paidimarry

69

Speech Enhancement Through an Extended Sub-band Adaptive Filter for Nonstationary Noise Environments . . . . . . . . . . . . . . . . . . . . . G. Amjad Khan and K. E. Sreenivasa Murthy

77

xxiii

xxiv

Contents

Design of Dayadi 1-bit CMOS Full Adder Based on Power Reduction Techniques . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Pabba Sowmya, D. Lakshmaiah, J. Manga, Gunturu Sai Shankar, and Desham Sai Prasad Size Deduced Printed Antenna for W-LAN Applications . . . . . . . . . . . . Patturi Ravali, Samiran Chatterjee, K. Radhika Reddy, Narmala Raju, Adhimulam Rohith Kumar, Gorla Haritha, Gantla Sai Kumar Reddy, and Akula Naresh

89

97

Design of 31.2/40.1667 GHz Dual Band Antenna for Future mmwave 5G Femtocell Access Point Applications . . . . . . . . . . . . . . . . . 104 V. Harini, M. V. S. Sairam, and R. Madhu Implementation of Master – Slave Communication for Smart Meter Using 6LOWPAN . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112 Navya Namratha Doppala and Ameet Chavan Dynamic Neural Networks with Semi Empirical Model for Mobile Radio Path Loss Estimation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119 Bhuvaneshwari Achayalingam, Hemalatha Rallapalli, and Satya Savithri Tirumala Throughput and Spectrum Sensing Trade-Off by Incorporating Self-interference Suppression for Full Duplex Cognitive Radio . . . . . . . 130 Srilatha Madhunala and Hemalatha Rallapalli IOT Based Monitor and Control of Office Area Using ZYBO . . . . . . . . 139 Priyanka Paka and Rajani Akula Robust Speaker Recognition Systems with Adaptive Filter Algorithms in Real Time Under Noisy Conditions . . . . . . . . . . . . . . . . . 147 Hema Kumar Pentapati and Madhu Tenneti Cognitive Radio: A Conceptual Future Radio and Spectrum Sensing Techniques - A Survey . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 155 N. Rajanish and R. M. Banakar Non-max Suppression for Real-Time Human Localization in Long Wavelength Infrared Region . . . . . . . . . . . . . . . . . . . . . . . . . . . 166 Anuroop Mrutyunjay, Pragnya Kondrakunta, and Hemalatha Rallapalli RF Energy Harvesting for Spectrum Management in Cognitive Radio Networks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 175 Suneetha, Harini, Yashaswini, and Hari Haran Design and Development of a Broadband Planar Dipole Antenna . . . . . 185 S. D. Ahirwar, D. Ramakrishna, and V. M. Pandharipande

Contents

xxv

Actuation System Simulation and Validation in Hardware in Loop Simulation (HILS) for an Aerospace Vehicle . . . . . . . . . . . . . . . . . . . . . 194 M. V. K. S. Prasad, Sreehari Rao Patri, and Jagannath Nayak A Real Time Low Cost Water Quality Progress Recording System Using Arduino Uno Board . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 201 Mohammad Mohiddin, Kedarnath Bodapally, Sravani Siramdas, Shainaz, and Sai Kumar Sriramula Evaluation of Scheduling Algorithms on Linux OS . . . . . . . . . . . . . . . . 210 Dhruva R. Rinku and M. Asha Rani Protection Against IED (Improvised Explosive Device) a Dreaded and Fearful Weapon of Terrorist – Problems, Solutions and Challenges . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 218 G. Kumaraswamy Rao and M. Madhavi Latha Design of Waveform for Airborne Radar in Sea Clutter Environment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 226 Balu Ramavathu and A. Bharathi Design and Development of High Performance Ceramic Packaging for Low Noise and High Gain GaAs MMIC . . . . . . . . . . . . . . . . . . . . . 235 Ravi Gugulothu, Sangam V. Bhalke, Anant A. Naik, Lalkishore K, and Ramakrishna Dasari Towards 5G: A Survey on Waveform Contenders . . . . . . . . . . . . . . . . . 243 G. Shyam Kishore and Hemalatha Rallapalli Leakage Current Reduction Techniques Using Current Mode Logic Circuits . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 251 K. A. Jyotsna, P. Satish Kumar, and B. K. Madhavi An Airborne Miniature Circularly Polarized Active Antenna for GPS and GLONASS Applications . . . . . . . . . . . . . . . . . . . . . . . . . . 259 Kumbha Sambasiva Rao, D. R. Jahagirdar, and K. Durga Bhavani Efficient Obstacle Detection and Guidance System for the Blind (Haptic Shoe) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 266 Rajani Akula, Bhagavatula Ramya Sai, Kokku Jaswitha, Molugu Sanjay Kumar, and Veeramreddy Yamini VLSI Design and Synthesis of Reduced Power and High Speed ALU Using Reversible Gates and Vedic Multiplier . . . . . . . . . . . . . . . . 272 M. Dasharatha, B. Rajendra Naik, N. S. S. Reddy, and Shoban Mude Performance Analysis of VLSI Circuits in 45 nm Technology . . . . . . . . 281 Sudhakar Alluri, B. Rajendra Naik, N. S. S. Reddy, and M. Venkata Ramanaiah

xxvi

Contents

Prediction of Differential GPS Corrections Using AR and ARMA Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 290 K. Madhu Krishna, P. Naveen Kumar, and R. Naraiah Positioning Parameters for Stand-Alone and Hybrid Modes of the Indian NavIC System: Preliminary Analysis . . . . . . . . . . . . . . . . 299 Devadas Kuna, Naveen Kumar Perumalla, and R. Anil Kumar An Annular Ring Antenna with Slotted Ground Plane for Dual Band Wireless Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 307 Tangalla Manoj Kumar, Namburi Randy Jonathan, Paritosh Peshwe, Srinivas Doddipalli, and Ashwin Kothari Local Ionospheric TEC Maps for DGPS Applications . . . . . . . . . . . . . . 314 E. Upendranath Goud and K. C. T. Swamy Comparison of VTEC Due to IRI-2016 Model and IRNSS over Low Latitude Region . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 320 D. Kavitha, P. Naveen Kumar, and K. Praveena An Experimental System Level Performance Analysis of Embedded Systems for GSM Application . . . . . . . . . . . . . . . . . . . . . 327 M. Rajendra Prasad and D. Krishna Reddy Calculation of Voltages and Power Losses with Multiple DG Units with Change of Load . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 335 S. Vijender Reddy and M. Manjula Node MCU Based Power Monitoring and Smart Billing System Using IoT . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 345 K. Lova Raju, K. Yaswanth Pavankalyan, Sk. Md. Khasim, and A. Naveen Real Time Neuro-Hysteresis Controller Implementation in Shunt Active Power Filter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 355 Pratap Sekhar Puhan and S. D. Sandeep A Novel Configuration of Multi-stage Outrunner Electromagnetic Launching for Aircraft Catapult System . . . . . . . . . . . . . . . . . . . . . . . . 364 Sandhya Thotakura, Kondamudi Srichandan, and P. Mallikarjuna Rao Analysis of CMV Reduction Methods for Three-Level NPC Inverter Fed Induction Motor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 373 R. Linga Swamy and R. Somanatham Blaze Averter – A Cooking Sustenant for Visually Challenged People . . . 383 Sai Venkata Alekhya Koppula, Suma Bindu Inumarthy, Devi Radha Sri Krishnaveni Korla, Pavani Mukkamala, and Padma Vasavi Kalluru

Contents

xxvii

Symmetrical and Asymmetrical Fault Response of DC Link in AC-DC Interface HVDC System . . . . . . . . . . . . . . . . . . . . . . . . . . . . 390 Maanasa Devi Atluri, N. V. L. H. Madhuri Ramineedi, and Revathi Devarajula Condition Assessment of Composite Insulator Removed from Service . . . 398 B. Sravanthi, K. A. Aravind, Pradeep Nirgude, M. Manjula, and V. Kamaraju Power Quality Improvement of Weak Hybrid PEMFC and SCIG Grid Using UPQC . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 406 G. Mallesham and C. H. Siva Kumar Comparison of State Estimation Process on Transmission and Distribution Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 414 M. S. N. G. Sarada Devi and G. Yesuratnam E-Mobility in India: Plug-in HEVs and EVs . . . . . . . . . . . . . . . . . . . . . 424 G. Sree Lakshmi, Vimala Devi, G. Divya, and G. Sravani Unified Simulation Test Facility for Aerospace Vehicles . . . . . . . . . . . . . 435 K. Rama Rao, B. Mangu, and Manchem Rama Prasanna Machine Learning Based Prediction Model for Monitoring Power System Security . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 443 N. Srilatha, B. Priyanka Rathod, and G. Yesuratnam In-depth Automation and Structural Analysis to Enhance the Power Distribution Reliability of an Urban Neighbourhood . . . . . . . 451 Ranjith Kumar Mittapelli, E. Tharun Sai, and Navya Pragathi Implementation of Ant-Lion Optimization Algorithm in Energy Management Problem and Comparison . . . . . . . . . . . . . . . . . . . . . . . . . 462 P. S. Preetha and Ashok Kusagur Modulated Frequency Triangular Carrier Based Space Vector PWM Technique for Spreading Induction Motor Acoustic Noise Spectrum . . . 470 Sadanandam Perumandla, Poonam Upadhyay, A. Jayalaxmi, and Jaya Prakash Nasam Hardware In Loop Simulation of Advanced Aerospace Vehicle . . . . . . . 481 A. Shiva Krishna Prasad and N. Susheela Analysis of Power Transmission System Using Quadrature Booster . . . 490 Nagasrinivasulu Mallepogu, Suresh Babu Daram, Kamal Kumar Usurupati, Jayachandra S., Sairam Seshapalli, and S. Venkataramu P.

xxviii

Contents

Backtracking Search Optimization Algorithm Based MPPT Technique for Solar PV System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 498 Mounika Sriram and K. Ravindra A New Topology of Interline Unified Power-Quality Conditioner for Multi Feeder System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 507 Surya Prakash Thota and Satish Kumar Peddapelli Adaptive NMPC Controller for Two-Link Planar Manipulator with Parameter Estimator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 520 B. Raja Gopal Redy, N. Karuppiah, Dubbaka Vinay Kumar Goud, and Samudrala Prashant Ganesh Solving the Complexity of Geometry Friends by Using Artificial Intelligence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 528 D. Marlene Grace Verghese, Suresh Bandi, and G. Jacob Jayaraj An Efficient Classification Technique for Text Mining and Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 534 Vb. Narasimha and Sujatha Clinical Big Data Predictive Analytics Transforming Healthcare: An Integrated Framework for Promise Towards Value Based Healthcare . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 545 Tawseef Ahmad Naqishbandi and N. Ayyanathan A Study of Different Techniques in Educational Data Mining . . . . . . . . 562 Nadia Anjum and Srinivasu Badugu A Study on Overlapping Community Detection for Multimedia Social Network . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 572 Sabah Fatima and Srivinasu Badugu A Review on Different Question Answering System Approaches . . . . . . 579 Tahseen Sultana and Srinivasu Badugu A Study of Malicious URL Detection Using Machine Learning and Heuristic Approaches . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 587 Aliya Begum and Srinivasu Badugu A Review on Network Intrusion Detection System Using Machine Learning . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 598 T. Rupa Devi and Srinivasu Badugu Review on Facial Expression Recognition System Using Machine Learning Techniques . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 608 Amreen Fathima and K. Vaidehi

Contents

xxix

Document Clustering Using Different Unsupervised Learning Approaches: A Survey . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 619 Munazza Afreen and Srinivasu Badugu A Study of Liver Disease Classification Using Data Mining and Machine Learning Algorithms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 630 Hajera Subhani and Srinivasu Badugu A Study of Routing Protocols for Energy Conservation in Manets . . . . 641 Aqsa Parveen and Y. V. S. Sai Pragathi Credit Risk Valuation Using an Efficient Machine Learning Algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 648 Ramya Sri Kovvuri and Ramesh Cheripelli Handwritten Mathematical Symbol Recognition Using Machine Learning Techniques: Review . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 658 Syeda Aliya Firdaus and K. Vaidehi Automatic Water Level Detection Using IoT . . . . . . . . . . . . . . . . . . . . . 672 Ch. V. S. S. Mahalakshmi, B. Mridula, and D. Shravani Optimal Scheduling of Tasks in Cloud Computing Using Hybrid Firefly-Genetic Algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 678 Aravind Rajagopalan, Devesh R. Modale, and Radha Senthilkumar A Novel Approach for Rice Yield Prediction in Andhra Pradesh . . . . . . 688 Nagesh Vadaparthi, G. Surya Tejaswini, and N. B. S. Pallavi Representation Techniques that Best Followed for Semantic Web - Web Mining . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 693 K. Vaishali and Sriramula Nagaprasad Isolated Health Surveillance System Through IoT Using Raspberry Pi . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 700 Sumayya Afreen, Asma Begum, G. Saraswathi, and Ayesha Nuzha Design and Implementation of RPL in Internet of Things . . . . . . . . . . . 707 M. V. R. Jyothisree and S. Sreekanth Detection and Tracking of Text from Video Using MSER and SIFT . . . 719 M. Manasa Devi, M. Seetha, S. Viswanada Raju, and D. Srinivasa Rao Blockchain Enabled Smart Learning Environment Framework . . . . . . . 728 G. R. Anil and Salman Abdul Moiz Global Snapshot of a Large Wireless Sensor Network . . . . . . . . . . . . . . 741 Surabhi Sharma, T. P. Sharma, and Kavitha Kadarala Author Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 753

About the Editors

Dr. Margarita N. Favorskaya is Professor and Head of Department of Informatics and Computer Techniques at Reshetnev Siberian State University of Science and Technology, Russian Federation. She is a member of KES organization since 2010, the IPC member and the chair of invited sessions of over 30 international conferences. She serves as Reviewer in international journals (Neurocomputing, Knowledge Engineering and Soft Data Paradigms, Pattern Recognition Letters, Engineering Applications of Artificial Intelligence); Associate Editor of Intelligent Decision Technologies Journal, International Journal of Knowledge-Based and Intelligent Engineering Systems, International Journal of Reasoning-based Intelligent Systems; Honorary Editor of the International Journal of Knowledge Engineering and Soft Data Paradigms; and Reviewer, Guest Editor and Book Editor (Springer). She is the author or the co-author of 200 publications and 20 educational manuals in computer science. She has co-authored/co-edited seven books for Springer recently. She supervised nine Ph.D. candidates and presently supervising four Ph.D. students. Her main research interests are digital image and videos processing, remote sensing, pattern recognition, fractal image processing, artificial intelligence and information technologies. Prof. Suresh Chandra Satapathy is currently working as Professor, School of Computer Engineering, KIIT Deemed to be University, Bhubaneswar, India. He obtained his Ph.D. in computer science and engineering from JNTU, Hyderabad, and M.Tech. in CSE from NIT, Rourkela, Odisha, India. He has 27 years of teaching experience. His research interests are data mining, machine intelligence and swarm intelligence. He has acted as program chair of many international conferences and edited over 25 volumes of proceedings from Springer series from LNCS, AISC, LNNS, LNEE, SIST, etc. He is also on the editorial board of few international journals and has over 130 research publications in international journals and conference proceedings.

xxxi

xxxii

About the Editors

Dr. K. Shyamala is working as Professor in Computer Science and Engineering, University College of Engineering, Osmania University, Hyderabad. She received B.E (CSE) and M.Tech (CSE) from Osmania University and Ph.D. from IIT Madras. She has published 24 papers in various national/international conferences and journals. Her areas of interests include mobile ad hoc network, wireless sensor nodes, routing protocols in MANETS parallel computer architecture and parallel algorithms. She is Active Member of IEEE, ACM and CSI. She is Member of Telangana State Council of Higher Education. She is reviewer of several journals and technical program committee member of various international conferences. Currently, she is guiding 11 Ph.D. scholars. e-mail: [email protected] Dr. D. Rama Krishna received his Bachelor of Technology (B.Tech) in electronics and communications engineering from Sri Krishnadevaraya University, Anantapur, Andhra Pradesh, India, and obtained his Master of Engineering (M.E.) and Doctor of Philosophy (Ph.D.) in electronics and communication engineering from Osmania University, Hyderabad, Telangana, India. He joined as Assistant Professor in the Department of ECE, University College of Engineering, Osmania University, in the year 2007; presently, he is working as Associate Professor. He served as Chairperson Board of Studies (Autonomous) for the Department of ECE, University College of Engineering, Osmania University, from March 2017 to March 2019; he has taught several undergraduate and graduate courses in communication engineering area and supervised nearly 25 UG and 60 PG students’ projects in the area of RF and microwave communication systems; currently, he is guiding 08 Ph.D. scholars at Osmania University. He successfully completed 03 sponsored research projects in the area of RF and microwave engineering and published 38 research papers in international journals/conference proceedings. His research areas of interest include multifunction antennas and antenna systems and microwave and millimeter-wave integrated circuits. He is Life Member of Institution of Engineers (IE), Institution of Electronics and Telecommunication Engineers (IETE), Indian Society for Technical Education (ISTE), Indian Society of Systems for Science and Engineering (ISSE) and Institute of Smart Structures and Systems (ISSS) and Member of Institute of Electrical and Electronics Engineers (IEEE), USA. He served as Secretary/Treasurer for the MTT/AP/EMC Society Joint Chapter of IEEE Hyderabad Section from January 2013 to December 2016. e-mail: [email protected] Dr. K. Srujan Raju is currently working as Dean Student Welfare and Heading Department of Computer Science & Engineering at CMR Technical Campus. He obtained his Doctorate in Computer Science in the area of Network Security. He has more than 20 years of experience in academics and research. His research interest areas include Computer Networks, Information Security, Data Mining,

About the Editors

xxxiii

Cognitive Radio Networks and Image Processing and other Programming Languages. Dr. Raju is presently working on 2 projects funded by Government of India, has filed 7 patents and 1 copyright at Indian Patent Office, edited more than 10 book proceedings published by Springer publications - AISC series, LAIS and other which are indexed by Scopus also authored 4 books, contributed chapters in various books and published more than 30 papers at reputed and peer-reviewed Journals and International Conference. Dr. Raju was invited as Session Chair, Key note Speaker, TPC and reviewer for many National and International conferences. His involvement with students is very conducive for solving their day to day problems. He has guided various student clubs for activities ranging from photography to Hackathon. He mentored more than 100 students for incubating cutting edge solutions. He has organized many conferences, FDPs, Workshops and Symposiums. He has established the Centre of Excellence in IoT, Data Analytics. Professor Raju has acted as reviewer and Technical Member for many conferences and is editorial member for few journals. Raju received Significant Contributor and Active Member awards by Computer Society of India - Hyderabad Chapter.

Random Walk with Clustering for Image Segmentation Mohebbanaaz(&), Madiraju Sirisha, and K. Joseph Rajiv ECE Department, Nalla Malla Reddy Engineering College, Hyderabad, India [email protected], [email protected], [email protected]

Abstract. We propose a method that uses k-mean clustering and Random walk algorithm for image segmentation. The use of the random walk algorithm is widespread as it segments thin and elongated parts and can produce a complete division of the image. However if there is any minute discontinuity the unwanted part may get segmented. To avoid this we first partition the image into group of clusters then apply random walk algorithm. Comparing the results of segmented image with clustering and without clustering it is shown that the proposed algorithm is most effective. Keywords: Random walk


k-mean clustering
Segmentation
Prior labels

1 Introduction In Image Processing applications image segmentation plays an important role. So any image segmentation algorithm must be quick in computing and editing and it must produce exact segmented image. We are using Random walk algorithm with k-mean clustering [11] to get the desired segmented image. In k-mean clustering an image is divided into k regions and each region (seed points) is given user defined label. Each seed point or pixel is located with user defined labels. In paper [1] random walker first considers a seed point which gives us the solution of Dirichlet problem and then seed point is fixed to unity. Remaining points or labels are set to zero. Fully discrete calculus [2] development connects random walks on graphs and discrete potential theory [7]. In this paper we are proposing an algorithm which uses k-mean clustering and Random walk (RW) algorithm. We have many RW-based algorithms like RW [3], RWR [4], LRW [5] and PARW [6]. By including Markov random fields with prior seeds we have already implemented a sub Markov random walk algorithm (subRW) [9]. It can solve the twig trouble it’s miles prolonged by using labels earlier. In subRW algorithm we first construct a frame work for photograph segmentation. Then it goes away a graph-cut G from any node i having probability ci and then it goes to adjacent nodes in same graph-cut G with Pa probability of 1-ci. This walk is changed to satisfy Markov transition probability ( qði; jÞ ¼ 1) which results in an expanded graph Ge . This Ge is built by including auxiliary staying nodes which are connected with seeded and unseeded nodes into graph G. Further we related these nodes with auxiliary killing nodes. We use this subRW algorithm with K-mean clustering. © Springer Nature Switzerland AG 2020 S. C. Satapathy et al. (Eds.): ICETE 2019, LAIS 4, pp. 1–9, 2020. https://doi.org/10.1007/978-3-030-24318-0_1

2

Mohebbanaaz et al.

2 Literature Survey Many approaches and algorithms have been proposed for segmentation of images. These are mainly classified in two categories. One is semi-automatic and other is fullyautomatic method. All of the Random walker algorithms comes under semi-automatic method. Many methods are proposed based on graph-cuts [7]. For segmentation of any region user should provide seed points. These separate foreground and background image pixels. While segmenting large images these methods are practically not preferred. In automatic methods, manual interaction is not needed. In past years many Methods have been proposed based on this criterion. Random walk can be extended for disconnected objects without labeling [7] and in the presence of inhomogeneity’s [10]. Theoretical analysis can be made by adding watershed segmentation to RW framework [8]. In [3] first the image is divided into nodes called as seed points. A random walker which can be at any location we find the probability of random walker to reach k seed points in less time. In [4] random walk with restarting probabilities (RWR) has been developed. In Random Walk with Restarting probability (RWR) algorithm a random walker with probability (1-c) walks to adjacent nodes and with a probability of c comes back to the starting node (initial seed point) [4]. In [5] according to lazy random walks algorithm a random walker stays at the node with a probability 1 − a and can walk along the edges connected with node with a probability a. In [9] it is proposed that subRW with prior labels can segment even thin and elongated parts. we can remove the noise if needed by using optimization techniques [12].

3 Proposed Method We usually consider an image into two parts during segmentation. One is required part and the other one is unwanted part. Actually some of the pixels in both parts are similar. Many Random Walk based rules have been proposed to differentiate these parts. but as the pixel value is same we are unable to differentiate it. Hence we first form a group of clusters by using k-mean clustering. Then we apply SubRW algorithms. Let the original image be I(x, y). Let the initial partitions obtained from the k-mean clustering be R ¼ fR1 ; R2 ; R3 . . . RN g; Where Ri denotes the ith partition and N is the total number of partitions. The size of each partition or cluster Ri be denoted by Ni. Calculate the mean intensity Mi of each partition Ri by: X Iðx; yÞ ð1Þ Mi ¼ ðx;yÞ Now find the mean intensities between the two partitions   Mij ¼  Mi  Mj 

ð2Þ

Random Walk with Clustering for Image Segmentation

Now find the difference between the intensities X    Bij ¼ I ð xi ; yi Þ  I xj ; yj ðxi ;yi Þðxj ;yj Þ

3

ð3Þ

where (xi, yi) 2 Ri and (xj, yj) 2 Rj. Now we find the clusters Cij which is the measure of similarity in intensities.   Cij ¼ 1=2 Mij þ Bij

ð4Þ

After determining Cij for all partitions i and j, we fix on the threshold Tc which Cij must satisfy before partitions i and j can be merged. If Cij is less than Tc, it implies that partition i and partition j are similar and hence they should be merged. Now we use the SubRW algorithm with prior labels as follows. For each cluster we set weight wihk as: wihk ¼ ð1  ci Þ h uki

ð5Þ

Where regularizing parameter k gives the measure of importance of the prior distribution. The transition probability is expressed as

ð6Þ

The probability that the random walker from any node position to reach or prior lk node hk or the mth staying node Slkm with label lk , is expressed as rim lk rim ¼ ð1  ci Þ

X

lk wij rim huki k þ ð 1  c Þ þ ci blim i ji d þ h di þ hgi i gi

ð7Þ

Now the present node hk is considered as a new node with label lk . So now we consider the reaching probability of hk as following vector k rmlk ¼ ðI  Dc Þprmlk þ ðI  Dc Þ uk þ Dc blim

¼ ðI  ðI  Dc ÞpÞ1 ððI  Dc Þ  uk þ Dc blimk Þ  1 ððI  Dc Þuk þ Dc blimk Þ ¼E

ð8Þ

4

Mohebbanaaz et al.

Modified vector notation rmlk can be given as   1  1 1 k lk r ¼ E ðI  Dc Þu þ Dc b Zk Mk lk

ð9Þ

The final labeling (segmentation) result including prior labels is as follows: Ri ¼ arg maxlk rilk

ð10Þ

 i represents the final node of an image. We can also optimize the Where R algorithm. Suppose 8i; 0\ci\1; then the objective function to optimize is as follows:

2 XN XN lk lk  lk ¼ 1 O w r  r i;j im jm i¼1 i¼1 2 X   1 N ðdi þ kgi Þci lk lk 2 w þ r  r i;j im im i¼1 ð1  c Þ 2 i XN 1 XN 1 XN lk lk 2 þ h uki ðrim  1Þ2 þ h utirim i¼1 t¼1;t6 ¼ k; i¼1 2 2

ð11Þ

The vector form of above equation is  lk ¼ 1 r lk T ðD  W Þr lk O m 2 m      1  lk T þ rm  blmk ðI  Dc Þ1 D þ h Dg Dc rmlk  blmk 2 T   h XK h  lk rm  e Dku rmlk  e þ r lk T Dtu rmlk þ t¼1;t6¼k; m 2 2

ð12Þ

The partial derivative of rmlk , optimizes the result.  lk  l  @O lk k ¼ ðD  W Þrmlk þ D1 g Da Dc rm  bm lk @rm ! k X þ k Dku rmlk þ Dtu rmlk  uk ¼ ¼

WÞrmlk



t¼1;t6¼k;

lk lk þ D1 ðDa  g D a D c r m  bm   1 lk D1 g Da ½ Dg þ Dc  Da Dg W rm



 lk ¼ D1 g Da Erm



 k uk

k  Dc blmk  D1 a Dg k u      Dg uk þ Dc blmk

ð13Þ

Random Walk with Clustering for Image Segmentation

5

By limiting the above vector, the achieving likelihood rmlk will be reliable with clustering. Adding clustering before applying RW algorithm produces some noise. Some distortions are produced due to these noises. To avoid this we can decrease parameter k. The twig part may get lost if the value of k is too small. So to reduce noise we combine labels prior value for each node before clustering. We perform coarse segmentation to avoid loss of twig part. We represent coarse segmentation as CRi ¼ arg maxk uki

ð14Þ

As we know much noise is added due to the coarse segmentation, but these noise is mostly present in unwanted area (background Region). So, the candidate areas are selected from cluster regions with seeds from the coarse segmentation. Next we give label to each candidate regions, which facilitates to keep the earlier records of small element as well as noise gets eliminated. ALGORITHM 1. An input image I(x, y) is divided using three GMM components into N clusters using k-mean clustering. 2. Calculate the mean intensity of each partition Ri find the mean intensities between two partitions Mij 3. Find the difference between the intensities Bij 4. Find the cluster Cij based on measure of similarities in intensities. 5. Apply Sub Markov Random Walk Algorithm. 6. Reset prior values of cluster if c 6¼ 1 7. Obtain segmentation results.

4 Simulation Results Figure 1 is the input image [9] which we consider for segmentation. Figure 2 gives us scribbled image which separate both foreground and background. Figure 3 gives us GMM components with a range of color maps. The segmentation outputs with clustering and without clustering are shown in Fig. 4(a) and (b).

Fig. 1. Input image

Fig. 2. Scribbled image

6

Mohebbanaaz et al.

Fig. 3. GMM components with range of color maps

Fig. 4. Segmented image using RW algorithm (a) without clustering and (b) with clustering

The segmented image from proposed method is compared with the manually segmented image to find how accurately our image is segmented. We extract the binary images from segmented images. Figure 5(a) and (b) are the extracted binary images of Fig. 4(a) and (b). Figure 6 is manually segmented image.

Fig. 5. Extraction of binary image (a) without clustering (b) with clustering

Random Walk with Clustering for Image Segmentation

7

Fig. 6. Manually segmented image

Let ‘M’ be the manual segmented image and ‘A’ be the segmented image. Let Fp (false positive), Tp (true positive) and Fn (false negative) be the regions considered. Fp = Unwanted region which is segmented. Tp = Required region which is segmented. Fn = Required region which is not segmented (Fig. 7).

A

M

Fp

Tp

Fn

Fig. 7. Evaluation metric

The Similarity Index (SI), Correct Detection Ratio (CDR), Under Segmentation Error (USE) and Over Segmentation Error (OSE) are used for efficient evaluation. SI is a measure which offers true segmented region relative to the total segmented region. CDR indicates the degree of trueness of the actual image. USE is the ratio of the number of pixels falsely identified as segmented portion by the proposed method to the manual segmented image. OSE is the ratio of number of pixels which are not identified by the proposed method to the manual segmented image. Total Segmentation Error (TSE) is sum of USE and OSE. The evaluation metrics SI, CDR, USE and OSE are obtained by equations (Table 1).

8

Mohebbanaaz et al.

SI ¼ CDR ¼ USE ¼ OSE ¼ TSE ¼

2Tp  100% 2Tp þ Fp þ Fn Tp  100% Tp þ Fn 2Fn  100% Tp þ Fn Fn  100% Tp þ Fn USE þ OSE

Table 1. Evaluation metric Image without clustering Image with clustering Similarity index (SI) 0.9285 0.9835 Correct detection ratio (CDR) 0.8917 0.9917 Under segmentation error (USE) 0.0289 0.0012 Over segmentation error (OSE) 0.1083 0.0031 Over segmentation error (OSE) 0.1372 0.0043

5 Conclusion We have conferred a completely unique framework which supports random walk with clustering for interactive image segmentation in this work. In the future, we’ll extend our algorithmic program to a lot of applications, such as extraction of multiple segments from single image. As we have done clustering prior RW noise of image has reduced due to which we are able to extract the particular region of an image with perfect edges and flexibility.

References 1. Kakutani S (1945) Markov processes and the Dirichlet problem. Proc Jpn Acad 21:227–233 2. Biggs N (1997) Algebraic potential theory on graphs. Bull London Math Soc 29:641–682 3. Grady L (2006) Random walks for image segmentation. IEEE Trans Pattern Anal Mach Intell 28(11):1768–1783 4. Kim TH, Lee KM, Lee SU (2008) Generative image segmentation using random walks with restart. In: Proceedings of ECCV, pp 264–275 5. Shen J, Du Y, Wang W, Li X (2014) Lazy random walks for superpixel segmentation. IEEE Trans Image Process 23(4):1451–1462 6. Wu X-M, Li Z, So AM, Wright J, Chang S-F (2012) Learning with partially absorbing random walks. In: Proceedings of NIPS, pp 3077–3085 7. Sinop AK, Grady L (2007) A seeded image segmentation framework unifying graph cuts and random walker which yields a new algorithm. In: Proceedings of IEEE ICCV, pp 1–8

Random Walk with Clustering for Image Segmentation

9

8. Couprie C, Grady L, Najman L, Talbot H (2011) Power watershed: a unifying graph-based optimization framework. IEEE Trans Pattern Anal Mach Intell 33(7):1384–1399 9. Mohebbanaaz, Vyshali S (2017) Image segmentation using sub-markov random walk algorithm with prior labels. Int J Innovative Res Sci Eng Tecnol 6(7). https://doi.org/10. 15680/IJIRSET.2017.0607243 10. Rajeyyagari S, Babu GA, Mohebbanaaz, Bhavana G (2019) Analysis of image segmentation of magnetic resonance image in the presence of inhomongeneties. Int J Recent Technol Eng 7(5):17–21. ISSN: 2277-3878 11. Ramakrishna Murty M, Murthy JVR, Prasad Reddy PVGD et al (2011) Dimensionality reduction text data clustering with prediction of optimal number of clusters. Int J Appl Res Inf Technol Comput 2(2):41–49. https://doi.org/10.5958/j.0975-8070.2.2.010 12. Himabindu G, Ramakrishna Murty M et al (2018) Classification of kidney lesions using bee swarm optimization. Int J Eng Technol 7(2.33):1046–1052

Design and Simulation of Microstrip Branch Line Coupler and Monopulse Comparator for Airborne Radar Applications Annapureddy Venkata Reddy(&) and V. G. Borkar Kalyani Centre for Technology and Innovation, Bharat Forge Limited, Gachibowli, Hyderabad 500032, Telangana, India [email protected]

Abstract. This paper presents the design and simulation of Microstrip branch line coupler and monopulse comparator at Ku-band frequency for airborne Radar applications and especially for missile applications. The proposed microstrip branch line coupler and comparator are low profile components and they have been designed and simulated using FEKO (Field Electromagnetic Computation) software by taking 2.2 as relative permittivity and RT Duroid 5880 as a substrate material. The microstrip monopulse comparator formed with four branch line couplers and four 90° bends. The monopulse comparator has great significance in missile guidance systems for finding the angular position of the target and it requires more accurate values to guide the missile system. The simulated results obtained are close to the specifications for required application. Keywords: Branch line coupler
Hybrids
Bends Substrate material
Microstrip
Band width


Dielectric constant


1 Introduction A Branch Line Coupler (Quadrature 90° Hybrid) is a four-port device with a 90° phase difference between two coupled ports. This microstrip coupler can be used as a Power splitter or as a power combiner. The 4-port branch line coupler is highly symmetrical in which any port can be used as input port, opposite two ports are output ports and other port is isolated port. Branch line couplers or the quadrature hybrid are 3 dB directional couplers with a 90° phase difference in the outputs. Microstrip structures are most widely used laminates because of its low profile, less weight, easy to fabricate and easy to installation. The proposed microstrip structure is a single layer structure. A parametric design analysis of branch line coupler and monopulse comparator has been discussed here. This section gives an introduction to the microstrip hybrid and monopulse comparator. In Sects. 2 and 3 specifications and selection of substrate material and thickness selection has been discussed. Section 4 presents detailed design procedure of branch line coupler and monopulse comparator. Sections 5 and 6 presents Simulation results and conclusion respectively.

© Springer Nature Switzerland AG 2020 S. C. Satapathy et al. (Eds.): ICETE 2019, LAIS 4, pp. 10–18, 2020. https://doi.org/10.1007/978-3-030-24318-0_2

Design and Simulation of Microstrip Branch Line Coupler

11

2 Specifications • • • • •

Frequency : Return Loss : Port isolation: Insertion loss: :

Upper Ku-band >−25 dB >−25 dB PAPRo}, is used to contemporary the array of PAPR as a probability of occurrence.

10

0

10

-1

OFDM OFDM-PTS R-PTS R-PTS-Mu law R-PTS-A law 10

-2

0

2

4

6 PAPR(dB)

8

10

12

Fig. 2. CCDF performance of Reduced complexity PTS with a A-law and l-law companding

10

0

0

10

CCDF

CCDF

10

-1

-1

10

OFDM OFDM-PTS R-PTS R-PTS+RCT 10

-2

-2

0

2

4

6 PAPR(dB)

8

10

12

10

0

OFDM OFDM-PTS R-PTS R-PTS+NERF 2

4

6 PAPR(dB)

8

10

12

Fig. 3. CCDF performance of reduced complexity PTS with RCT and NERF companding

74

T. Sravanti et al.

Figure 2 analyses the PAPR [11] at CCDF = 10−2, the OFDM PAPR is *11.5 dB, using PTS is *10.5 dB, using reduced complex PTS is *6.6 dB and hybrid method with A-law is *3.5 dB and l-law is *5.0 dB for A = 87.6 and l = 255. 0

10

CCDF

CCDF

10

-1

10

10

0

-1

OFDM OFDM-PTS R-PTS R-PTS+COS

OFDM OFDM-PTS R-PTS R-PTS+AEXP

-2

10

0

2

4

6 PAPR(dB)

8

10

10

12

-2

0

2

4

6 PAPR(dB)

8

10

12

Fig. 4. CCDF performance of reduced complexity PTS with a Cos and AEXP companding

10

0

10

CCDF

CCDF

10

-1

10

0

-1

OFDM OFDM-PTS R-PTS R-PTS+logR

Original PTS red-pts R-pts+tanhR -2

10

0

2

4

6 PAPR(dB)

8

10

12

10

-2

0

2

4

6 PAPR(dB)

8

10

12

Fig. 5. CCDF performance of reduced complexity PTS with a tanhR and logR companding

Figure 3 analyses the PAPR at CCDF = 10−2, the OFDM PAPR is *11.5 dB, using PTS is *10.5 dB, using reduced complex PTS is *6.6 dB. The hybrid method is designed by modified error functions. PAPR of RCT is *4 dB for R = 0.5, in literature as R increases the PAPR increases and PAPR of NERF is *3.8 dB. From

Reduced Complexity Hybrid PAPR Reduction Schemes

75

these, the optimal PAPR is realized by designing hybrid method by NERF as it doesn’t depend on any parameter. The analyses in Fig. 4, the PAPR at CCDF = 10−2, the OFDM PAPR is *11.5 dB, using PTS is *10.5 dB, using reduced complex PTS is *6.6 dB. The hybrid method is designed by modified exponential functions. PAPR of COS is *3.0 dB for y = 0.5, here the parameter y is directly correlated to COS function and AEXP, PAPR is *3.5 dB for d = 1.4 as parameter d increases, the PAPR increases because the absolute value of input is taken to avoid distortions. The Fig. 5 analyses the PAPR at CCDF = 10−2, the OFDM PAPR is *11.5 dB, using PTS is *10.5 dB, using reduced complex PTS is *6.6 dB. The hybrid method is designed by modified tangent functions. PAPR of tanhR is *4.0 dB for y = 1, k = 5 and logR is *3.8 dB for y = 0.5, k = 10. In tanhR, the rate of improvement is for k < 10 whereas in logR as parameter k increases, the rate of improvement increases and parameter y controls the companding level. From these, the optimal PAPR is achieved by designing hybrid method by logR.

5 Conclusions A novel hybrid method with 16-QAM, N = 64 subcarriers is executed by coalescing reduced complexed PTS together with modified non-linear companding techniques to decrease the PAPR and improve OBI with low complex design. To reduce PAPR of reduced complex PTS for PAPR *6.6 dB technique, a novel companding methods are pinned after reduced complex PTS. The hybrid system with RCT the PAPR is 4.0 dB, NERF is 3.8 dB hence, hybrid method using NERF is better companding with optimum PAPR. Modified exponential companding COS PAPR is 3.0 dB, AEXP reduces to 3.5 dB, therefore hybrid with COS results low PAPR. The modified tangent companding tanhR reduces to 4.0 dB, logR lessens to 3.8 dB. A hybrid technique with logR companding technique is additionally effective with minus complexity. Therefore, these hybrid systems are considered as per the system requirements based on efficiency of PAPR and complexity. These type of companding methods combined to PTS technique are more appropriate for OFDM application in future broadcasting systems that doesn’t have high-level processor, subsequently it accepts suggestive decrease in PAPR with truncated cost, computational difficulty with insignificant performance deprivation. Acknowledgement. The above work has been carried out under the project entitled “Study and Implementation of Self-Organized Femtocells for Broadband Services to Indoor Users in Heterogeneous Environment” sponsored by AICTE, New Delhi under Research Promotion Scheme (RPS), Vide sanction letter No: 8-30/RFID/RPS/POLICY-1/2016-2017, Dated: 2 August 2017.

76

T. Sravanti et al.

References 1. Farhang-Boroujeny B, Moradi H (2016) OFDM inspired waveforms for 5G. IEEE Commun Surv Tutor 18(4):2474–2492 Fourth quarter 2. Rahmatallah Y, Mohan S (2013) Peak-to-average power ratio reduction in OFDM systems: a survey and taxonomy. IEEE Commun Surv Tutor 15(4):1567–1592 Fourth Quarter 3. Cimini Leonard J, Sollenberger Nelson R (2000) Peak-to-average power ratio reduction of an OFDM signal using partial transmit sequences. IEEE Commun Lett 4(3):86–88 4. Han SH, Lee JH (2004) PAPR reduction of OFDM signals using a reduced complexity PTS technique. IEEE Sig Process Lett 11(11):887–890 5. Jiang T, Xiang W, Richardson PC, Guo J, Zhu G (2007) PAPR reduction of OFDM signals using partial transmit sequences with low computational complexity. IEEE Trans Broadcast 53(3):719–724 6. Ghassemi A, Gulliver TA (2008) A low-complexity PTS-based radix FFT method for PAPR reduction in OFDM systems. IEEE Trans Sig Process 56(3):1161–1166 7. Ku S-J, Wang C-L, Chen CH (2010) A reduced-complexity PTS-based PAPR reduction scheme for OFDM systems. IEEE Trans Wirel Commun 9(8):2455–2460 8. Wang X, Tjhung TT, Ng CS (1999) Reduction of peak-to-average power ratio of OFDM system using a companding technique. IEEE Trans Broadcast 45(3):303–307 9. Anjaiah Ch, Krishna H, Prasad P (2015) Mu-law companded PTS for PAPR reduction in OFDM systems. In: 2015 IEEE international conference on electrical, computer and communication technologies (ICECCT), Coimbatore, pp. 1–4 10. Al-Hashemi ZSH (2015) An overview: peak to average power ratio (PAPR) in OFDM system using some new PAPR techniques (with mat lab code), June 2015 11. Sravanti T, Vasantha N (2017) PAPR reduction in OFDM using reduced complexity PTS with companding. In: 2017 Third International Conference on Advances in Electrical, Electronics, Information, Communication and Bio-Informatics (AEEICB), Chennai, pp. 371–374

Speech Enhancement Through an Extended Sub-band Adaptive Filter for Nonstationary Noise Environments G. Amjad Khan1(&) and K. E. Sreenivasa Murthy2 1

Rayalaseema University, Kurnool, Andhra Pradesh, India [email protected] 2 G. Pullaiah College of Engineering and Technology, Kurnool, Andhra Pradesh, India

Abstract. A new speech enhancement algorithm is proposed in this paper with an aim of reducing the non-stationary noises added on the clean speech signals. The suppression of nonstationary noise is a serious problem. The attributes of noise differ according to the type of noise and environment in which the noise occurs. To solve the issue of nonstationary noises, a novel nonstationary noise suppression mechanism based on sub-band adaptive filtering (SAF) is proposed in this paper. The performance of sub band adaptive filtering is excellent in the case of speech signals when combined at very low SNR conditions. In addition to SAF, a noise classification mechanism is proposed in this paper to reduce the additional computational complexity in noise identification. Extensive simulations are performed according to the proposed mechanism by using different speech signals with different noises at different signal-to-noise ratios. The performance is evaluated in terms of the performance metrics, signal distortion, background intrusiveness, and overall quality. The proposed mechanism exhibits an outstanding performance. Keywords: Speech enhancement
Nonstationary noise Sub-band adaptive filtering
PESQ
SegSNR
WSS




1 Introduction In recent years, the use of speech-oriented applications, such as modern mobile communications, automatic speech recognition systems, and computer–human interaction systems has increased considerably. To enhance the quality and clarity of speech, unnecessary noises in the speech must be suppressed [1]. The presence of unnecessary noises makes speech-oriented applications ineffective in the real world. For example, in the hands free phone system, the microphone is placed at a definite distance from the mouth of speaker. In these situations, additional noises, such as crowd and vehicle sounds, are added to the speech. Such speech signals cannot deliver considerable information at the end user side. Hence, speech enhancement by suppressing unnecessary noises is an important research subject. Generally, speech processing systems are constructed with an inbuilt noise suppression system. © Springer Nature Switzerland AG 2020 S. C. Satapathy et al. (Eds.): ICETE 2019, LAIS 4, pp. 77–88, 2020. https://doi.org/10.1007/978-3-030-24318-0_10

78

G. Amjad Khan and K. E. Sreenivasa Murthy

Several techniques have been proposed to suppress the noises in a noisy speech signal. Spectral subtraction is a popular noise suppression technique. This technique was developed to reduce the effects of noise in a speech signal. In this method, the spectral magnitude is adjusted such that the noise from noisy speech reduces [2]. Spectral subtraction primarily involves identifying the noise spectrum from the nonspeech segments, which result in the loss of intelligibility and occurrence of residual noise in the enhanced speech. Sub-band-based speech enhancement is another method for reducing the noise in speech signals [3]. In this article, an adaptive speech enhancement algorithm based on sub-band adaptive filtering (SAF) is proposed. Conventional speech enhancement approaches are inefficient under nonstationary noisy environments. Therefore, in this study, the best-fit noise part was obtained using the normalized least mean square (NLMS) algorithm. A novel SAF technique was used for noise suppression after determining the best-fit noise part from the noisy speech by using the NLMS algorithm. The proposed SAF technique is a novel extension of conventional SAF techniques. The main objective of this method is to eliminate the nonstationary noises from a noisy speech signal with a low computational complexity and fast convergence rate. Extensive simulations were performed with the proposed approach by using different speech signals and different noises, such as babble, restaurant noises, and factory noises. The performance of the proposed method was measured through performance metrics such as the perceptual evaluation of speech quality (PESQ), log likelihood ratio (LLR), segmental signal-tonoise ratio (SegSNR), and weighted spectral slope (WSS). The rest of the paper is organized as. Section 1 describes a brief Introduction. Section 2 provides the particulars of the conventional method or previous methods. The particulars of the proposed mechanism are provided in Sect. 3. The simulation results are included in Sect. 4. Finally, the conclusions are stated in Sect. 5.

2 Literature Survey Various approaches have been developed for enhancing the speech signal in different noise environments. There exist two major classes of speech enhancement approaches, namely time-domain speech enhancement approaches, such as subspace methods [4– 6], and frequency-domain approaches, such as spectral subtraction methods [7–11]. To enhance a speech signal corrupted by environmental noise, a novel signalsubspace-based approach was presented in [4]. In this novel approach, the perceptual Karhunen–Loève transform (KLT) technique was improved by using the variance of the reconstruction error (VRE) to optimize the subspace decomposition model. An optimally modified log-spectral amplitude (OM-LSA) system was proposed in [7] by varying the gain function of LSAs according to two hypotheses. This system exhibited significant improvement in speech enhancement. To reduce mixed nonstationary noises, a single-channel speech enhancement method was proposed in [8]. The method proposed in [8] was based on a wavelet packet and an ideal binary mask thresholding function. Although the aforementioned methods can suppress the noise in a noisy speech signal, the convergence factor is very low. Furthermore, the computational complexity

Speech Enhancement Through an Extended Sub-Band Adaptive Filter

79

is very high. For achieving an increased quality and low computational complexity in speech enhancement applications, a new class of approaches known as adaptive algorithms have attracted increased interest [13–15]. These algorithms are effective in acoustic noise control and nonstationary noise suppression. Least mean square (LMS) is the most standard adaptive algorithm [17]. The LMS algorithm has low computational complexity and is very simple to implement [14]. However, the speech signal reconstructed after suppressing the noise from noisy speech exhibits convergence problems. The convergence problems are highly sensitive to the variations in the step size. For a small step size, the convergence is slow and the MSE is small. The MSE is high and the convergence is very fast for a large step size.

3 Proposed Approach To solve the problems associated with conventional approaches, a novel band dependent variable-step-size sign SAF (BD-VSS-SSAF) technique is proposed.

Input Noisy Speech

Noise N1

Feature Extraction

N2 N3

Noise N8 Best fit noise Noise

BD-VSS-SSAF

Optimal weights and Step size Selection

Output Enhanced Speech Fig. 1. Flowchart of the developed system.

The proposed method is an extension of the SAF method [21]. The complete working procedure is displayed in Fig. 1. As displayed in Fig. 1, the input noisy speech signal is initially subjected to noise classification through the SVM algorithm. Twelve

80

G. Amjad Khan and K. E. Sreenivasa Murthy

types of noises are used to train the SVM algorithm. In the testing phase, the input noisy speech is processed for detecting the noise type. After the noise is detected, the optimal weights and step sizes are selected. The input noisy speech is correlated with the standard noise signal to extract the best-fit noise part (i.e., the noise-dominant regions in the noisy speech). Because processing a complete signal results in increased complexity, the proposed best-fit noise evaluation part extracts only the noise-dominant regions. Only the noise-dominant regions are processed by the proposed noise suppression mechanism (i.e., BD-VSS-SSAF). The details of the noise suppression mechanism are provided in the following subsections. 3.1

BD-VSS-SSAF

Consider a desired signal d ðnÞ obtained from an unknown system. The signal is represented as follows: d ðnÞ ¼ vðnÞ þ uT ðnÞw0

ð1Þ

Fig. 2. Sub band adaptive filtering system.

where w0 denotes a weight vector that must be calculated through the adaptive filter, uðnÞ is an input signal vector, T indicates the transpose, and vðnÞ indicates the additive noise. In this noise suppression system, the desired signal d ðnÞ and input signal uðnÞ are partitioned into N sub-band signals [di ðnÞ and ui ðnÞ; i ¼ 0; 1; 2; . . .; N  1] by the analysis filter bank ½Hi ðzÞ; i ¼ 0; 1; 2; . . .; N  1]. and synthesis filter bank ½Gi ðzÞ; i ¼ 0; 1; 2; . . .; N  1]. Then, the sub-band signals yi ðnÞ and di ðnÞ are decimated such that the decimated sub-band signals are represented as yi;D ðk Þ and di;D ðkÞ, where n denotes the original signal and k represents the decimated signal.

Speech Enhancement Through an Extended Sub-Band Adaptive Filter

81

The sub-band error signal ei;D ðk Þ is calculated as follows: ei;D ðkÞ ¼ di;D ðk Þ  uTi ðk ÞwðkÞ

ð2Þ

For i ¼ 0; 1; 2; . . .; N  1 Where wðk Þ ¼ ½w1 ðk Þ; w2 ðkÞ; . . .; wM ðkÞ denotes the weight vector, di;D ðkÞ ¼ di ðkN Þ; and ui ðkÞ ¼ ½ui ðkN Þ; ui ðkN  1Þ; . . .; ui ðkN  M  1Þ Equation (2) can be represented in the vector form as follows: eD ðkÞ ¼ dD ðk Þ  U T wðkÞ
 where dD ðkÞ ¼ d0;D ðkÞ; d1;D ðkÞ; . . .; dN1;D ðkÞ and

ð3Þ

U ðk Þ ¼ ½u0 ðkÞ; u1 ðkÞ; . . .; uN1 ðkÞ: According to [16], the weight vector of the original SSAF is obtained as follows: U ðkÞsgnðeD ðkÞÞ wðk þ 1Þ ¼ wðkÞ þ l qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi : PN1 T u ð k Þu ð k Þ þ  i i i¼0

ð4Þ

where l is the step size,  is a small arbitrary constant to avoid division by zero, and sgn(.) denotes the sign function of every element in eD ðk Þ. In the case of equal weight factors, 1 k1 ¼ k2 ¼; . . .; kN1 ¼ qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi PN1 T i¼0 ui ðk Þui ðk Þ þ 

ð5Þ

In the case of nonstationary noises, an equal weight factor does not enable superior performance for noise suppression in a speech signal because each of the sub-bands provide an irreplaceable contribution to the overall convergence performance. Every sub-band has its own significance. Thus, every sub-band causes considerable enhancement in the suppression of nonstationary noises from speech signals. If individual weight factors are assigned for every sub-band, Eq. (6) changes as follows: 1 ffi ki ¼ pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi T ui ðkÞui ðk Þ þ 

ð6Þ

The weight vector update changes as follows:   XN1 ui ðkÞsgn ei;D ðkÞ pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi w ðk Þ þ l i¼0 uTi ðkÞui ðk Þ þ 

ð7Þ

82

G. Amjad Khan and K. E. Sreenivasa Murthy

According to Eq. (9), every single sub-band in the modified SSAF has an individual weighting factor ki , which enhances the noise suppression of nonstationary noises in speech signals. Allotting individual weights for every sub-band enhances the performance of the noise suppression algorithm. Unlike conventional approaches that use a constant step size, the step size is varied in the proposed method according to the following criterion: lðkÞ ¼ lðk  1Þ þ ueðkÞeðk  1ÞxT ðk  1ÞxðkÞ

ð8Þ

where u is a small arbitrary constant that controls the adaptation in the step size. 3.2

Best-Fit Noise Part

In the proposed method, the best-fit noise part is extracted from noisy speech by comparing the noisy speech signal with a standard reference noise signal. For obtaining the best-fit noise part, the standard reference noise signal is processed frame by frame. Initially, the reference noise signal is partitioned into N frames, and every frame is compared with the noisy speech signal to obtain the best-fit noise part. After the noise signal is split into N frames, frame-to-frame comparison is performed through the NLMS algorithm. According to the NLMS [21], the error updates and weight vector update are evaluated as follows: eq ðkÞ ¼ dq ðkÞ  xTi ðk Þwq ðk Þ wq ðk þ 1Þ ¼ wq ðk Þ þ

ð9Þ

lq eq ðkÞxq ðkÞ xTi ðkÞxq ðkÞ

ð10Þ

where q 2 ½1M  and M denotes the number of frames obtained after splitting the noise reference signal into frames of length L. The pure LMS has a high sensitivity to the scaling of input signals, which leads to the misguidance of the weight vectors (Fig.3). Frame with diverging weights

Frame with converging weights

0.1

0.3

0.08

0.25

0.06 0.2

0.02

Amplitude

Amplitude

0.04

0 -0.02

0.15 0.1 0.05

-0.04 0

-0.06 -0.05

-0.08 -0.1

0

20

40

60

80

100

120

140

160

180

200

-0.1

0

20

40

60

80

100

weights

weights

(a)

(b) Fig. 3. Weight convergence

120

140

160

180

200

Speech Enhancement Through an Extended Sub-Band Adaptive Filter

83

As a result, selecting a step size that guarantees the stability of the algorithm is difficult. Hence, the NLMS is used in the proposed method to determine the best-fit noise part. 3.3

Classification of the Noise Type

To achieve a high SNR in the enhanced speech signal, a machine leaning algorithm was used in this study after suppressing the nonstationary noises. The type of noise is initially identified by testing the input noisy speech signal with the SVM classifier. In the proposed method, the 256-point short-time Fourier transform (STFT) features are extracted and the energy of the translated features is then mapped to the bark domain. Furthermore, the SVM classifier is used to classify the noise present in the noisy speech signal. Because the SVM classifier is a binary classifier, it can classify only two classes at a time. In the proposed method, a multiclass SVM (MC-SVM) was used to classify different types of noises. For a system with N classes, the number of classifiers required is N(N − 1)/2. After the noise type is identified, the best-fit noise part is extracted and the noisy speech is passed through the proposed BD-VSS-SSAF method.

4 Simulation Results The execution of the proposed methodology is depicted in this section. Ten male and 10 female sounds were arbitrarily chosen from the TIMIT database to perform extensive computer simulations for determining the efficiency of the proposed algorithm. Seven noise types were considered from the NOISEX-92 database, namely cockpit noise, high frequency (HF) channel noise, floor noise, pink noise, car interior noise, destroyer operations room noise, destroyer engine room noise, jet cockpit noise 1, tank noise, and military vehicle noise. These noises were mixed with the clean speech signals received from the TIMIT database at various SNRs, such as 0, 5, and 10 dB. All the acquired clean speeches had a duration of at 2 s. Twenty clean speech tests were considered for execution assessment and are made out of 23000 examples on a normal. In the noise classification phase, totally 8 types of standard noise signals train the SVM algorithm. Each noise signal was divided into frames, and then a 128-point STFT was applied over every 15 frames. Furthermore, the STFT features were subjected to 18-dimensional bark feature extraction. The obtained bark features were used to train the system. In the testing phase, noisy speech of 2 s duration was processed for noise classification by using the SVM classifier. The approximate elapsed time of the noise signals is 10 s, and the approximate time of noisy speech is 2 s. After the noise type was identified, the noisy speech was evaluated for the best-fit noise part according to the correlating properties. The proposed BD-VSS-SSAF method was then used to perform noise suppression. To evaluate the performance of the proposed approach, four performance metrics, namely the PESQ, LLR, SegSNR, and WSS were evaluated. According to these performance metrics, three composite metrics were derived. The composite metrics derived for the proposed method were compared with those of conventional approaches, such as OM-LSA, MMSE-BC, and IMCRA [20]. The simulation results of the enhanced speech signals with the proposed approach are illustrated in Fig. 4.

84

G. Amjad Khan and K. E. Sreenivasa Murthy

(a)

(b) 5000

0.4

4500 0.3

4000 3500

Frequency

Amplitude

0.2

0.1

0

3000 2500 2000 1500

-0.1

1000 -0.2

-0.3

500

0

0.5

1

1.5

2

2.5

3

0

0.2

0.4

0.6

0.8

(c)

1

1.2

1.4

1.6

1.8

2

2.2

time

time

(d)

Fig. 4. Enhanced speech signals: (a) input noisy speech (car interior noise at SNR = 0 dB); (b) spectrogram of (a); (c) Output enhanced speech; and (d) spectrogram of (c).

In the PESQ test, 20 randomly selected listeners heard the original and enhanced speech signals and scored the signals from 1 to 5. The PESQ is a subjective measure, with high scores indicating superior quality. The PESQ values obtained using the proposed approach were higher than those obtained using the conventional approaches. The PESQ results include mean opinion scores (MOS) that cover a scale from 1 (bad) to 5 (excellent). The Weighted spectral distance, which is a direct measure of the spectral distance, is determined by comparing the smooth out spectra from the clean and warped speech samples. Methods such as Linear Predictive analysis, cepstrum filtering and filter bank analysis can be used to obtain the smoothed spectra. The Weighted Spectral distance can be stated as follows:

dWSS

 2 1 XM1 W ðj; mÞ Sc ðj; mÞ  Sd ðj; mÞÞ ¼ PK m¼0 M j¼1 W ðj; mÞ

ð11Þ

where K is the number of bands, M is the total number of frames, Sc ðj; mÞ and Sd ðj; mÞ are the spectral slopes (typically the spectral differences between neighboring bands) of the jth band in the mth frame for clean and distorted speech, respectively, and W ðj; mÞ is the weight, which can be calculated according to [21]. Furthermore, the SegSNR was also considered for evaluating the performance of the proposed method

Speech Enhancement Through an Extended Sub-Band Adaptive Filter

! PN 2 Xk þ N1 10 XM1 i¼1 x ðiÞ SegSNR ¼ log10 PN 2 m¼0 i¼k M i¼1 ðxðiÞ  yðiÞÞ

85

ð12Þ

where N is length of the segment, M is the number of segments, xðiÞ is the original speech, and yðiÞ is the enhanced speech. According to the PESQ, LLR, SegSNR and WSS, three composite metrics, namely the signal distortion (Csig ), background intrusiveness (Cbak ), and overall quality (Covl ), were derived. The composite metrics are represented by the following equations: Csig ¼ 3:093  1:029SLLR þ 0:603SPESQ  0:009SWSS

ð13Þ

Cbak ¼ 3:634 þ 0:4785SPESQ  0:007SWSS þ 0:063SSegSNR

ð14Þ

Covl ¼ 1:594 þ 0:8055SPESQ  0:512SLLR  0:007SWSS

ð15Þ

After measuring the aforementioned composite routing metrics, the metrics were scored over a five-point scale. The obtained results are described in the following tables.

Table 1. Results for the signal distortion (Csig ). Noise HF channel

SNR 0 5 10 Pink 0 5 10 Car interior 0 5 10 Destroyer operations 0 5 10 Destroyer engine 0 5 10 Tank 0 5 10 Military vehicle 0 5 10

OM-LSA 2.58 3.25 3.79 2.63 3.22 3.77 4.99 5.22 5.43 2.82 3.40 3.92 2.80 3.41 3.91 3.25 3.84 4.43 3.92 4.34 4.80

MMSE-BC 2.55 3.14 3.67 2.45 3.08 3.63 4.80 5.08 5.30 2.51 3.15 3.71 2.86 3.45 3.95 3.05 3.61 4.22 3.72 4.19 4.65

OMLSA-IMCRA 2.89 3.44 3.94 2.90 3.45 3.95 5.02 5.23 5.43 2.85 3.46 3.98 3.00 3.56 4.04 3.37 3.92 4.46 3.95 4.37 4.83

Proposed 2.97 3.65 4.18 3.04 3.67 4.11 5.09 5.33 5.61 2.96 3.69 4.09 3.14 3.71 4.18 3.49 4.15 4.61 4.11 4.52 4.89

86

G. Amjad Khan and K. E. Sreenivasa Murthy Table 2. Results for the background intrusiveness ðCbak Þ. Noise HF channel

SNR 0 5 10 Pink 0 5 10 Car interior 0 5 10 Destroyer operations 0 5 10 Destroyer engine 0 5 10 Tank 0 5 10 Military vehicle 0 5 10

OM-LSA 2.20 2.67 3.11 2.27 2.72 3.17 4.13 4.46 4.80 2.33 2.76 3.16 2.24 2.68 3.10 2.59 3.06 3.57 2.94 3.30 3.75

MMSE-BC 2.09 2.56 3.02 2.13 2.62 3.10 4.01 4.38 4.72 2.18 2.66 3.08 2.28 2.75 3.23 2.48 2.93 3.48 2.88 3.28 3.71

OMLSA-IMCRA 2.33 2.75 3.16 2.39 2.83 3.26 4.13 4.45 4.80 2.34 2.79 3.19 2.33 2.76 3.18 2.65 3.09 3.57 2.96 3.30 3.77

Proposed 2.49 2.86 3.29 2.51 2.97 3.39 4.35 4.68 4.96 2.47 2.89 3.32 2.49 2.89 3.29 2.78 3.19 3.68 3.17 3.56 3.89

Table 3. Results for the overall quality ðCovl Þ. Noise HF channel

SNR 0 5 10 Pink 0 5 10 Car interior 0 5 10 Destroyer operations 0 5 10 Destroyer engine 0 5 10

OM-LSA 2.16 2.76 3.26 2.27 2.80 3.29 4.42 4.64 4.84 2.45 2.97 3.42 2.34 2.87 3.34

MMSE-BC 2.16 2.69 3.18 2.17 2.73 3.23 4.27 4.54 4.74 2.25 2.82 3.30 2.46 2.99 3.45

OMLSA-IMCRA 2.43 2.92 3.37 2.51 2.99 3.45 4.43 4.65 4.85 2.48 3.01 3.46 2.52 3.00 3.46

Proposed 2.59 3.15 3.52 2.67 3.18 3.67 4.51 4.71 4.93 2.59 3.16 3.57 2.68 3.23 3.69 (continued)

Speech Enhancement Through an Extended Sub-Band Adaptive Filter

87

Table 3. (continued) Noise Tank

Military vehicle

SNR 0 5 10 0 5 10

OM-LSA 2.80 3.33 3.87 3.35 3.76 4.17

MMSE-BC 2.70 3.20 3.75 3.26 3.69 4.11

OMLSA-IMCRA 2.91 3.39 3.89 3.39 3.78 4.20

Proposed 3.05 3.48 4.03 3.61 3.87 4.41

As presented in Table 1, the signal distortion in the proposed approach was high for every noise type. The proposed approach involves a novel SAF technique, which effectively suppresses the nonstationary noises in noisy speech. Therefore, the quality of the improved speech should be high, which allows the listener to hear the speech clearly. Compared with the Csig for the OM-LSA, MMSE-BC, and OMLSA-IMCRA approaches, The results presented in Table 2 reveal the effectiveness of the proposed approach in suppressing nonstationary noises. A higher value of Cbak indicates better performance The obtained overall quality values are shown in Table 3. The proposed method attained a higher overall quality than the conventional approaches.

5 Conclusion In this paper, a new speech enhancement approach is proposed by combining SAF with a machine learning algorithm. In the proposed approach, the best-fit noise part from a noisy speech signal is identified through the NLMS algorithm. Furthermore, the proposed approach can determine the type of noise present in the input noisy speech signal, which considerably increases the speed of the system by decreasing the number of samples to be processed for suppression. The simulation results for the Csig , Cbak , and Covl metrics indicate the effective performance of the proposed approach. The values obtained for the performance metrics indicate that the proposed approach enhances the suppression of nonstationary noises and successfully preserves speech quality and intelligibility.

References 1. Sankar A, Beaufays SF, Digalakis V (1995) Training data clustering for improved speech recognition. In: Proceedings of European conference on speech communication and technology (EUROSPEECH), Madrid, Spain, pp 1–4 2. Boll SF (1979) Suppression of acoustic noise in speech using spectral subtraction. IEEE Trans Acoust Speech Sig Process 27(2):113–120 3. Ephraim Y, Malah D (1984) Speech enhancement using a minimum mean-square error short-time spectral amplitude estimator. IEEE Trans Acoust Speech Sig Process 32(6):1109– 1121

88

G. Amjad Khan and K. E. Sreenivasa Murthy

4. Saadoune A, Amrouche A, Selouani SA (2014) Perceptual subspace speech enhancement using variance of the reconstruction error. Digit Sig Process Rev J 24(1):187–196 5. Jensen JR, Benesty J, Christensen MG, Jingdong Chen A (2013) class of optimal rectangular filtering matrices for single-channel signal enhancement in the time domain. IEEE Trans Audio Speech Lang Process 21(12):2595–2606 6. Saadoune A, Amrouche A, Selouani SA (2013) MCRA noise estimation for KLT-VREbased speech enhancement. Int J Speech Technol 16(3):333–339 7. Cohen I, Berdugo B (2001) Speech enhancement for non-stationary noise environments. Sig Process 81(11):2403–2418 8. Singh S, Tripathy M, Anand RS (2014) Single channel speech enhancement for mixed nonstationary noise environments. Adv Sig Process Intell Recogn Syst 264:545–555 9. Bharti SS, Gupta M, Agarwal S (2016) A new spectral subtraction method for speech enhancement using adaptive noise estimation. In: 2016 3rd international conference on recent advances in information technology (RAIT), pp 128–132 10. Lu C-T, Tsen K-F, Chen Y-Y, Wang L-L, Leita C-L (2016) Speech enhancement using spectral subtraction algorithm with over-subtraction and reservation factors adapted by harmonic properties. In: 2016 international conference on applied system innovation (ICASI), pp 1–5 11. Cohen I, Bergudo B (2002) Noise estimation by minima controlled recursive averaging for robust speech enhancement. IEEE Sig Process Lett 9(1):12–15 12. Siddapaji, Sudha KL (2014) Performance analysis of new time varying LMS (NTVLMS) adaptive filtering algorithm in noise cancellation system for speech enhancement. In: 2014 4th world congress on information and communication technologies (WICT 2014), pp 224– 228 13. Wenchao X, Guangyan W, Lei C (2017) Speech enhancement algorithm based on improved variable-step LMS algorithm in cochlear implant. J Comput Appl 37(4):1212–1216 14. Hadei S, Iotfizad M (2010) A family of adaptive filter algorithms in noise cancellation for speech enhancement. Int J Comput Electr Eng 2 15. Nataraj VS, Athulya MS, Savithri SP (2017) Single channel speech enhancement using adaptive filtering and best correlating noise identification. In: 2017 IEEE 30th Canadian conference on electrical and computer engineering (CCECE) 16. Ni J, Li F (2010) Variable regularization parameter sign subband adaptive filter. Electron Lett 46(24):1605–1607 17. Widrow B, Glover J, McCool JM, Kaunitz J, Williams CS, Hearn RH, Zeidler JR, Dong E, Goodlin R (1975) Adaptive noise cancelling: principles and applications. Proc IEEE 63:1692–1716 18. Haykin S (2014) Adaptive filter theory, 4th edn. Pearson Education Asia, LPE, London 19. Karthik GVS, Ajay Kumar M, Rahman MdZU (2011) Speech enhancement using gradient based variable step size adaptive filtering techniques. Int J Comput Sci Emerg Technol 2 (1):168–177 (E-ISSN 2044-6004) 20. Yuan W, Xia B (2015) A speech enhancement approach based on noise classification. Appl Acoust 96:1119 21. Gerkmann T, Hendriks R (2012) Unbiased MMSE-based noise power estimation with low complexity and low tracking delay. IEEE Trans Audio Speech Lang Process 20(4):1383– 1393

Design of Dayadi 1-bit CMOS Full Adder Based on Power Reduction Techniques Pabba Sowmya, D. Lakshmaiah(&), J. Manga, Gunturu Sai Shankar, and Desham Sai Prasad Vignana Bharathi Institute of Technology, Hyderabad, India [email protected]

Abstract. Full adder being the basic building block that performs many operations like addition, division, multiplication and other similar arithmetic process in VLSI systems. Dayadi 1-bit CMOS FA implemented by using CMOS circuits and transmission gate circuits is presented. Design of Dayadi 1-bit CMOS full adder Based on Power Reduction Techniques. The sum and carry generation circuits are designed by novel logic style. The circuit was implemented using micro wind tools in cmos-90 nm technology. Factors such as no of transistors, propagation delay, PDP, and chip area were compared with the existing designs such full adder, TGL and so on. The ultimate aim of designing Dayadi 1-bit CMOS full adder is to reduce power dissipation and to decreased the delay compared to existing work performance. Keywords: Power
High-speed Area Dayadi full adder


Power delay product


1 Introduction The requirement for low-power VLSI systems is constantly increasing because of the endless applications emerging in mobile communication and compact devices. Today’s compact devices are usually battery operated for example, mobile phones, PDA’s [1], which demands VLSI with less power consumption. So designers and developers are facing more problems regarding high performance, rapid speed, low-power consumption and narrow silicon space. Thus constructing high performance low-power adder cells [2] are having enormous importance. Therefore in this project, a wellorganized approach for understanding the adder construction and working is given. It is listening carefully on splitting the whole FA into several smaller modules. Every single module is constructed, reduced the transistor count and simulated individually [3, 4]. Multiple FA cells are formed by joining this smaller module. In organize to get better the presentation parameters of Dayadi 1-bit CMOS full adder we need to design hybrid logic by utilizing CMOS logic [5] and transmission gate logic (TGL) [6]. CMOS technique is generally implemented by using PMOS pull-up and NMOS pull-down MOSFETs by keeping the MOSFETs in series at output side. The major drawback of this circuit is poor driving capability [7, 8]. To overcome this problem buffers need to be used. Besides this there are several advantages of C-CMOS logic style was gives full voltage hang which is very much desirable in designing complex © Springer Nature Switzerland AG 2020 S. C. Satapathy et al. (Eds.): ICETE 2019, LAIS 4, pp. 89–96, 2020. https://doi.org/10.1007/978-3-030-24318-0_11

90

P. Sowmya et al.

circuits. it gives enhanced quality throughput despite the fact of transistor sizing and voltage scaling [9]. Layout area of this logic design is efficient. Voltage degradation problem in the current logic is overcome by using transmission gate logic which comprises of only 20 transistors [10]. Here transmission gate usually acts as relay or analog switch. The functionality in this logic design is when PMOS is high NMOS will be low and vice versa [11]. The projected Dayadi 1-bit CMOS full adder is designed by using the aforementioned two logic styles [12].power delay product degrades considerably in the spill mode of process but the correctly intended Buffers are not incorporated [13]. Whereas some adders can be constructed by developed additional logic style. Cognate architectures are named as hybrid-logic circuit. These hybrid logic circuits makes use of beneficial features of on top of mentioned logic styles to improve the general carrying out of the FA [14]. still although this hybrid logic circuit style offers talented carrying out, a large portion of these designs encounter a poor driving capacity which results in the definite decrease in their execution in surge form of performance but realistically circuits buffers are excluded (Fig. 1).

Fig. 1. (a) Block diagram of full adder. (b) Existing XNOR module. (c) Existing Carry generation module. (d) Complete circuit diagram of full adder

The major purpose of this paper is to design Dayadi 1-bit CMOS full adder whose performance parameters are optimized in terms of power, Layout area, switching Delay and power delay product (PDP) and capacitance. Transistor count is reduced i.e. only 12 transistors are used and the present design is compared with the existing logic designs and performance parameters are compared. The design is employed using 65nm and 90-nm technology by using micro wind cadence tools. On the other hand, the layout area excluding buffer (350.25 lm2) and the delay of the circuit (3.98 ns), in 90nm technology, and we need 16 gates to build the circuit design and power consumed by 21.24 lW.

2 Proposed Dayadi 1-bit CMOS Full Adder Model The proposed Dayadi 1-bit CMOS full adder circuit is implemented by using three blocks as shown in Fig. 2 sum output (SUM) is generated by using module one and module two. These two modules are basically XNOR modules which will generate sum output and module three generates the carry output signal (Cout). These three modules are designed individually. Here we designed module 1 and module 3 separately by

Design of Dayadi 1-bit CMOS Full Adder Based on Power Reduction Techniques

91

reducing number of transistors (transistor count) to optimize parameters such power, delay and area. These modules are discussed below in detail. Hybrid technologies are generally to design full adders are formed by implementing circuits with more than one logic style. This improvement in power, delay and layout area was obtained using this logic style here we can observe the each module designs in Fig. 2 complete circuit diagram of proposed full adder, Fig. 3: Modified XOR circuit design, Fig. 4: Carry Generation Module2 and Fig. 5 Modified sum Module. Now we can see the each design module as shown in below with clear explanation with circuit diagrams

Fig. 2. CMOS circuit of proposed full adder

3 Modified XOR Module 1 In the proposed FA circuit, XNOR circuit reduced to only four gates designed by gates p1, p2, n1 and n2 compare to existing XOR have six transistors need to design the XOR module answerable for most of the power and delay consumption of the whole FA circuit. Consequently, this module is intended to reduce the delay to the most excellent probable extend with avoiding the propagation delay. Figure 3 shows in the XOR circuit anywhere the delay is reduced significantly by purposeful use of pathetic inverter.

Fig. 3. Modified CMOS XOR circuit design

Fig. 4. Carry generation Module 2 multiplexer

92

P. Sowmya et al.

A. Carry Generation Module 2 Multiplexer In the proposed circuit, the output carry signal is implemented by the transistors p4, p5, n4, and as in Fig. 4. The input carry signal (Cin) propagates only through a single transmission gate (n4 and p4), reducing the overall carry propagation path significantly. The intentional use of strong transmission gates guaranteed further decrease in propagation delay of the carry signal. B. Generation of SUM Module 3 In the proposed circuit, the output carry signal is implemented by the transistors p6 and n6 as shown in Fig. 5. Here also we can reduce the number of gates for increasing the performance compare to existing technology.

Fig. 5. Modified sum module

The noise margin of PMOS and NMOS gates and also we can observe the how much of power 0.225 mW consumption is done.

4 Reduction of Power Techniques Power consumption is the vital element in many electronic devices starting from mobile phones to computer systems. It plays key main role in determining the efficiency of any electronic system. In order to regulate and control power dissipation in electronic devices we need to adopt new design procedures. Generally preferred logic style is low power logic style in VLSI systems. The scale of integration is increasing day by day as highly developed electronic devices are employed on a very large scale integration (VLSI) design. As scale of integration increases; number of transistors increases and size of chip is decreased. Power consumption can be optimized through various techniques. If we minimize chip area we can achieve desired performance. It is necessary to handle trade-off between area and speed. However speed is design element and without disturbing functionality. Leakage Power Dissipation Non- zero reverse leakage and sub-threshold currents are the two common problems raised due to the P-type and N-type transistors in CMOS full adder. The circuit consists of transistors the current leakage of these transistors leads to power dissipation when the transistors are not performing switching action. Leakage power is mainly due to reverse bias diode leakage current between source terminal and drain terminal and

Design of Dayadi 1-bit CMOS Full Adder Based on Power Reduction Techniques

93

sub-threshold current is due to transistor twisted off channel. Computer system consumes static power along with dynamic power. Static power in CMOS design is considered as leakage power, to reduce power dissipation leakage power techniques are introduced. There are mainly three types of techniques 1. Ground Gating Technique 2. Power Gating Technique 3. Hybrid Gating technique.

4.1

Ground Gating Technique

In this ground gated technology we have to connect NMOS transistor in connection between GROUND and circuit diagram. NMOS transistor has low resistive nature and it has less parasitic elements like (RLC). NMOS transistor has high switching activity compared to PMOS (Fig. 6).

Fig. 6. Ground gated circuit diagram

4.2

Power Gating Technique

In this power gated technology we have to connect PMOS transistor in connection between power supply(VDD) and circuit diagram. PMOS transistor has high resistive nature and it has more parasitic elements like (RLC). PMOS transistor is used to resist the VDD power (Fig. 7).

Fig. 7. Power gated circuit diagram

94

4.3

P. Sowmya et al.

Hybrid Gating Technique

In this technology we have to connect the both GROUND gated and POWER gated technique. In this case both functionalities work at a time so that we can obtain low power usage and high accuracy without distributing functionality of one bit full adder. But in this technology both NMOS and PMOS transistors are used to construct the hybrid gated technique (Fig. 8).

Fig. 8. Hybrid gated circuit diagram

5 Performance and Simulation Results Compare the existing full adder models our proposed full adder models are better results like PDP area and delay (Figs. 9, 10, 11 and Table 1).

Fig. 9. Delay of full adder

Fig. 10. Area of full adder

Design of Dayadi 1-bit CMOS Full Adder Based on Power Reduction Techniques

95

Fig. 11. FA of simulation results Table 1. Comparison table of Existing and Proposed Models in 90 nm technology Parameter

Existing full adder [1] IEEE 2015 21.24 350.25 84.3

Power (µW) Area (µm2) PDP femito (w/s) performance Delay (ns) 3.98 Capacitance(fF) 18.35 at sum Transistors 16 count

Proposed Proposed full adder Ground gated-FA 62.16 54.68 224.17 248.325 80.6 13.57

Proposed Power gatedFA 43.623 295.375 10.26

Proposed Hybrid gatedFA 41.91 279.24 11.70

1.310 12.25

1.124 12.46

1.124 12.45

1.124 4.76

12

13

13

14

6 Conclusion In this research work Dayadi 1-bit CMOS FA implemented by using CMOS logic and transmission gate logic (TGL) is presented. The circuit was implemented using micro wind tools in cmos-90 nm technology. Factors such as no of transistors, propagation delay, PDP, and chip area were compared with the existing work full adder, TGL and so on compared to exiting technology performance is increased by two times and area was decreased by of the existed technology. Therefore the proposed structure of 1-bit FA uses only 12T’s instead of 16T’s compared with the previous work. Also power can be reduced by leakage power reduction techniques such as ground gating, hybrid gating and power gating techniques.

7 Future Work As a future scope, improving the execution of 1 bit Dayadi full adder can be executed by changing the value of W/L proportions. Utilizing the design of 1 bit proposed FA blocks, we can implement a two bit, four bit, eight bit, 16 bit, 32 bit, 64 bit Subtract or circuit, Adder circuit, and multiplier circuits. These adders can also be design and

96

P. Sowmya et al.

differentiate using different possible nm technologies like 90 nm, 65 nm, 32 nm, 22 nm, and so on. Acknowledgment. The author here by acknowledgment their deep gratitude to the management of VBIT institute of technology, Hyderabad to have proved the laboratory facilities for developing this project and also special thanks to Dr. Dayadi Lakshmaiah and Mrs. Manga J. for the proper guidance and special thanks to, our Vignana Bharathi Institute of Technology, Hyderabad, India chairmen Goutham Rao sir, secretary Manohar Reddy sir principal, Jayanth Kulakarni sir, ECE HOD Dr. Y. Srineevas sir and R&D Department.

References 1. Bhattacharyya P, Ghosh S, Kumar V (2015) Performance analysis of a low-power highspeed hybrid 1-bit full adder circuit. IEEE Trans Very Large Scale Integr (VLSI) Syst 23 (10):2001–2008 2. Megha V (2017) Performance analysis of a low-power high-speed hybrid 1-bit full adder circuit using CMOS technologies using Cadance. Int Res J Eng Technol (IRJET) 04 (08):1931–1938 e-ISSN 2395-0056 3. Tung C-K, Hung Y-C, Shieh S-H, Huang G-S (2014) A low-power high-speed hybrid CMOS full adder for embedded system. In: Proceedings of IEEE conference on design diagnostics electronic circuits and systems, vol 13, April 2014, pp 1–4 4. Goel S, Kumar A, Bayoumi MA (2013) Design of robust, energy efficient full adders for deep-submicrometer design using hybrid-CMOS logic style. IEEE Trans Very Large Scale Integr (VLSI) Syst 14(12):1309–1321 5. Weste NHE, Harris D, Banerjee A (2012) CMOS VLSI design: a circuits and systems perspective, 3rd edn. Pearson Education, Delhi 6. Rabaey JM, Chandrakasan A, Nikolic B (2012) Digital integrated circuits: a design perspective, 2nd edn. Pearson Education, Delhi 7. Radhakrishnan D (2001) Low-voltage low-power CMOS full adder. IEE Proc-Circ Devices Syst 148(1):19–24 8. Zimmermann R, Fichtner W (1997) Low-power logic styles: CMOS versus pass-transistor logic. IEEE J Solid-State Circ 32(7):1079–1090 9. Chang CH, Gu JM, Zhang M (2005) A review of 0.18-lm full adder performances for tree structured arithmetic circuits. IEEE Trans Very Large Scale Integr (VLSI) Syst 13(6):686– 695 10. Aranda ML, Báez R, Diaz OG (2010) Hybrid adders for high-speed arithmetic circuits: a comparison. In: Proceedings of 7th IEEE International Conference on Electrical Engineering Computing Science and Automatic Control (CCE), Tuxtla Gutierrez, NM, USA, September 2010, pp 546–549 11. Vesterbacka M (2003) A 14-transistor CMOS full adder with full voltage swing nodes. In: Proceedings of IEEE Workshop Signal Processing Systems (SiPS), Taipei, Taiwan, October 1999, pp 713–722 12. Zhang M, Gu J, Chang C-H (2003) A novel hybrid pass logic with static CMOS output drive full-adder cell. In: Proceedings of International Symposium on Circuits and Systems, May 2003, pp 317–320

Size Deduced Printed Antenna for W-LAN Applications Patturi Ravali(&), Samiran Chatterjee, K. Radhika Reddy, Narmala Raju, Adhimulam Rohith Kumar, Gorla Haritha, Gantla Sai Kumar Reddy, and Akula Naresh ECE Department, Jyothishmathi Institute of Technology and Science (Affiliated to JNTU, Hyderabad), Nustulapur, Karimnagar 505481, Telangana, India [email protected], [email protected], [email protected], [email protected], [email protected], [email protected], [email protected], [email protected] Abstract. An integrated top layer with single feed compact size MSA designed for the desired communication like Wireless local area network (WLAN) which combines the network along with wireless technology. Resonant frequency has been achieved by use of simple rectangular patch and circular patch deducted from the top layer. The simulated antenna characteristics designed by using method of moment based EM solution software named IE3D. In this paper we are analyze S11 parameter in dB, beam-width and absolute gain (dBi) for the designed antenna structure. Keywords: Compact Gain


Feed
WLAN
Layer
Radiation pattern
VSWR


1 Introduction In new age of wireless communication, the microstrip design creates a challenge with large bandwidth for communication engineers [1]. For wireless transmissions, it requires lightweight and small compact antenna. So, for this the compact with deduced size is mandatory for communication. For microwave connectivity as well as data communication, minimum two frequencies are required per day. The Resonant frequencies are used because communication engineers use different frequency bands for communication. Therefore, engineers have recently designed antennas with multiple factors which will help the communication in future. Reducing size is modern technique where the designed antenna size is same as conventional antenna. To reduce size, effective method is to subtract some portion from traditional antenna in terms of any structure [2–5]. Compressing of size is done using very low resonance frequency of the cleaved antenna compared to the traditional antenna [6–8]. Unlike designed antennas, other antennas also used to compress size of traditional antenna like DRA (Dielectric Resonator Antenna), fractal antenna, etc. to reduce the size of antenna. But for the above antennas, design complexity for fractal is more and large values dielectric constant requires for DRA and the large values of er is difficult to get from market. © Springer Nature Switzerland AG 2020 S. C. Satapathy et al. (Eds.): ICETE 2019, LAIS 4, pp. 97–103, 2020. https://doi.org/10.1007/978-3-030-24318-0_12

98

P. Ravali et al.

Today, compact size of the microstrip antenna increased the demand for applications in communications, mainly 900 MHz–30 GHz communications [9, 10]. My work is also to design a compress size microstrip antenna by subtracting two rectangular slots with unequal dimensions and one circular slot from the patch on top layer to increase S11 parameter and gain of antenna with ratio of resonant frequency. For compressing the size we require large values of dielectric constant [11, 12]. Our final achievement is to compress size of antenna with large operational bandwidth. The simulation was performed by IE3D [13] using Method of Moment. Due to compress in nature this antenna is applicable for satellite systems and microwave relay systems.

2 Antenna Structure Proposed antenna designed by cutting two unequal rectangular and one circular slot from the patch on the top layer which is shown in Fig. 1 and displayed with the PTFE substrate. Dimensions and position of the radius = 0.8 mm are indicated in figures. The insulation materials specified here are epoxy polytetrafluroethelene based substrate with FR4. The feeding point for co-axial is given at point (1.5, −1) where the correction center is located at (0, 0). The designed antenna of top layer is shown in Fig. 1.

Fig. 1. Top view of designed antenna

The antenna designed with a same PTFE insulation material. The antenna designed with rectangular patch of 10 mm  12 mm.

Size Deduced Printed Antenna for W-LAN Applications

99

3 Analysis with Simulated Results Now, a different study of the designed antenna is performed and shown in this section. Detailed analysis of parameters for designed antenna analyzed for bandwidth, gains and S11 parameter of antenna. The designed antenna’s simulated return loss is displayed in Fig. 2. From the advantage of MSA, it is one advantage that the MSA must be operating for more than one resonant frequency. So, according to MSA design rules, we are design the proposed antenna and it is verifying the characteristics of MSA.

Fig. 2. Designed antenna return loss

As slots are cut in the correct position of the antenna, the resonance frequency obtained with large frequency ratio with a deep return loss. The 1st resonance obtained at f1 = 5.60 GHz with −18.64 dB of reflection co-efficient. The next resonance obtained at f2 = 6.61 GHz with −13.50 dB of reflection co-efficient. The 10 dB corresponding width obtained for the proposed antenna in f1 and f2 are 115.93 MHz and 33.81 MHz, respectively. 3.1

Simulated Radiation Pattern

The E-plane and H-plane radiation patterns for proposed antenna are displayed in Figs. 3, 4, 5 and 6 for all the resonance (Tables 1 and 2).

100

P. Ravali et al.

Fig. 3. Simulated radiation pattern for E-field at 5.60 GHz

Fig. 4. Simulated radiation pattern for H-field at 5.60 GHz

Size Deduced Printed Antenna for W-LAN Applications

Fig. 5. Simulated radiation pattern for E-field at 6.61 GHz

Fig. 6. Simulated radiation pattern for H-field at 6.61 GHz

101

102

P. Ravali et al. Table 1. Analyzed radiation pattern for designed antenna

Structure Frequency of of antenna resonanance (GHz) Designed f1 = 5.60 f2 = 6.61

Ratio of resonance frequency f2/f1 = 1.18

Beam width of half power (deg)

Gain with respect to isotropic antenna (dBi)

170.58° 165.79°

5.07 2.08

Table 2. Analyzed return loss for designed antenna Structure of antenna Designed

Frequency of resonanance (GHz) f1 = 5.60 f2 = 6.61

S11 parameter (dB) −18.64 −13.50

10 DB width (MHz) 115.93 33.81

4 Conclusion Single co-axial feed with one sided compress printed antenna whose theoretical investigations were performed using a MoM based software IE3D. When cutting two unequal rectangular and one circular slot from the patch of the top layer, the significant improvements achieved with −18.64 dB of S11 parameter in dB. Other observations for designed antenna are three-dimensional half power beam width of about 170.58°, is wide beam for intended applications. By changing the feed point in designed antenna, we achieved a low bandwidth of 10 dB with low signals. In 6.61 GHz, we can get two types of polarization i.e. co-polarization and cross polarization. At 0° no power is radiated by the proposed antenna. So, at 0° we get a output zero and we achieve a conical shape. All the values are written in this paper is correct and there will be no modification is required. Acknowledgement. All members are grateful for the support provided by JITS-1, Nustulapur, Karimnagar and all faculty members of the ECE department to complete this paper successfully.

References 1. Sarkar I, Sarkar PP, Chowdhury SK (2009) A new compact printed antenna for mobile communication. In: 2009 Loughborough antennas & propagation conference, Loughborough, UK, 16–17 November 2009 2. Chatterjee S, Paul J, Ghosh K, Sarkar PP, Chanda (Sarkar) D, Chowdhury SK (2011) A compact microstrip antenna for WLAN communication. In: National conference of electronics, communication and signal processing, Paper ID: 116 3. Chakraborty U, Chatterjee S, Chowdhury SK, Sarkar PP (2010) Triangular slot microstrip patch antenna for mobile communication. In: 2010 annual IEEE India conference (INDICON), Paper ID: 511, pp 4–7

Size Deduced Printed Antenna for W-LAN Applications

103

4. Jan J-Y, Tseng L-C (2004) Small planar monopole antenna with a shorted parasitic invertedL wire for wireless communications in the 2.4, 5.2 and 5.8 GHz bands. IEEE Trans Antennas Propag 52(7):1903–1905 5. Chatterjee S, Chakraborty U, Sarkar I, Chowdhury SK, Sarkar PP (2010) A compact microstrip antenna for mobile communication. In: 2010 annual IEEE India conference (INDICON), Paper ID: 510, pp 1–3 6. Danideh A, Fakhr RS, Hassani HR (2008) Wideband coplanar microstrip patch antenna. Prog Electromagn Res Lett (PIER) 4:81–89 7. Chatterjee S, Paul J, Ghosh K, Sarkar PP, Chowdhury SK (2011) A printed patch antenna for mobile communication. In: Convergence of optics and electronics conference, Paper ID: 15, pp 102–107 8. Bahl J, Bhartia P (1980) Microstrip antennas. Artech House, Dedham 9. Chakraborty U, Chatterjee S, Chowdhury SK, Sarkar PP (2011) A compact microstrip patch antenna for wireless communication. Prog. Electromagn Res C 18:211–220 10. Fallahi R, Kalteh A-A, Roozbahani MG (2008) A novel UWB elliptical slot antenna with band-notched characteristics. Prog. Electromagn Res C 82:127–136 11. Chatterjee S, Chowdhury SK, Sarkar PP, Sarkar DC (2013) Compact microstrip patch antenna for microwave communication. Indian J Pure Appl Phys 51:800–807 12. Balanis CA (1989) Advanced engineering electromagnetic. Wiley, New York 13. Zeland Software Inc. IE3D: MOM-based EM simulator. http://www.zeland.com

Design of 31.2/40.1667 GHz Dual Band Antenna for Future mmwave 5G Femtocell Access Point Applications V. Harini1,2(&), M. V. S. Sairam3, and R. Madhu1 1

Department of ECE, JNTU Kakinada, Kakinada, AP, India [email protected] 2 Department of ECE, Vardhaman College of Engineering, Hyderabad, Telangana, India 3 Department of ECE, G.V.P. College of Engineering (Autonomous), Visakhapatnam, AP, India

Abstract. For 5G Femtocell Access Point applications, a high gain antenna is proposed which is used in mm-wave communications. Based on reflection coefficient, gain, E-field distribution, radiation pattern the performance of this antenna is analyzed when once some new rectangular and circular slots are introduced within the patch at each frequency respectively. The dimensions of the patch are given by 4.9 mm
4.4 mm. A Rogers RT duriod 5880 with a dielectric constant of 2.2 and thickness of the substrate is 0.508 mm is used for fabrication of this antenna. The proposed antenna is radiated two resonating frequencies 31.2 GHz and 40.1667 GHz and attained a gain of 6.632 dB and 7.34 dB correspondingly which is able to cover 5G applications. The proposed antenna design may well be appropriate for mm-wave 5G femtocell access point applications. Keywords: Dual-band antenna

 Femtocells  Millimeter wave  5G

1 Introduction Femtocell, additionally called Femtocell Access Point (FAP), may be a short-ranged, weak, low cost base station. Femtocell is comparable to a wireless net router and straight forward to put in offices and residences. It’s a mini base station for the indoor coverage purpose and an extension of out of doors network [1]. A particular helpful application of indoor system antennas is the plan of distributed antennas. A sensible ceiling mounted antennas for indoor coverage want necessities of terribly wide beam width, consistent with a distinct appearance, thus this specific antenna has been designed to appear kind of like a smoke detector [2]. Microstrip printed patch antennas are more attractive for indoor system design antennas which have uniform coverage. Figure 1 describes proposed frequency bands for 5G communications from 20 GHz to 50 GHz. Future era works with mmwave band because of high capability, high data rate and lower latency mostly demanded by normal mobile users [3]. To achieve this, a novel patch antenna is designed which give high gain and efficiency [4, 5]. To integrate the proposed bands in a single device, a development of © Springer Nature Switzerland AG 2020 S. C. Satapathy et al. (Eds.): ICETE 2019, LAIS 4, pp. 104–111, 2020. https://doi.org/10.1007/978-3-030-24318-0_13

Design of 31.2/40.1667 GHz Dual Band Antenna

105

Fig. 1. Frequency bands for 5G communications

dual-band antenna has been made. In a contemporary literature, it’s shown that, one band antenna is intended which will operate only at one frequency that is 28 GHz having a gain of 6.92 dB [6]. The antenna impedance and bandwidth covers the 28 GHz mmwave band however the scale of the antenna was high relatively with planned antenna [7]. A 28 GHz band antenna is designed to handle losses of air substrates and which also deals with the effects of low relative permittivity substrates [8, 9]. A 24/38 GHz dual-band antenna is proposed to work at high gain for future wireless communication. The structure of paper describes as follows, Sect. 2 narrates about the design of antenna, and Sect. 3 sketches the results in detail. Finally, Sect. 4 concludes the design and its applications.

2 Antenna Design The geometry of mmwave patch antenna size is 5.9 mm
8.6 mm and is designed using Rogers RT duriod 5880 substrate material with er ¼ 2:2 and a dielectric loss tangent of 0.0009 with the thickness of the substrate ℎ = 0.02” (0.508 mm). The design is done in commercially available Ansoft HFSS version 15. Figure 2 shows the proposed antenna structure. There is a rectangular slot on the top most side of patch which has a major effect in generation of mmwave frequency band. Two rectangular slots along with one circular slot in center have introduced in the middle of the patch so that it gives another frequency band. In this simple edge feed excitation has been used. The fundamental design procedure of antenna is given as follows The width of patch antenna 1 Wp ¼ pffiffiffiffiffiffiffiffiffi 2f res e0 l0

rffiffiffiffiffiffiffiffiffiffiffiffiffi rffiffiffiffiffiffiffiffiffiffiffiffiffi 2 c 2 ¼ ere þ 1 2f res ere þ 1

Effective dielectric constant ereff ¼ ere 2þ 1 þ qffiffiffiffiffiffi 1 Effective Length Leff ¼ 2fc ereff

ere  1 2 ½1

1 12h 2 þ W  p

res

Wp þ 0:264Þ hs The extension length h ¼ Wp þ 8Þ s ðereff  0:258Þð hs Actual length of patch L ¼ Leff  2DL Ground plane dimensions Lgnd = 6hs + L; Wgnd ¼ 6hs þ Wp where hs = thickness of substrate; L = patch length; Leff = effective length. The front and side view of the antenna is demonstrated in Figs. 2 and 3. The proposed mmwave antenna dimensions are given in Table 1. DL

0:412ðereff þ 3Þð

106

V. Harini et al.

L ss Lp Ls1 Ws1

L ups Wp

R

W ups

W ss

L dns Wdns

Ws2 Ls2

W fl Lfl

Fig. 2. Geometry of proposed mmwave antenna

Fig. 3. Side view of proposed mmwave antenna

The proposed antenna is designed with the following steps, In the first step only one slot at the top most side of the patch is inserted with dimensions 4.1 mm
0.6 mm as shown in below Fig. 4. And achieved frequency 43.333 GHz with S11 = −17.91 dB.

Design of 31.2/40.1667 GHz Dual Band Antenna

107

Table 1. Design parameters of proposed mmwave antenna. Parameter LG/LSS LP LS1 L ups L dns LS2 L fl R

Value (mm) Parameter 5.9 WG/WSS 4.9 WP 4.1 WS1 3 W ups 3 W dns 0.45 WS2 0.4 W fl 0.5

Value (mm) 8.6 4.4 0.6 0.5 0.5 0.6 4

Fig. 4. Antenna with one top edge slot and its S11 plot

In the second step, a upper rectangular slot with dimensions 3 mm
0.5 mm is inserted and achieved two frequencies 40.233 GHz with S11 = −16.58 dB and 30.933 GHz with S11 = −18.96 dB as shown in Fig. 5.

Fig. 5. Antenna with one rectangular slot and its S11 plot

108

V. Harini et al.

In the third step, a lower rectangular slot with dimensions 3 mm
0.5 mm is inserted and achieved a frequency 38.8 GHz with S11 = −14.86 dB as shown in Fig. 6. Actually this is a good design but only single frequency is able to achieve which has been done in various literatures. So to attain dual bands, a circular slot was inserted in between rectangular slots.

Fig. 6. Antenna with two rectangular slots and its S11 plot

In the final step, a circular slot with radius 05 mm and achieved required frequencies 31.2 GHz with S11 = −16.9 dB and 40.1667 GHz with S11 = −14.98 dB as shown in Fig. 7.

Fig. 7. Antenna with two rectangular slots and one circular slot and its S11 plot

3 Results and Discussion The performance of the mmwave antenna is evaluated in terms of reflection coefficient (S11), radiation characteristics, gain, efficiency and surface current distributions. Voltage Standing Wave Ratio (VSWR) describes the efficient use of RF power transmitted from source. The impedance mismatch is represented by reflection coefficient at the load. The negative logarithmic value of reflection coefficient gives return loss. For the sensible applications, VSWR = 2 is suitable because the return loss would be −9.54 dB or −10 dB.

Design of 31.2/40.1667 GHz Dual Band Antenna

109

Based on various inclusions of slots step by step as discussed in antenna design, a design has been finalized which has reflection coefficients crossed at two different frequencies 31.2 GHz having a return loss of −16.9052 dB with a bandwidth of 200 MHz and 40.1667 GHz having a return loss of −14.98 dB with a bandwidth of 1.033 GHz, and these two bands are covered in 5G candidate bands. The simulated return loss is shown in Fig. 8.

Fig. 8. Reflection Coefficient (S11 dB) of proposed mmwave antenna

The comparison of various S11 in dB is shown in Fig. 9, where inclusion of various slots has been done.

Fig. 9. Comparison of S11 in dB of various slots in proposed mmwave antenna

The simulated elevation radiation pattern for the designed antenna at the two resonant frequencies 31.2 GHz and 40.1667 GHz are illustrated in Fig. 10 to Fig. 11. It is observed from the figures that the radiation patterns are stable at the two bands and are almost produce directional pattern. The simulated gain and efficiency for designed antenna is plotted in Figs. 12 and 13. At 31.2 GHz and 40.1667 GHz the doable gain and efficiencies of proposed antenna is 6.632 dB, 99.45% and 7.34 dB, 110.35% respectively. The Electric field distributions of the proposed antenna are mainly contributed by upper rectangular slot at two resonating frequencies as shown in Fig. 14.

110

V. Harini et al.

Fig. 10. Simulated Gain and radiation Pattern of proposed antenna @31.2 GHz.

Fig. 11. Simulated Gain and radiation Pattern of proposed antenna @40.1667 GHz

Fig. 12. Simulated Gain vs frequency

Fig. 13. Simulated Efficiency vs frequency

Fig. 14. Surface current distribution @31.2 GHz and 40.1667 GHz for the proposed mm wave antenna

Design of 31.2/40.1667 GHz Dual Band Antenna

111

4 Conclusion A high gain dual band antenna is designed, simulated and results are observed at 31.2 GHz and 40.1667 GHz frequencies which are compatible with desired 5G frequency bands (20 GHz to 50 GHz). The simulated results are more effective in terms of gain, efficiency, current distribution and reflection coefficient parameters. Elevated radiation pattern of proposed antenna describes a directional pattern having high peak gain at a particular direction. Directionality leads to reduction in interference in femtocell networks. The antenna features a sensible result at frequencies 31.2 GHz having a return loss of −16.9052 dB with a bandwidth of 200 MHz and 40.1667 GHz having a return loss of −14.98 dB with a bandwidth of 1.033 GHz, and these two bands are lined in 5G candidate bands Sensible performance, easy configuration and low profile create this antenna significantly enticing for transportable furthermore as fastened communication devices, together with femtocell access point which has same characteristics of normal base station which is used in mobile communications.

References 1. Chandrasekhar V, Andrews JG, Gatherer A (2008) Femtocell networks: a survey. IEEE Commun Mag 46(9):59–67 2. Chow P, Karim A, Fung V, Dietrich C (1994) Performance advantages of distributed antennas in indoor wireless communication systems. In: Proceedings of IEEE vehicular technology society conference, vol 3, pp 1522–1526 3. Şeker C, Güneşer MT (2018) A single band antenna design for future millimeter wave wireless communication at 38Ghz. In: ICEF III, international conference on engineering and formal sciences, Amsterdam, 11–12 May 2018, pp 36–40 4. Haraz OM, Ali MMM, Alshebeili S, Sebak AR (2015) Design of a 28/38 GHz dual-band printed slot antenna for the future 5G mobile communication Networks. In: 2015 IEEE international symposium on antennas and propagation & USNC/URSI national radio science meeting. IEEE, July 2015, pp 1532–1533 5. Gampala G, Reddy CJ (2016) Design of millimeter wave antenna arrays for 5G cellular applications using FEKO. In: 2016 IEEE/ACES international conference on wireless information technology and systems (ICWITS) and applied computational electromagnetics (ACES). IEEE, March 2016, pp 1–2 6. EL_Mashade MB, Hegazy EA (2018) Design and analysis of 28 GHz rectangular microstrip patch array antenna. WSEAS Trans Commun 17:1–9 7. Morshed KM, Esselle KP, Heimlich M (2016) Dielectric loaded planar inverted-F antenna for millimeter-wave 5G hand held devices. In: 2016 10th European conference on antennas and propagation (EuCAP). IEEE, April 2016, pp 1–3 8. Haraz OM, Ashraf M, Alshebeili S (2015) Single-band PIFA MIMO antenna system design for future 5G wireless communication applications. In: 2015 IEEE 11th international conference on wireless and mobile computing, networking and communications (WiMob), Abu Dhabi, United Arab Emirates, October 2015, pp 608–612 9. Outerelo DA, Alejos AV, Sanchez MG, Isasa MV (2015) Microstrip antenna for 5G broadband communications: overview of design issues. In: 2015 IEEE international symposium on antennas and propagation & USNC/URSI national radio science meeting. IEEE, July 2015, pp 2443–2444

Implementation of Master – Slave Communication for Smart Meter Using 6LOWPAN Navya Namratha Doppala(&) and Ameet Chavan Department of ECE, Sreenidhi Institute of Science and Technology, SNIST, Yamnampet, Ghatkesar, Hyderabad 501301, Telangana, India [email protected], [email protected]

Abstract. Advanced electrical energy reading device, which collects data using various serial ports and transfers acquired data to a server using 2G/3G communication technologies, has gained global popularity. Limitations with traditional techniques are periodical trips followed by physical data collection implemented through a physical cable connection. This paper presents a Smart meter which uses 6LoWPAN to collect the data wirelessly and implemented as a Master-Slave system. The System controller will decode the data collected through slaves. The decoded data will be transferred to the server using GPRS (2G/3G) technology. This device will overcome the limitations posed by the advanced electrical energy reading device like the physical presence of utility personnel and increasing cost expenses of wired connectivity. Keywords: IPv6


RPL
Smart meter
6LoWPAN
IoT

1 Introduction Electricity meter operates continuously measuring the voltage and current to give energy (used in joules, KWH). Meters for smaller services can be connected inline between source and customer whereas meters for huge loads greater than 200 A, the current transformers are used (it may be located other than in-line). Traditional electromechanical meters uses the principle of electromagnetic induction to measure energy consumption. The next generation electronic meters use the metering engine to show through an LCD display. In the view of 2G and 3G communication, optical sensors are connected directly to the meter which includes periodic trips, increasing the energy usage of each reading and loss of cost for the domestic users. A new idea to simplify the measurement of energy usage for each meter and to eliminate the wired connectivity and periodic trips, the concept of the smart meter with 6LoWPAN is introduced. Smart meter collects the data, diagnosis it and the status will be transferred to the central database for further process like billing and analysing the parameters. Whereas 6LoWPAN have one or more nodes connected in a network shares the same Internet Protocol version 6 which in turn expands up to 2^128 IP addresses that are equal to 50 octillion addresses [1].

© Springer Nature Switzerland AG 2020 S. C. Satapathy et al. (Eds.): ICETE 2019, LAIS 4, pp. 112–118, 2020. https://doi.org/10.1007/978-3-030-24318-0_14

Implementation of Master – Slave Communication for Smart Meter

113

2 Background Work The existing techniques were studied and the limitations of each paper are briefly explained. [2] The Automatic meter reading system which is based on GSM for immediate billing, explains the use of electricity in the energy meter. This was developed in Microsoft.Net framework. The meter is enabled with the GSM module. It consists of GSM modem on one end and GSM receiver on the other end to collect the data periodically. The limitations in this method are power factor improvement and no initialization message option while sending the data. [3] Proposes a system of digital meter to find out the accurate reading by using Image capturing technique. It involves the user to capture an image of Optical Character Recognition. Additionally cropping of the image and resizing can be done by the software. These captured data is sent to the server and it will generate the amount of bill which communicates through online. The main drawback is limited internet connectivity. It also requires 2 MP to HD camera efficiency for stability and the same android phone should be registered for the application. [4] Discusses the traditional technique of the electromechanical energy meter. The manual reading of power consumption data is made available with the power line. It involves a conductor to carry the data which can later be used for the AC electric power distribution. Before the processing of data, the PLC transmitter holds the data and sends it to the PLC receiver end. The latency and complexity of wired connections are the main drawbacks. [5] The aim of this method is to reduce such a tedious job by the process of collecting data from the consumer’s end. To ensure the accuracy of energy consumption per meter, a GSM modem is integrated along with the meter to send alerts to the particular consumer. There is an option of disconnecting the power supply by the controller if in case the payment is not done by the user. [6] It presents the automatic meter reading of data from consumer end with the help of Zigbee and GSM module. The system mainly consists of two units one is individual consumer unit and the other in the central office section. The consumer unit contains the GSM module and the main server contains Zigbee modem. The wireless sensor network in between both modules gives the energy reading of the meter. The transmission distance is of 85 m. Hence Zigbee is the good option for low data rate transmission but the drawback is limited connectivity of nodes.

3 Proposed System Architecture The authors proposed to design a SMART METER, which uses a 6LoWPAN to collect the wireless data. The architecture consists of a 6LoWPAN Master-Slave system, system controller, power management IC, General Packet Radio Service (GPRS), RAM and SD card. The 6LoWPAN module is inbuilt in the Master and Slave units. The 6LoWPAN comprises of Microcontroller Unit along with RF transceiver which is useful for sending and receiving the data on RF band. Figure 1 shows the block diagram of the proposed system.

114

N. N. Doppala and A. Chavan

Fig. 1. Block diagram of the proposed system

3.1

6LoWPAN Module for Proposed System

A. System Stack – The main capability of 6LoWPAN is to operate in star and mesh network topology with the 64bit addressing modes. The master devices route the data mapped to a slave whereas the slaves are like end devices which enables to route data to other devices. While considering the Open System Interconnection (OSI) stack, 6LoWPAN makes a tremendous difference by including the adaptation layer in between the Media Access Layer (MAC) and Network layer. The need of adaptation layer is to allow the transmission of IPv6 datagram over IEEE 802.1.5.4 radio links by header compression, fragmentation & reassembly and stateless autoconfiguration. The network routing in a Master-Slave system is controlled by an IETF standard called Routing Protocol for Low power and Lossy networks (RPL) [7–11]. B. Hardware Design – The hardware architecture comprises of CC1310 Wireless Microcontroller Unit and very low power RF transceiver (RF core). The powerful 48 MHz cortex–M3 CPU in the platform supports multiple RF Standards and physical layers. The RF core will autonomously handle the time-critical aspects of radio protocol to take the load off the main CPU and leaves more resources for user application. A dedicated radio controller ARM cortex-M0 handles low-level RF protocol command that will be stored in RAM or ROM. This MCU includes a sensor controller which can monitor the sensors or performs various tasks simultaneously thus certainly reduces power utilization and offloads the main cortex-M3 CPU. The flash memory arranges non-volatile storage for data and code. The debug is done through JTAG. To ensure the synchronization while communicating in the master and slave devices the MCU units are in-built from the manufacturer [12].

Implementation of Master – Slave Communication for Smart Meter

3.2

115

System Controller (ARM Cortex A8 Processor)

The Microprocessor Unit sub-system of the device exchanges the transactions among the ARM Cortex-A8 Processor and interrupt controller (INTC). The MPU sub-system was a hard macro with the additional logic for emulation, protocol conversion, debug enhancements and interrupt handling. The ARM Cortex-A8 has ARMv7 compatibility. It is used for communicating through an AXI bus which allows receiving of an interrupts from this sub-system interrupt controller (MPU INTC). Power Management IC manages the power flow among the various power sources to the power loads. GPRS connects to the GSM network using SIM for data connectivity to send the data to the server and to retrieve data when the request is sent. 3.3

Communication in-between the Master and Slave Units

• Initialize the Serial Port, Baud rate and RF channel for the Meter (slave) communication. • The above mentioned specifications must be similar for both Master and Slave units (meters), if not the connection cannot be established in between them.

(a) Data flow in Master Unit

(b) Data flow in Slave unit

Fig. 2. Shows the data flow in Master and Slave units

116

N. N. Doppala and A. Chavan

• When all the devices like meters and the master unit are powered ON, then RPL starts establishing the connection between them as shown in the Fig. 2. • After a new UDP connection is created between the meter and master unit, they are bind to a specified local port. • Now the connection between the master unit and the meter is established successfully and it said to be ready for exchanging of data. • An application will be run in the master unit. • We can get the response from the connected meter (slave node) by placing the request through the commands as specified in the application.

4 GUI Application and Results It is more important for consumers to know about the availability of energy consumption. So there is an extendable usage of creating an application which gives the information of energy usage in KWH with the certain meter ID. The enhancement of the GUI application is to be implemented for this system. The mathematical expression for the traditional technique of electromechanical meters is given by Power, P¼

3600  Kh t

ð1Þ

Where Kh denotes one revolution of the disc in the unit of watt-hour per revolution, t is denoted for the time taken by the disc for one revolution. Whereas the Smart meters used in this project are Single phase Static Watt-Hour Meter which is also a 6LoWPAN LPRF communication device. As per our convenience, the energy consumption data is recorded for every 60 s which will be stored in the Load register of MCU. The data from the slave units are sent to the master unit [13] which is acknowledged. The Calculation is made to get the meter reading at the end of every month. The remaining energy consumed by a meter is given by the expression Z

t

E ðtÞ ¼

ðPsðtÞ  PcðtÞÞ:dt

ð2Þ

0

Where E(t) is remaining energy of the slave unit, Ps(t) is total output power from the energy source at time t, Pc(t) is total energy consumption for Ps(t) at same time t. To display the result through a handheld device (Master Unit). We need to develop an application for consumer convenience. The meter reading in the master unit will be displayed as seen in the Fig. 3 (screenshots). 1. Begin the application LPR module and select the Main menu and go to SCAN option to display the available meters around the network.

Implementation of Master – Slave Communication for Smart Meter

117

2. Then the meters which are available is shown. By selecting the meter ID it will generate the energy consumption details of that particular meter. 3. Upload the data by selecting the UPLOAD DATA option, it will send the data to the server. Then the data is uploaded successfully.

(a) Main menu in LPR module application

(c) Display the Power consumption in KWH

(e) Upload request

(b) Display of Meter IDs

(d) Uploading the data to the server

(f) Data uploaded to the server

Fig. 3. Screenshots of meter reading at the master unit

118

4.1

N. N. Doppala and A. Chavan

Advantages and Limitations of the Proposed System

Accurate billing on a real-time basis is being implemented. By amending the safety risk and to extend the range of devices in a network, one can easily opt for the smart meters with 6LoWPAN. The limitation in this system is health and safety concerns. In a research-oriented for the health of a person, the risk factor is high due to the RF radiation emitted by wireless smart meters.

5 Conclusion To ensure higher levels of real time service, reduction in latency and cost effectiveness the proposed work has used IoT as the base technology. The proposed work further employs 6LoWPAN wireless technology to complement and meet the typical constraints set for the Internet of Things. In order to develop many nodes in one network either it is star or mesh network, scalability and security have been studied and applied. This further plays an important role in managing the fraud/unauthorized access and raise subscribers in a better way. The security aspect is implemented by generating a unique key and encrypts data with AES encryption. Other main features of the work include simpler software and low complex hardware for the ease of maintenance and service.

References 1. Olsson J (2014) 6LoWPAN demystified. Texas Instruments (TI), October 2014 2. Ashna K, George SN (2013) GSM based automatic energy meter reading system with instant billing. 2013 international multi conference on automation, computing, communication, control and compressed sensing (iMac4s). IEEE Press, June 2013, pp 65–72 3. Ghongade Aniruddha S, Khode Chaitanya R, Darekar Mahesh P, Gaikwad Siddhant S (2017) Consumer app with automatic image capturing and processing for meter reading and billing. Multidiscip J Res Eng Technol 5(4):7598–7601 4. Gunasekaran R, Karthikeyan C, Pavalam J, Mohanapriya P, Preethi A, Indhumathi V (2017) Power line carrier communication using automated meter reading. Bioprocess Eng 1(4):104–109 5. Anjaly Joseph T, Elizabeth P, Bail JP (2017) Wireless automatic meter reading system with power factor correction. Int J Adv Eng Res Dev 4(6):697–702 6. Sreelakshmi RV, Thomas N (2015) Automatic energy meter reading system using Zigbee and GSM. Int J Eng Sci Comput 1212–1215 7. IEEE Std 802.15.4, 2011 specifications for low-rate wireless personal area networks (WPANs). IEEE Standard, June 2011 8. Garg R, Sharma S (2017) A study on need of adaptation layer in 6LoWPAN protocol stack. Int J Wirel Microwave Technol (IJWMT) 7(3):49–57 9. RFC 6550, RPL: IPv6 Routing Protocol for Low-Power and Lossy Networks, Internet Standard, March 2012 10. RFC 7102, RPL: IPv6 Routing Protocol for Low power and Lossy Networks, January 2014 11. Kuryla S (2010) RPL: IPv6 routing protocol for low power and lossy networks. In: Networks and distributed systems seminar, March 2010 12. CC1310 Simplelink Wireless MCU. Texas Instruments (TI), February 2015 13. Musa A (2013) 6LoWPAN- based wireless home automation: from secure system development to building energy management. Smart Comput Rev 3(2)

Dynamic Neural Networks with Semi Empirical Model for Mobile Radio Path Loss Estimation Bhuvaneshwari Achayalingam1(&), Hemalatha Rallapalli2, and Satya Savithri Tirumala3 1

3

Deccan College of Engineering and Technology, Hyderabad 500001, T.S, India [email protected] 2 Osmania University, Hyderabad 500007, T.S, India Jawaharlal Nehru Technological University, Hyderabad 500085, T.S, India

Abstract. The paper evaluates the performance of Focused Time Delay, Distributed Time Delay (DTD), Layer Recurrent, and Nonlinear Autoregressive dynamic neural networks for estimating path loss of mobile radio signals, by training a semi empirical model. The path losses predicted with Walfisch Bertoni, Walfisch-Ikegami (WI), Sakagami and Xia models are compared with the path loss extracted from measured mobile signal strengths at frequency of 947.5 MHz in Hyderabad city of India. The best suited, Walfisch-Ikegami model is trained with feedback neural networks using Levenberg Marquardt, Scale Conjugate Gradient and Resilient Propagation training algorithms. The DTD neural network, trained by Levenberg algorithm has good performance with MSE (3.33) and correlation (0.987). The highlight is the proposed Hybrid-Neural Network Walfisch-Ikegami (H-NNWI) model which implements DTD neural network with a modified Walfisch-Ikegami model. The results prove the efficient performance of proposed model with least MSE (2.73), highest correlation (0.989) and improved path loss estimation. Keywords: Neural network
Path loss
Walfisch-Ikegami model
Focused Time Delay
Distributed Time Delay
Levenberg Marquardt Hybrid-neural network model




1 Introduction The estimation of mobile radio path loss has a significant role in designing RF link budget for new cellular networks or analyzing the existing ones. The path loss describes the average attenuation of signal strengths and is well quantified using propagation models [1]. Empirical models have a simpler implementation, require additional correction factors and lack accuracy. Alternately, deterministic models require excessive computational time, detailed information of the environment and increase the complexity [2]. Therefore semi empirical models are preferred, and are merged with neural network training to achieve precise path loss computation. The © Springer Nature Switzerland AG 2020 S. C. Satapathy et al. (Eds.): ICETE 2019, LAIS 4, pp. 119–129, 2020. https://doi.org/10.1007/978-3-030-24318-0_15

120

B. Achayalingam et al.

neural network models integrate the benefits of empirical and deterministic models and enables processing of large data with high speed. The application of neural networks for path loss estimation can be considered as the nonlinear mapping of input variables such as the propagation distance, terrain, and frequency. The estimated path loss is the output variable on which the input parameters are mapped. In the earlier works, a hybrid model is presented which combines the Cost Walfisch-Ikegami model with an adaptive neural network for propagation loss modeling [3]. Similar hybrid approaches are implemented using Genetic algorithm or error correction models [4, 5]. The major drawback of the conventional model is it does not well describe the non-linear nature of propagation loss in realistic conditions, thus leading to certain prediction errors. This drawback is focused and a neural network trained hybrid model is proposed which minimizes the difference between model estimated and measured path losses. Initially, the semi empirical models are implemented to estimate the mobile radio path loss, and the Walfisch-Ikegami (WI) model is chosen for the neural network implementation. The selection of the WI model is made from the existing models, by comparing its performance with respect to the path loss obtained from measurements. In the neural network implementation, the WI model is trained with FTD, DTD, LRN, NARX neural networks. A comparative analysis is made and DTDNN is proved to be more suitable network. In order to improve the path loss performance, a modified WI model is proposed and trained using Distributed Time Delay Neural Network (DTDNN). The proposed Hybrid-Neural Network Walfisch-Ikegami model (H-NNWI) is evaluated and its efficient performance in improving the accuracy of path loss estimation is statistically verified.

2 Semi Empirical Models for Path Loss Estimation The semi empirical path loss models are preferred as they increase the accuracy of prediction with the available resources and minimize the complexity. The Table 1 gives the commonly used semi empirical models [1]. The Walfish Bertoni, Walfish-Ikegami, Xia and Sakagami models are implemented to estimate the mobile radio path loss. In the next step, a drive test is performed in an urban area, at the city of Hyderabad and mobile signal strengths are recorded at 947.5 MHz downlink frequency, covering ten cellular sites. The experimental details and data collection process are described in detail [7]. Knowing the received signal strengths and base station transmitter power, the path loss is extracted from measurements with suitable pre processing. The measured and model predicted path losses are compared and based on MSE, RMSE and Standard Deviation of error, WalfishIkegami is chosen as the best model for path loss estimation. The next phase is the neural network implementation of the selected model.

Dynamic Neural Networks with Semi Empirical Model Table 1. Semi empirical path loss models

121

122

B. Achayalingam et al.

3 Dynamic Neural Networks The dynamic feedback networks are more powerful networks compared to the feed forward ones since they include a memory and are trained with definite or time varying patterns [8]. The architectures of dynamic neural networks are described in detail [9]. The Neural Network toolbox in Matlab is employed to implement any network if it is arranged in the form of Layered Digital Network (LDDN). The dynamic feedback networks arranged as LDNN are trained with commonly used algorithms such as Levenberg-Marquardt (LM), Scaled Conjugate Gradient (SCG) and Resilient back propagation (RP). The major features of the dynamic neural networks are in Table 2 and features of the algorithms used for training are in Table 3. Table 2. Dynamic neural networks Focused Time Delay (FTDNN)

Distributed Time Delay (DTDNN)

Layer Recurrent Neural Network (LRNN)

Non Linear Auto Regressive Network (NARX)

• The FTDNN comprises of a delay line tapped at the input • The network is designed so that every synapse works as a FIR filter • Most applicable for the prediction of time varying sequences [9] • It trains faster than other dynamic networks • DTDNN is a network of distributed tapped delay lines • It employs the dynamic back propagation • Trains slower compared to FTDNN as delays are present throughout the network • The LRNN generalizes the Elman network • Every layer of the network includes a delay with a feedback loop, with an exception for the last layer [10] • It has arbitrary number of layers and transfer function in each layer • Trained with gradient-based algorithms • Employed in filtering and modeling applications • NARX network belongs to a recurrent category of dynamic Networks • Various layers of the network have feedback connections. [11] • The succeeding output values are regressed on preceding values • The NARX works in Series-parallel mode (SP) or Parallel (P) mode • Suitable to model nonlinear time series systems

Dynamic Neural Networks with Semi Empirical Model

123

Table 3. Neural network training algorithms Levenberg Marquardt (LM)

Scaled Conjugate Gradient (SCG) Resilient back propagation (RP)

P 2 • E (w) is an objective error function. EðwÞ ¼ m i¼1 ei ðwÞ 2 • e2i ¼ ðydi  yi Þ Where ydi is the expected value of output neuron i, yi is the obtained output of neuron with m error terms • New weight vector wk þ 1 wk þ 1 ¼ wk þ dwk • compute E(wk þ dwk ) • The line search for every iteration is replaced by scaling • The steepest descent direction is merged with the preceding search direction to create a new search direction. [12] • Training with RP is faster than training with back propagation • RP eliminates the variations of weights and biases as a result of a small magnitude of the gradient in SCG • The weight and bias are increased if two iterations have the same sign of the derivative • They are reduced if the signs differ from the previous search

4 Hybrid-Neural Network Walfisch-Ikegami Model The step wise procedure for the proposed Hybrid-Neural Network Walfisch-Ikegami (H-NNWI) model is presented. (1) The path loss is calculated using Walfish Ikegami model as [13, 14]
PL ¼

Lo þ Lrts þ Lmsd Lo

forLrts þ Lms  0 forLrts þ Lms\0

ð1Þ

The Lo, Lrts, Lmsd given in the Table 1 are suitably estimated. (2) The Walfish Ikegami model is trained using Distributed Time Delay Neural Networks and the Root Mean Square Error (RMSE) are computed. pffiffiffiffiffiffiffiffiffiffi 1 RMSE ¼ MSE ¼ N

rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi XN ½d  yi 2 i¼1 i

ð2Þ

‘N’ is the number of observations, di is the neural network predicted path loss and yi is the measured output. The DTDNN gives good performance compared to other networks and hence selected for training. (3) In the next step, the Walfish Ikegami (WI) path loss model is modified as PLðdBÞModified ¼ PLðdBÞWalfischIkegami  RMSEðDTDNNÞ

ð3Þ

PLðdBÞModified ¼ Lo0 þ Lrts0 þ Lmsd0

ð4Þ

The path loss for the modified WI model is estimated as per Eq. 3. (4) The modified Walfish Ikegami model ðPLðdBÞModified Þ is trained using DTDNN, resulting in the Hybrid Neural Network trained WI model (H-NNWI). The LM

124

B. Achayalingam et al.

algorithm is employed for training the DTDNN and error function to be minimized is given as PLðdBÞHNNWI ¼ E2i ¼ ðPLmeasured;i  PLModified;i Þ2

ð5Þ

The proposed Hybrid Neural Network WI model is implemented and verified.

5 Results and Discussions The Walfisch-Bertoni (WB), Walfisch-Ikegami (WI), Xia and Sakagami semi empirical models are implemented in Matlab to determine the path losses. The major simulation parameters are given in Table 4 and Fig. 1 shows the measured and model estimated path losses. Table 4. Simulation parameters Parameters Transmitter power of base station (PT) Transmitter gain Effective Isotropic Radiated Power (EIRP) Frequency (downlink) (f) Transmitter antenna height (hb) Receiver antenna height (hm) Street width Building height (hroof) Street orientation angle (/) Tx-Rx distance

Values 43.01 dBm 11.78 dBm 54.8 dBm 947.5 MHz 35 m 1.5 m 7m 24 m 30o d

Fig. 1. Semi-empirical path loss models

Constraints 4 < hb  70 1 < hm  3 10 < w  25 hroof < hb 0 < /  90o 0.02 km–5 km

Dynamic Neural Networks with Semi Empirical Model

125

The mobile radio field strengths collected in the urban region are processed by computing the local mean at typical lengths between 10k to 40k. The averaging ensures the removal of fast variation components to extract the measured path loss. Comparing the simulations with the measured path losses, it is seen that the Walfisch Ikegami (WI) model performs better than other models. The error metrics of MSE, RMSE and Standard Deviation of Error (SD) are computed and results are shown in Table 5. Table 5. Error metrics Metrics Semi empirical path loss Walfisch Bertoni model MSE 1.212 RMSE 1.013 SD 1.095

models Walfisch Ikegami model Xia model 0.518 1.742 0.888 1.320 0.884 1.313

Sakagami model 1.635 1.278 1.272

Analysing the results, it is inferred that the WI path loss model has lower errors, in comparison with other models. Therefore, the WI model is chosen for neural network implementation to determine the path loss. 5.1

Comparison of Dynamic Neural Networks

The implementation process consists of creating a neural network in a predictive mode using the path loss estimated with the Walfisch-Ikegami model as input and the path loss obtained from measurements as the target. Initially, the inputs and desired outputs are converted to a standard neural network cell array to have zero mean and unit standard deviation. An appropriate training algorithm is selected and neural network prediction is done. The goal is to reduce the error between the target and trained neural network output. The WI model is trained with FDTN, DTDNN, LRNN and NARX dynamic neural networks for 100 epochs at a frequency of 947.5 MHz. The Levenberg, Scale Conjugate and Resilient back propagation training algorithms are compared. The performances are evaluated, in terms of MSE, correlation, hidden layer neurons and computation time. The results are given in Tables 6, 7, 8 and 9. Table 6. Focused time delay neural network Training Algorithms LevenbergMarquardt (LM) Neurons 10 20 30 Correlation 0.953 0.973 0.985 MSE 11.3 6.52 3.72 Time(s) 1.81 2.64 3.14

Scale conjugate algorithm (SCG) 10 20 30 0.834 0.861 0.865 33.9 32.0 31.3 3.13 4.91 6.87

Resilient-back propagation (RP) 10 20 30 0.842 0.858 0.863 36.1 32.8 31.7 1.71 2.73 3.74

126

B. Achayalingam et al. Table 7. Distributed time delay neural network Training Algorithms LevenbergMarquardt (LM) Neurons 10 20 30 Correlation 0.958 0.970 0.987 MSE 10.6 7.49 3.33 Time (s) 3.03 8.43 8.97

Scale conjugate algorithm (SCG) 10 20 30 0.854 0.864 0.899 34.9 32.8 24.8 3.10 5.18 7.33

Resilient-back propagation (RP) 10 20 30 0.853 0.858 0.892 35.1 33.9 26.2 1.76 2.87 3.98

Table 8. Layer recurrent neural network Training Algorithms LevenbergMarquardt (LM) Neurons 10 20 30 Correlation 0.945 0.959 0.980 MSE 13.9 10.4 5.12 Time(s) 2.82 16.06 73.47

Scale conjugate algorithm (SCG) 10 20 30 0.849 0.857 0.875 36.9 34.3 30.3 3.52 6.20 8.78

Resilient-back propagation (RP) 10 20 30 0.841 0.843 0.859 38.0 37.5 33.8 2.08 3.40 4.72

Table 9. Non linear auto regressive neural network Training Algorithms LevenbergMarquardt (LM) Neurons 10 20 30 Correlation 0.938 0.961 0.975 MSE 15.4 9.76 6.35 Time(s) 3.99 4.48 5.03

Scale conjugate algorithm (SCG) 10 20 30 0.860 0.873 0.883 33.6 30.7 28.3 7.51 13.13 17.93

Resilient-back propagation (RP) 10 20 30 0.873 0.880 0.896 30.7 29.1 25.2 4.54 7.23 9.61

Analyzing the results, it is found that with the increase in the hidden layer neurons, MSE value decreases and the correlation between the estimated and measured path loss is increased. This is obtained with increased time of computation. Inferring the training algorithms, it is found that LM is the most suited one compared to SCG and RP, as it reduces the MSE and improves the correlation for the implemented neural networks. Comparing the performances of neural networks, it is observed that the DTDNN has the best performance with a least MSE (3.33) and a highest correlation (0.987), for training done with 30 hidden layer neurons. The computation time in DTDNN is marginally increased with respect to FTDNN as the delays are distributed throughout the network compared to a single input delay for FTDNN. The NARX performs better than LRNN, but the computation time is increased. Each layer connection of LRNN has a tap delay involved with it and the presence of the feedback loop increases the delay. The overall comparison suggests DTDNN as the preferred network for training the path loss model. The training, regression plot and the time series response of DTDNN for the best performance are given in Figs. 2, 3 and 4 respectively.

Dynamic Neural Networks with Semi Empirical Model

Fig. 2. Training performance

127

Fig. 3. Regression plot

Fig. 4. Time series response of DTDNN for Walfisch Ikegami model

5.2

Hybrid-Neural Network Walfisch-Ikegami Model

The proposed Hybrid-Neural Network Walfisch-Ikegami (H-NNWI) model is implemented with DTDNN using Levenberg algorithm. The experimentally determined path loss is compared with the path loss obtained from the Walfisch Ikegami model and the suggested hybrid model as depicted in Fig. 5. The original WI model has a standard deviation of error (0.884 dB), whereas the H-NNWI model gives (0.865 dB). There is a reduction of 0.019 dB in the standard deviation of error for the proposed model. It can be seen that the Modified Walfisch Ikegami estimates the path loss in a better way than the Walfisch Ikegami model. The neural network performance of WI and proposed H-NNWI model is in Table 10.

128

B. Achayalingam et al.

Fig. 5. Hybrid Neural network WI model

Table 10. Neural network performance Path Loss models Correlation MSE Time (s) Walfisch Ikegami (WI) 0.987 3.33 18.97 Hybrid-Neural Network W-I model 0.989 2.73 18.02

From the results, it is concluded that the proposed model has an improved path loss performance with the least MSE (2.73), appreciable correlation (0.989) and a reduced computation time. The suitable neural network implementation merged with the traditional semi empirical modeling improves the accuracy of path loss estimation.

6 Conclusion The paper evaluates the performance of Focused Time Delay, Distributed Time Delay, Layer Recurrent and Nonlinear Autoregressive dynamic neural networks in training a semi empirical Walfisch Ikegami model for mobile radio path loss estimation. The selection of the WI model for neural network implementation is justified by comparing the path loss obtained from measurements. The performance comparisons prove that Distributed Time Delay neural network implemented with Levenberg-Marquardt training algorithm has the best performance with MSE (3.33) and correlation (0.987). The proposed Hybrid-Neural Network Walfisch Ikegami model improves the accuracy of path loss estimation. The least MSE (2.73), the highest correlation (0.989) between the model estimated and measured path losses and a reduced computation time proves the effective performance of the suggested Hybrid neural network model for estimating the path loss.

Dynamic Neural Networks with Semi Empirical Model

129

References 1. Saunders S, Aragon-Zavala A (2007) Antennas and propagation for wireless communication systems. Wiley, Hoboken 2. Lasisi H (2017) Development of a propagation path loss prediction model for mobile communication networks deployment in Osogbo, Nigeria. Eur J Eng Res Sci 2(11):13–17 3. Gschwendtner BE, Landstorfer FM (1995) Adaptive propagation modeling using hybrid neural technique. Electron Lett 32(3):162–164 4. Yuen CW, Wong WK, Qian SQ, Chan LK (2009) Fung EHK (2009) A hybrid model using genetic algorithm and neural network for classifying garment defects. Expert Syst Appl 36 (2):2037–2047 5. Cavalcanti BJ, Cavalcante GA, de Mendonça LM, Cantanhede GM, de Oliveira MMM, D’Assunção AG (2017) A hybrid path loss prediction model based on artificial neural networks using empirical models for LTE and LTE-A at 800 MHz and 2600 MHz. J Microwaves Optoelectron Electromagn Appl 16(3):708–722 6. Xia HH (1997) A simplified analytical model for predicting path loss in urban and suburban environments. IEEE Trans Veh Technol 46(4):1040–1046 7. Bhuvaneshwari A, Hemalatha R, Satyasavithri T (2013) Development of an empirical power model and path loss investigations for dense urban region in Southern India. In: 2013 IEEE Malaysia international conference on communications (MICC). IEEE, pp. 500–505 8. Chiang Y-M, Chang L-C, Chang F-J (2004) Comparison of static-feed forward and dynamic-feedback neural networks for rainfall–runoff modeling. J Hydrol 290(3–4):297–311 9. Bhuvaneshwari A, Hemalatha R, Satyasavithri T (2016) Performance evaluation of Dynamic Neural Networks for mobile radio path loss prediction. In: 2016 IEEE Uttar Pradesh section international conference on electrical, computer and electronics engineering (UPCON). IEEE 10. Giles CL, Kuhn GM, Williams RJ (1994) Dynamic recurrent neural networks: theory and applications. IEEE Trans Neural Netw 5:153–156 11. Diaconescu E (2008) The use of NARX neural networks to predict chaotic time series. Wseas Trans Comput Res 3(3):182–191 12. Baghirli O (2015) Comparison of Lavenberg-Marquardt scaled conjugate gradient and Bayesian regularization back propagation algorithms for multistep ahead wind speed forecasting using multilayer perceptron feed forward neural network. Master of Science with a Major in Energy Technology with Focus on Wind Power, Department of Earth Sciences, Uppsala University 13. Popescu I, Nafomita P, Constantinou P, Kanatas A, Moraitis N (2001) Neural networks applications for the prediction of propagation path loss in urban environments. In: 2001 IEEE VTS 53rd vehicular technology conference, spring proceedings, Rhodes, Greece, vol 1, pp 387–391 14. Ambroziak SJ, Katulski RJ (2014) Statistical tuning of Walfisch-Ikegami model for the untypical environment. In: 2014 8th European conference on antennas and propagation. IEEE, pp 2087–2091

Throughput and Spectrum Sensing Trade-Off by Incorporating Self-interference Suppression for Full Duplex Cognitive Radio Srilatha Madhunala1(&) and Hemalatha Rallapalli2 1

2

Department of Electronics and Communication Engineering, Vardhaman College of Engineering, Shamshabad, Hyderabad 501218, Telangana, India [email protected] Department of Electronics and Communication Engineering, Osmania University, Hyderabad, Telangana, India

Abstract. As the growth of wireless applications increasing very rapidly, the requirement of radio frequency spectrum also increases. But due to limited frequency spectrum, it has become overcrowded. In order to optimize spectrum utilization, cognitive radio has emerged, which allows secondary users to coexist in licensed users spectrum band and when found primary user in RF channel, secondary user required to evacuate that channel for a specific time interval and introduces significant latency for primary user. In traditional half duplex systems, secondary user first sense the channel and then transmission will happen. In this paper, full duplex communication with self interference suppression is proposed to optimize detection probability of primary user and secondary user throughput. In the presence of self-interference, waveform based spectrum sensing technique for WLAN channel usage is adopted to allow secondary user to detect primary user with high accuracy. Our objective is to decrease latency and improve throughput subject to varying self-interference of secondary user. Results of the proposed method implemented using Matlab shows reduced latency by a factor of 2.1 in comparison with full duplex slotted method and 14 in comparison with half duplex method by considering zero SIS and non-zero SIS. Keywords: Full duplex cognitive radio
Self-interference suppression Waveform based spectrum sensing
Latency




1 Introduction Usage of spectrum has been increasing exponentially in wireless technologies and become crowded in-turn cannot accommodate future services resulting to spectrum scarcity. However, licensed users are not utilizing their spectrum effectively thus resulting white spaces in space and time. To gain benefits from underutilized spectrum, cognitive radio has been emerged [1]. During the end of 2003, FCC proposed a rule stating that cognitive radio will be one which dynamically shares spectrum of primary users. IEEE then formed a group called 802.22 WRAN standard which utilizes TV band for providing wireless internet access without causing any disturbance as shown in Fig. 1 [2]. © Springer Nature Switzerland AG 2020 S. C. Satapathy et al. (Eds.): ICETE 2019, LAIS 4, pp. 130–138, 2020. https://doi.org/10.1007/978-3-030-24318-0_16

Throughput and Spectrum Sensing Trade-Off

131

Fig. 1. Topology of IEEE 802.22 WRAN [2]

Figure 2 depicts the frame structure of conventional CR which is presumed to be half duplex. Each frame includes sensing time slot as well as transmission time slot. Secondary user senses the primary user spectrum for s time slot and starts transmitting data for remaining T-s time slot. A trade-off exists in frame structure between spectrum sensing time slot and data transmission time slot, in turn throughput of cognitive radio. As per detection theory [3], if sensing time slot increases, probability of detection will be higher, which inturn improves unused spectrum utilization. But, if sensing time increases, data transmission time slot decreases and attainable throughput also gets decreased. Author in [4] addressed sensing-throughput trade-off by finding optimal sensing time slot which maximizes average attainable throughput under, single high target probability of detection constraint to protect primary user, but it has to suffer from either lost transmission capability or extra channel uses. The difficulty faced to achieve full duplex communication [12] is that the secondary user transmitter power is generally much larger than receiver power, so while receiving, signal which is transmitted by itself is considered as self-interference.

Fig. 2. Conventional half duplex cognitive radio frame structure

[4, 5] shows a method which periodically stops secondary user transmission while sensing the spectrum by introducing blind intervals, so that primary user transmissions can be detected. Using different sensing methods like energy detection, waveform based detection [9], power throughput trade-off [6, 10] are used to maximize secondary user throughput in presence of residual self-interference. Different authors [7, 8, 10, 11] proposed alternative approaches to enable simultaneous transmission and sensing, but faultless SIC is difficult to achieve and also it is above the noise floor. In this paper, we focus on improving latency and throughput by using full duplex scheme under self-interference suppression and primary user protection. Simulation results shows the improvement in latency and throughput.

132

S. Madhunala and H. Rallapalli

Outline: contribution of remaining paper is organized as follow: full duplex communication system model is described in Sect. 2. In Sect. 3, we derive probability of detection and false alarm using waveform based spectrum sensing technique and the formulation of secondary user decision process under performance analysis is done in Sect. 4. Finally, presenting simulation results and conclusion of the paper in Sects. 5 and 6 respectively.

2 System Model Conventional cognitive radio user will access the primary user spectrum channel without causing interference to primary user network. As primary user will not notify to secondary users about the start and end point of transmission and also do not wait while secondary users use the channel. So, secondary users should detect white spaces for transmission and backoff when primary user appears. For efficient interference cancellation and improving latency, throughput, full duplex cognitive radio is used [7]. Secondary user with self-interference capability will operate in full duplex communication either in slotted mode or sliding mode (simultaneous sensing and transmission) in comparison with half duplex which operates as either sensing mode or transmission mode as shown in Fig. 3. Let the secondary user senses a frequency band with carrier frequency fc, its bandwidth W and the received signal will be sampled at sampling frequency fs, which should be much greater than Nyquist rate [5].

Fig. 3. Frame structure of cognitive radio with half duplex and full duplex capability under selfinterference cancellation

Two scenarios will occur: first, if the licensed user is utilizing the spectrum, then the received signal at the secondary user is expressed as

Throughput and Spectrum Sensing Trade-Off

Y ðnÞ ¼ sðnÞ þ wðnÞ

133

ð1Þ

Equation (1) is called as hypothesis H1 and the second is when the primary user is not utilizing the spectrum, then received signal is expressed as Y ðnÞ ¼ wðnÞ

ð2Þ

Equation (2) is called as hypothesis H0. Where, s(n) is the signal which has to be detected, w(n) is the Additive White Gaussian Noise and n is the sample index. Detection algorithm performance will be analyzed with two probabilities: hypothesis H1 is a metric used for detecting the presence of licensed user activity and is defined by probability of detection, Pd and hypothesis H0 which is used for detecting the licensed user, when actually licensed user is not there and is defined by probability of false alarm Pf. Whenever the probability of detection is high, effective protection of licensed user is received. However, when the probability of false alarm is low, then the secondary user can use the frequency band more frequently when it is available. So, detecting probability should be as much as possible and probability of false alarm should be as minimum as possible in order to ensure better performance of a system.

3 Waveform Based Spectrum Sensing As energy based spectrum sensing cannot differentiate users signal and noise, selfinterference may lead to miss indicate licensed user activity, in spite of its simplicity. In this paper, waveform based sensing method with self-interference cancellation is adopted, which uses a known pattern in licensed user signal like pilot patterns or preamble and correlates the output signal with a known signal of itself [5, 8]. In terms of reliability and convergence time, waveform based method outperforms compared to energy based method. In this section, full duplex operation is analyzed by assuming self-interference cancellation factor v where v 2 ½0; 1 using waveform based method. If self-interference is completely suppressed (perfect SIS) then v = 0, otherwise suppresses a fraction 1 − v is imperfect SIS. Considering same model given in (1), the hypothesis test to check the status of a channel for full duplex is
Y ð nÞ ¼

vsðnÞ þ wðnÞ; H0 PðnÞ þ vsðnÞ þ wðnÞ; H1

 ð3Þ

The decision metric for waveform based spectrum sensing is " M ¼ Re

N X

# 

Y ðnÞs ðnÞ

ð4Þ

n¼1

Where * represents conjugate operation. When licensed user signal is absent, then the decision metric becomes

134

S. Madhunala and H. Rallapalli

" M ¼ Re

N X

# 

wðnÞs ðnÞ

ð5Þ

n¼1

and when licensed user signal is present, then the decision metric becomes M¼

N X n¼1

" 2

jsðnÞj þ Re

N X

# wðnÞs ðnÞ

ð6Þ

n¼1

Decision metric M is compared with a given threshold c to detect licensed user activity. If M > c, then licensed user is present, otherwise licensed user is absent and the channel can be used. The performance of sensing method will be measured by probability of detection and probability of false alarm which will be identified by the hypothesis test [4, 5] and is defined as Pf ¼ Pr ½M [ c=H0  ¼ 1  FM ðcÞ

ð7Þ

Pd ¼ Pr ½M\c=H1  ¼ 1  FM ðcÞ

ð8Þ

And the probability of miss-detection, Pm is defined as Pm ¼ Pr ½M\c=H1  ¼ 1  Pd

ð9Þ

In the perception of secondary user, two scenarios will exist: i. When secondary user is not transmitting, half duplex and full duplex (slotted and sliding) methods performance is comparable due to absence of self-interference. Simply secondary user senses the spectrum. ii. When secondary user is transmitting, half duplex method cannot sense the spectrum while transmitting, so it misses detecting primary user transmission (Pm = 1), whereas full duplex can simultaneously sense and transmit the data by cancelling its self-interference signal. As cognitive user is able to quickly detect licensed user activity using full duplex, latency will be reduced and secondary user throughput increases as compared to half duplex.

4 Performance Analysis Sensing latency and throughput trade off under self-interference suppression is discussed in this section. Using waveform based spectrum sensing method, with full duplex, there exist optimal sensing time to achieve best trade-off. Figure 3 depicts the cognitive radio frame structure with half duplex as well as full duplex capability under self-interference cancellation as proposed in [9] and [10]. Let sensing time duration is s, frame duration is T, C0 indicates capacity of secondary user by assuming primary user is not transmitting and C1 indicates capacity of secondary user while licensed user is transmitting.

Throughput and Spectrum Sensing Trade-Off

C0 ¼ log2 ð1 þ SNRSU Þ   SNRSU C1 ¼ log2 1 þ 1 þ SNRPU

135

ð10Þ ð11Þ

Where SNRSU indicates signal-to-noise ratio of secondary user which is measured at receiver node and SNRPU indicates signal-to-noise ratio of licensed user which is received at receiver node of secondary user. There are two possibilities where the secondary user can utilize licensed user’s frequency channel [4, 10]. Case 1: Total throughput of secondary user with full duplex capability when licensed user not transmitting and no false alarm raised by secondary user. RrD ¼ C0 ð1  Pr Þ

ð12Þ

Whereas, half duplex system will have extra loss in throughput due to the overhead of sensing RHD ¼

 T  ts  C0 1  pf T

ð13Þ

Case 2: Total throughput of secondary user with full duplex capability while licensed user transmitting which is not detected by secondary user. RFD ¼ C1 ð1  pd Þ

ð14Þ

Whereas, half duplex system will have extra loss in throughput due to the overhead of sensing RHD ¼

T  ts C1 ð1  pd Þ T

ð15Þ

It is clearly shown that for full duplex, due to self-interference, there is an increase in probability of false alarm. So, to maintain throughput close to C0, full duplex system with self-interference has to operate with lower probability of detection. Results indications that sliding window allows lower latency while maintaining high secondary user throughput in comparison with slotted full and half duplex.

5 Simulation Results The performance evaluation of full duplex scheme compared to half duplex scheme under SIS is shown in simulation results. Perfect self-interference suppression (no self-interference) is assumed to analyse the results. Figure 4 depicts the substantial improvement in sliding full duplex scheme latency in comparison with slotted full duplex and conventional half duplex schemes.

136

S. Madhunala and H. Rallapalli

For half duplex as the system is unable to sense and transmit simultaneously, latency is nearly half of secondary user frame length. As slotted full duplex scheme will take decisions for every ts time period, the latency is low with a minimum of ts/2. Sliding full duplex scheme will identify the primary user, and decisions can be taken very quickly as compared to other two schemes. Figure 5 depicts the quantiles of 97% and 99% for both sliding window full duplex and slotted window full duplex schemes for sensing latency and throughput.

Fig. 4. Latency-throughput for three schemes under perfect self-interference suppression

Increasing self-interference suppression which is measured relative to noise floor. Figure 6 illustrate the improvement in latency for sliding and slotted full duplex schemes for varying self-interference. For all values of SIS, lower latency can be achieved for sliding window full duplex and also it is more difficult to self-interference by factor of 2.1 improvement as compared to slotted scheme. Figure 7 illustrate the improvement in throughput for sliding and slotted full duplex schemes for varying selfinterference. Throughput is calculated for the same primary user protection for both the full duplex schemes and for all values of self-interference, higher throughput is achieved for sliding window full duplex.

Fig. 5. Latency-throughput for 97% and 99% quantile of full duplex schemes

Throughput and Spectrum Sensing Trade-Off

137

Fig. 6. Latency of full duplex schemes under imperfect self-interference suppression

Fig. 7. Throughput of full duplex schemes under imperfect self-interference suppression

6 Conclusion We projected full duplex cognitive radio system which increases spectrum utilization, secondary user throughput and spectrum latency by performing sensing the spectrum and transmission of data at the same time. Particularly, we study the framework under the condition where primary users will be protected under perfect SIS and imperfect SIS by using waveform based method. In order to provide tradeoff between throughput and latency, we proposed full duplex sliding window scheme which will takes decisions at every sample. Furthermore, simulation results shows that proposed sensing framework can achieve high throughput and low latency over conventional half duplex scheme under self-interference suppression.

References 1. FCC (2002) Spectrum policy task force report. FCC 02155 2. IEEE standard for information Technology-Local and metropolitan area networks-specific requirements-part 22: cognitive wireless RAN medium access control (MAC) and physical layer (PHY) specifications: policies and procedures for operation in TV Banks, in IEEE standard 802.22-2011, pp 1–680 (2011)

138

S. Madhunala and H. Rallapalli

3. Poor HV (1998) An introduction to signal detection and estimation, 2nd edn. Springer, Berlin 4. Liang YC, Zeng Y, Peh ECY, Hoang AT (2008) Sensing-throughput tradeoff for cognitive radio networks. IEEE Trans Wirel Commun 7(4):1326–1337 5. Yucek T, Arslan H (2009) A survey of spectrum sensing algorithms for cognitive radio applications. IEEE Commun Surv Tutor 11(1):116–130 6. Liao Y, Song L, Han Z, Li Y (2015) Full duplex cognitive radio: a new design paradigm for enhancing spectrum usage. IEEE Commun Mag 53(5):138–145 7. Riihonen T, Wichman R (2014) Energy detection in full duplex cognitive radios under residual self-interference. In: Proceedings of cognitive radio oriented wireless networks and communications, June 2014, pp 57–60 8. Afifi W, Krunz M (2014) Adaptive transmission-reception-sensing strategy for cognitive radios with full duplex capabilities. In: IEEE international symposium on dynamic spectrum access networks, April 2014, pp 149–160 9. Afifi W, Krunz M (2015) Incorporating self-interference suppression for full duplex operation in opportunistic spectrum access systems. IEEE Trans Wirel Commun 14 (4):2180–2191 10. Liao Y, Wang T, Song L, Han Z (2014) Listen and talk: full duplex cognitive radio networks. In: 2014 IEEE global communications conference, December 2014, pp 3068– 3073 11. Ahmed E, Eltawil A, Sabharwal A (2012) Simultaneous transmit and sense for cognitive radios using full duplex: a first study. In: Proceedings of IEEE international symposium antennas propagation, July 2012, pp 1–2 12. Duarte M, Sabharwal A (2010) Full duplex wireless communications using off-the-shelf radios: feasibility and first results. In: Proceedings of the ASILOMAR 2010 conference, November 2010, pp 1558–1562

IOT Based Monitor and Control of Office Area Using ZYBO Priyanka Paka(&) and Rajani Akula JNTUH College of Engineering Hyderabad, Hyderabad, India [email protected], [email protected]

Abstract. Now a days, everyone is worried about their safety in hazardous environments. In applications ranging from household to industrial environments, monitoring and control has become important. Using the Internet as the common interface, many devices can be controlled. IOT (Internet of Things) monitors and controls the devices using the World Wide Web. In this paper temperature and light intensity are monitored and the devices Fan and light are controlled remotely from a safe distance by using web technology. The monitoring and controlling system uses ARM cortex A9 processor in ZYBO (Zynq Board) as core of the hardware. The Web server application and Linux, both are ported on ARM cortex A9 processor. The GUI (Graphical User Interface) application is designed by using Qt for displaying temperature and light intensity values and for controlling the devices. The sensor control unit sends the sensor data from microcontroller through wireless medium with the help of Zigbee technology to ARM cortex A9 processor. The data from web server is received through HTTP (Hyper Text Transfer Protocol) and displayed on web page. The temperature and light intensity values are displayed on the monitor and also on the web page. The devices fan and light are controlled remotely from a web page. Keywords: Sensors


ZYBO
Qt creator
Web server

1 Introduction In everyday life, monitoring and control of systems from remote locations is becoming necessary now a days. It makes easy to control and monitor the condition of devices from any location at any time. Previously embedded systems uses microcontrollers or microprocessor with low-cost 32-bit processors with limited capabilities, now a days they are using modern 32-bit system on a chip (SOC), which contain a reduced instruction set (RISC) processor with volatile memory, non-volatile flash memory and a wide range of standard I/O devices on a single chip [1]. The IOT (Internet of things) refers to the wireless connectivity between objects. It can be defined as the connection of different types of objects such as smartphones, personal computers to the Internet. The Internet provides an immediate solution to connect devices and is used to access remote information and remotely control the devices [2]. The spread of the computer environment undoubtedly changes our way of life. Our working environments have improved. Advanced home automation, for example, can create a © Springer Nature Switzerland AG 2020 S. C. Satapathy et al. (Eds.): ICETE 2019, LAIS 4, pp. 139–146, 2020. https://doi.org/10.1007/978-3-030-24318-0_17

140

P. Paka and R. Akula

comfortable home atmosphere. Both those seeking a luxurious lifestyle and people with special needs can benefit from home automation, as wireless control can help them carry out their daily activities easily and accurately [3]. Single core controller with a limited datarate is previously used. ZYBO is a system on chip which has a dual core ARM cortex A9 processor. Data is acquired in parallel because it has a high processing speed. Compared to ARM9 and ARM11, the performance is fast [4]. The Monitoring and control system consists of Sensor control unit and Monitoring Unit. The sensor control unit consists of a microcontroller used to interface sensors, relays are used to control the devices, while Zigbee transfers sensor data using wireless communication. The monitoring unit consists of ZYBO, and Zigbee. ARM Cortex A9 processor in ZYBO is used as major controller. Web server Application and Linux is ported on Arm cortex A9 processor. The temperature and light intensity are measured using sensors. Sensors are interfaced with Analog to digital converter (ADC). The data is transmitted from microcontroller to Arm cortex A9 processor in ZYBO with the help of Zigbee Technology to Monitoring unit. The industrial parameters temperature and light intensity are monitored continuously. The user can request a particular web page from a web server. By entering unique IP address the data is received from web server through HTTP protocol and displayed on web page.

2 System Overview The monitoring and control system proposed in this paper is composed of two units, the sensor control unit and the monitoring unit. Sensor Control Unit is shown in the Fig. 1. LDR (Light Dependent Resistor) and Temperature sensors are interfaced with 8 channel ADC. ADC is connected to the 8051 microcontroller. Two relays are connected to microcontroller to control the devices fan and light. The data from Temperature sensor and LDR sensor is an analog voltage. Sensors are interfaced with ADC0808. The digital data is 8 bit data from ADC is given as an input to the input port of the microcontroller. Microcontroller stores the sensor data in serial buffers in digital form. Zigbee is connected to the microcontroller for transmission and reception of digital data. Zigbee transmits the sensor data to the monitoring unit via a wireless medium, both temperature and light intensity.

Fig. 1. Sensor control unit block diagram

IOT Based Monitor and Control of Office Area Using ZYBO

141

Monitoring Unit is shown in Fig. 2. Arm Cortex A9 processor in ZYBO is used as major controller. The communication between the sensor control unit and the monitoring unit is carried out by the Zigbee module, i.e. the sensor control unit reads the temperature and light intensity values and sends them to the monitoring unit using Zigbee. The temperature and light intensity values of the sensor can be seen on the monitor using the GUI and on webpage using standard web browser. Based on the monitored values decision is taken by using threshold value for temperature is 35° centigrade and for light intensity threshold is 100 lm. The devices Fan and light are controlled from web page.

Fig. 2. Monitoring unit block diagram

2.1

Hardware

The heart of the sensor control unit is 8051 microcontroller. 8051 Microcontroller is an 8-bit Microcontroller with 40 pins DIP (dual in line package), 4 kb storage of ROM and 128 bytes storage of RAM. It consists of four parallel 8-bit ports are programmable as well as addressable as per the requirement. It consists of On-chip crystal oscillator having crystal frequency of 12 MHz. It has two buses one for program and other for data of 8 bit. It also has 8 bit accumulator and 8 bit processing unit. The sensors used for monitoring the temperature and light intensity are LM35 Temperature sensor and LDR sensor respectively. Temperature sensor LM35 is used. The LM35 series are integrated-circuit precision temperature sensors, whose output voltage is linearly proportional to the celsius (Centigrade temperature). The sensor has a sensitivity of 10 mV/C. To detect the presence or absence of light, the LDR sensor is used. Light intensity is measured in Lumen. A typical LDR has a resistance in the darkness of 1 MΩ, and in the brightness a resistance of a couple of Kohm. To transmit the data through wireless medium Zigbee is used. A Zigbee wireless communication protocol based on the IEEE 802.15.4 protocol. Due to low power consumption, its transmission distance varies from 10 to 100 m, depending on the output and environmental characteristics. It transmits data over long distances by passing information through a mesh network of intermediate devices to reach more distant ones. Zigbee has data rate 250 kbit/s. ZYBO is used as main controller in Monitoring Unit. ZYBO is a Xilinx All Programmable SOC architecture (APSOC), easy-to-use embedded software and digital circuit development platform built around the Xilinx Zynq-7000 family of smaller members. Xilinx APSOC architecture integrates a dual-core ARM Cortex-A9

142

P. Paka and R. Akula

processor with Xilinx7-series Field Programmable Gate Array logic. It has Ethernet port. It has a number of peripherals for multimedia and connectivity. It is used from a complete OS to design systems of any complexity. It contains a power supply regulated by 5 V/2.5 A and a micro SD card of 16 GB. Jumper is necessary for all sources for each supply. Four slide switches and push buttons, four LEDs directly connected to the Zynq, two push buttons and one led via Multiplexed pins are connected via serial resistors to prevent short circuit damage. 2.2

Software

(1) Linux Operating System: The Linux operating system is loaded into the ARM cortex processor using the files Boot loader, Linux Kernel and Root file system. Boot loader is a small program that loads the kernel of the operating system into the On chip ROM memory and transfers it to the kernel. The main role is to initialize hardware equipment (including I/O, a special function register), to create a space map for memory and to bring the hardware and software environment of the system to an appropriate state. The kernel of the Linux operating system supports the ARM cortex A9 processor well and manages the components connected to the periphery of the controller. (2) Design of GUI: The GUI design for embedded systems differs from conventional data computer class software, which often handles mouse or keyboard events and other external devices. The Monitoring and Control System uses Qt/Embedded as its GUI application development platform under embedded Linux, so that it can meet the limitations of embedded system resources. Qt uses C++ as its programming language, programming can be implemented with Linux C. The header files are the library of Qt-API and the Linux system call library. Device drivers are required to read and write a specific device file and to write operational interface functions and modify the sensor drivers. To achieve the graphical display of data collected by sensors, the monitoring and control system uses Qt to complete the GUI. By adjusting the size and position on that window, we developed a relevant GUI window is shown in Fig. 3 using Qt designer. The design of the GUI contains buttons and labels. The values of temperature and light intensity are displayed using labels. The devices are controlled by radio buttons (ON/OFF).

Fig. 3. GUI window

IOT Based Monitor and Control of Office Area Using ZYBO

143

3 Workflow of the Proposed System The monitoring and control system using Linux as operating system on ARM cortex A9 processor. Temperature and light intensity are measured by using Temperature and LDR sensors. Sensor gives analog voltage. So, sensor data is converted into digital data. with the help of QT designer GUI is created with labels temperature and light for displaying the temperature and light intensity values on monitor and for controlling the devices Fan and Light radio buttons are created. Web page is created by using hypert ext markup language (HTML) language for monitoring of industrial parameters and controlling the devices. The workflow of proposed monitoring and control system is as shown in the Fig. 4.

Fig. 4. Flowchart of monitoring and controlling

144

P. Paka and R. Akula

4 Results The sensor control unit containing 8051 microcontroller, Zigbee, Sensors and cooling Fan is as shown in the Fig. 5 and Monitoring unit with Zigbee is as shown in the Fig. 6. The Linux operating system is ported on ARM cortex processor in ZYBO board. GUI application is created by using Qt creator. The temperature and LDR sensors readings are sent from sensor control unit to monitoring unit by using File transfer protocol with the help of zigbee. The sensor readings temperature and light values are displayed on monitor as shown in Fig. 7 and also on web page as shown in Fig. 8. Web page receives information from web server through HTTP protocol. The status of sensor readings are monitored from web page and user can perform any corresponding control action if needed. If threshold value for temperature is 35° centigrade and for light intensity threshold is 100 lm is crossed then the devices Fan and Light gets turned ON/OFF. If the threshold of the temperature is increased then the fan is turned on. If the light intensity value is below the threshold then the light is turned on. The devices Fan and light can be controlled from web page.

Fig. 5. Sensor control unit

Fig. 6. Monitoring unit

IOT Based Monitor and Control of Office Area Using ZYBO

Fig. 7. Display on monitor

Fig. 8. Display on web page

145

146

P. Paka and R. Akula

5 Conclusion This paper introduces a wireless solution for the easy management and monitoring of office units based on the Internet protocol. Temperature and light intensity values are monitored remotely from remote location through internet by entering unique IP address in a web browser. The devices fan and light in office are controlled from web page. The temperature and light intensity are checked for crossing the threshold values and controlled immediately by controlling the sensors remotely.

References 1. Kullarkar P, Dhote K (2014) Embedded web server based open laboratory monitoring and controlling system using arm 11. Int J Adv Res Comput Sci Softw Eng 4(5) 2. Pavithra D, Balakrishnan R (2015) IOT based monitoring and control system for home automation. In: 2015 global conference on communication technologies (GCCT). IEEE, pp 169–173 3. Manssor SAF, Osman AA, Awadalkareem SD (2015) Controlling home devices for handicapped people via voice command techniques. In: International conference on computing, control, networking, electronics and embedded systems engineering 4. Keerthana K, Prashanthi R (2016) Flexible and reconfigurable SOC for sensor network under ZYNQ processor. Int J Sci Technol Eng (IJSTE) 3(01) 5. Vijayalakshimi K, Maha Lakshmi DG (2016) Real time weather monitoring from remote location using Raspberry Pi. Int J Adv Res Electr Electron Instrum Eng (IJAREEIE) 5(12) 6. Joaquinito R, Sarmento H (2016) A wireless biosignal measurement system using a SoC FPGA and bluetooth low energy. In: 6th international conference on in consumer electronics-Berlin (ICCEBerlin). IEEE, pp 36–40 7. Ferrari P, Marioli D, Sisinni E, Taroni A (2005) An internet based interactive embedded data acquisition system for real time application. IEEE Trans Instrum Meas 54 (6) 8. Manivannan M, Kumaresan N (2011) Design of online interactive data acquisition and control system for Embedded real time application. In: Proceedings of IEEE international conference on emerging trends in electrical and computer technology, pp 551–556 9. Guan M, Gu M (2012) Design and implementation of an embedded web server based on ARM. In: IEEE international conference on computer science education, pp 479–482 10. Tarange PH, Mevekari RG, Shinde PA (2015) Web based automatic irrigation system using wireless sensor network and embedded Linux board. In: International conference on power and computing technologies (ICCPCT). IEEE, pp 1–5

Robust Speaker Recognition Systems with Adaptive Filter Algorithms in Real Time Under Noisy Conditions Hema Kumar Pentapati(&) and Madhu Tenneti Department of ECE, Swarnandhra Institute of Engineering and Technology, Narasapur, West Godavari District, Andhra Pradesh, India [email protected]

Abstract. At present, Speaker recognition systems are being widely used in speaker detection, authentication and authorization to perform secure transactions in various personal, commercial and industrial applications. As speaker recognition in noisy environments is becoming increasingly difficult, a novel system is developed using MFCC and Vector Quantization. The training data is collected from 15 native speakers. LMS, NLMS and RLS adaptive filters are used to reduce the noise in the speech signal. The performance of all the speaker recognition systems is rated by calculating the Equal Error Rate (ERR) and Euclidian distance. Keywords: Mel Frequency Cepstral Coefficients (MFCC)
Least mean square
Normalised LMS Recursive Least Squares Euclidian distance
Equal Error Rate
Vector Quantization




1 Introduction The task of speaker recognition is to recognizing an input speaker among the reference models of different speakers by comparing the feature vectors. To identify the person, we have the process of matching the voice of the unknown speaker with the reference models in the trained database, while in verification it performs a binary decision which consists of determining whether the person speaking is the same person he/she claims to be. In simple words we are calling it as verifying their identity [1]. Generally, the features of the input utterance get changed under noisy environment. So, it is very difficult to construct the speaker recognition system in noisy conditions. Due to noise, the undesired features of the speaker results, this makes the performance of the systems gets changed or degraded especially when the input speech signal is applied through the microphone. The main objective of this paper is to develop the different improved speaker recognition systems such that improved recognition from one system to another. When the input utterance is passed through the specific adaptive filter that tends to suppress the noise and giving the signal with reduced noise. This process of noise cancellation can be done by three filters called LMS Adaptive filter, NLMS Adaptive filter and RLS Adaptive filters, when input is through microphone [2]. So, from the system with LMS © Springer Nature Switzerland AG 2020 S. C. Satapathy et al. (Eds.): ICETE 2019, LAIS 4, pp. 147–154, 2020. https://doi.org/10.1007/978-3-030-24318-0_18

148

H. K. Pentapati and M. Tenneti

adaptive filter to the system with NLMS adaptive filter, the performance of the system can be improved due to reduction of noise from speech signal. Finally, the system with RLS adaptive filter gives the predominant increase in performance and improved recognition when compared with the systems with LMS and NLMS adaptive filters.

2 Proposed Method For speaker recognition system, we have two phases: feature extraction phase and feature matching phase. Feature extraction deals with extraction of important characteristics from speech signal, these characteristics are unique to each person. So, these features of input speech signal or unknown speaker are compared with the speakers in trained database for speaker identification, so this process is feature matching [3, 4]. In this proposed method, MFCC is used for feature extraction and Vector Quantization technique is used for feature matching but feature extraction produces the unique and exact features of a speaker only in the controlled environment. But in the presence of external or surrounding noise, the MFCC yields the undesired features of that particular speaker. To overcome this, the speech signal is passed through preprocessing stage that is through one particular adaptive filter before applying the utterance to next phase called feature extraction. So, Speaker recognition system is implemented using the following stages (i) Pre-processing stage (ii) Feature extraction (iii) Speaker modelling 2.1

Pre-processing Stage

Three different speaker recognition systems are developed with pre-processing stage as LMS, NLMS and RLS adaptive filters for three systems respectively. Adaptive Filter The main aim of an adaptive filter in noise cancellation is to increase the signal to noise ratio by separating the noise from a signal. It has Adaptation algorithm for adjusting parameters for improved performance that monitors the surroundings such that varies the filter transfer function. The process of noise cancellation which requires the help of algorithms such that characteristics to converge very rapidly. LMS and NLMS are adaptive which are used for signal enhancement in such a way that noise can be reduced. The drawback of the LMS algorithm is selection of step size parameter. So, that it is very difficult to choose step size. This problem can be solved by NLMS adaptive filter, here the step size parameter is normalized [6]. Normalized least mean square (NLMS) is next to the LMS algorithm. NLMS adaptive algorithm is a time varying step size algorithm. Very fast convergence rate and its efficient RLS algorithm in noise cancellation is key factors and key advantages of RLS adaptive filter. The recursive Least Squares algorithm is more complex than compared to LMS and NLMS adaptive filters because it requires high computational

Robust Speaker Recognition Systems with Adaptive Filter Algorithms

149

requirement than LMS and NLMS filters, but behaves much better in terms steady state MSE compared to remaining filters [7]. The Recursive least squares (RLS) adaptive filter is an algorithm which recursively finds the filter coefficients that minimize a weighted linear least squares cost function relating to the input signals and it is shown in below figure. Because speech is a nonstationary signal, for such type of signals the RLS adaptive gives the better performance but some stability problems. Compared to other adaptive filters, the MSE is minimized, so that it gives the better performance than that of other adaptive filter algorithms [7] (Fig. 1).

Fig. 1. Standard adaptive filter

2.2

Feature Extraction

Most of the Automatic speaker recognition systems uses the method called Mel Frequency Cepstral Coefficients (MFCCs) for feature extraction, which is more effective and robust under noisy conditions. These are coefficients that represent the individual speaker. In MFCC the frequency bands are positioned logarithmically (on the Mel scale) which approximates the human auditory system’s response more closely than the linearly spaced frequency bands. To achieve this, the triangular filters are placed linearly upto 1000 Hz and placed logarithmically above 1000 Hz [3]. 2.3

Feature Matching

Pattern recognition is used to classify objects of interest into certain number of classes. The objects of interest are called patterns and here feature vectors that are extracted from the person’s speech utterance using different available techniques. The classification procedure is applied on extracted features, it can be also referred to as feature matching. The feature matching techniques used in speaker recognition includes Dynamic Time Warping (DTW), Hidden Markov Modelling (HMM), and Vector Quantization (VQ). In order to have simpler implementation and accuracy, the VQ approach is used in this system for feature matching, due. VQ is a process of mapping vectors from a large vector space to a finite number of regions in that space. Each region is called a cluster and can be represented by its center called a codeword. The collection of all codewords is called a codebook [4].

150

H. K. Pentapati and M. Tenneti

3 Methodology Each System is developed by selecting the LMS, NLMS and RLS adaptive filters as pre-processing stage. In the first system, this pre-processing stage is selected as LMS Adaptive filter, which reduces the noise from input speech signal and then given to feature extraction phase [5]. In the second system, the pre-processing stage which is LMS adaptive filter is replaced by NLMS adaptive filter then applied to feature extraction phase. Now the features of input speech signal is compared with the features of the trained database and each time the Euclidian distance is computed, so speaker with smallest VQ distortion is selected as desired speaker corresponding to input speech. So, the performance of the system is measured by means of Equal Error Rate (EER) and Euclidian distance and this is compared with the previous system. Finally, the pre-processing stage is selected as Recursive Least Square (RLS) adaptive filter and much more cleaned signal that is negligible noise in speech signal is passed through next stage. The Euclidian distance is computed and this has smallest VQ distortion when compare to above mentioned systems. As explained above, the same proposed method is shown in below Fig. 2.

INPUT SPEECH SIGNAL

PRE PROCESSING STAGE

FEATURE EXTRACTIO N PHASE

DECISION ACCEPT/REJECT

VECTOR QUANTIZATION

IDENTIFICATI ON RESULT

Fig. 2. Block diagram of all three different speaker recognition systems

Note: PREPROCESSING STAGE: FOR FIRST SYSTEM, THE PREPROCESSING STAGE IS LMS ADAPTIVE FILTER FOR SECOND SYSTEM, THE PREPROCESSING STAGE IS NLMS ADAPTIVE FILTER FOR THIRD SYSTEM, THE PREPROCESSING STAGE IS RLS ADAPTIVE FILTER

Robust Speaker Recognition Systems with Adaptive Filter Algorithms

151

4 Experiments On all the 3 different speaker recognition systems, the experiments were conducted using single word called “hello” with 20 speakers in trained database in presence of noisy environment. Firstly, the adaptive filter filters the signal such that the cleaned signal is passed through the feature extraction phase that gives Mel frequency Cepstral Coefficients (MFCC) and then the code book generated for each VQ model and thus finally Euclidian distance was computed for each speaker with corresponding VQ code book. The same procedure was repeated for all the 3 speaker recognition systems, which were implemented using each adaptive filter. So, the performance of each system was evaluated in terms of Euclidian distance and Equal error rate (ERR), False acceptance rate (FAR) and False Reject Rate (FRR) [2]. Finally, from one system to another, we observed the improvement in recognition. The System with RLS adaptive filter gives the best results compared to other systems with LMS and NLMS Adaptive filter (Fig. 3).

Fig. 3. Speech signal with RLS adaptive filter

5 Results and Discussions The Euclidian distance gives the similarity between the given speaker and the template, that is, the input utterance is compared with every known template gives rise to Euclidian distance. The system performance was measured in terms ERR, FAR and FRR. If it has lower EER, we can conclude that the performance of the system is high. In an uncontrolled environment, if we are not using any filter, the Equal Error Rate of 68% was calculated in the system. The Equal Error Rate of 45% was achieved by the system using LMS Adaptive filter. Also observed the improved results in the system with NLMS Adaptive filter and further, the performance increases for the system with RLS Adaptive filter. If the speaker repeats the same utterance for 2times or 2repetitions. The below table is for the utterance by the speaker for 1st time (Table 1).

152

H. K. Pentapati and M. Tenneti

Table 1. Output in noisy environment (speaker-1st time) in terms of Euclidian distance for different systems. Speaker recognition system System without filter System with LMS adaptive filter System with NLMS adaptive filter System with RLS adaptive filter

Euclidian distance Result 1.21 Speaker does not found 0.93 Speaker found 0.82 Speaker found 0.77 Speaker found

Also the same procedure was applied for the same speaker but for 2nd time. The below table is for the utterance by the speaker for 2nd time (Table 2). Table 2. Output in noisy environment (speaker-2nd time) in terms of Euclidian distance for different systems. Speaker recognition system System without filter System with LMS adaptive filter System with NLMS adaptive filter System with RLS adaptive filter

Euclidian distance Result 1.15 Speaker does not found 0.96 Speaker found 0.84 Speaker found 0.75 Speaker found

The improvement in Euclidian distance is clearly observed from the above systems and such that the system with RLS adaptive filter gives better Euclidian distance. The improvement in equal error rate (ERR) was observed in the system with RLS adaptive filter compared to the systems with LMS adaptive filter and NLMS adaptive filter. The curves of FAR and FRR, when they are overlapped, they intersect at one particular point, the value at that point is called Equal Error Rate (EER). The below table shows the comparison of performance of different systems in terms of Equal Error Rate (ERR), FRR and FAR (Table 3). Table 3. Output in noisy environment using different systems with adaptive filters Speaker recognition system System without filter System with LMS adaptive filter System with NLMS adaptive filter System with RLS adaptive filter

FRR 40.32 30.41 24.15 13.72

FAR 54.19 38.43 31.27 28.16

EER 64 34 29 24

Generally the performance of the system in real time with adverse conditions may not be same every time, but the system with RLS adaptive filter gives almost same results in different environmental conditions. In front end, extraction of the desired features even in noisy environment is implemented with RLS adaptive filter and with no change in speaker modeling end, after that it is further extended to different adaptive

Robust Speaker Recognition Systems with Adaptive Filter Algorithms

153

filters and tested the overall performance of the system. The performance measure with Euclidian distance as a metric and the Equal Error rates of different systems which are developed using various filters is shown below Figs. 4 and 5.

Equal Error Rate

Fig. 4. Improvement in the performance of system with different filters

80

System without filter

60 40

System with LMS filter2

20 0 Clean Environment

Noisy Environment

System with NLMS filter System with RLS filter

Environment

Fig. 5. Variation in the performance of different systems in two environments

6 Conclusion This paper gives development of the three speaker recognition systems using adaptive filters and also the performance comparison of all three speaker recognition systems in noisy conditions. The system with Recursive Least Squares (RLS) adaptive filter gives

154

H. K. Pentapati and M. Tenneti

the better performance compared to other systems, it achieves the Equal Error Rate (ERR) of 20% which is better than other systems. It gives the accurate results in terms of Euclidian distance, which was 0.74 for this system. Further, this work can be extended by improving the feature extraction phase to extract the desired features and by selecting the desired features from the extracted features.

References 1. Nijhawan G, Soni MK (2014) Speaker recognition using MFCC and vector quantisation. Int. J. Recent Trends Eng Technol 11(1):211 2. Hemakumar P, Srinivas V, Madhu T (2016) Improved dynamic speaker recognition system using NLMS adaptive filter. Int J Comput Appl (0975–8887) 148(10):39–43 3. Kekre HB, Bharadi VA, Sawant AR (2012) Speaker recognition using vector quantization by MFCC and KMCG clustering algorithm. In: 2012 international conference on communication, information & computing technology (ICCICT), 19–20 October 2012 4. Srinivasa Kumar Ch, Mallikarjuna Rao P (2011) Design of an automatic speaker recognition system using MFCC, vector quantization and LBG algorithm. Int J Comput Sci Eng (IJCSE) 3(8). ISSN 0975-3397 5. Abdul Samad S, Hussin A, Anuar Ishak K (2015) Improving hybrid speaker verification in noisy environment using least mean square adaptive filters. In: The 5th International Conference on Information and Communication Technology for The Muslim World (ICT4M), 17–18 November 2014. ISBN 978-1-4799-6242-6 6. Poddar A, Sahidullah Md, Saha G (2015) Performance comparison of speaker recognition systems in presence of duration variability. 978–1–4673–6540–6/15/$31.00 @ 2015 IEEE 7. Dhiman J, Ahmad S, Gulia K (2013) Comparison between adaptive filter algorithms (LMS, NLMS and RLS). Int J Sci Eng Technol Res (IJSETR) 2(5):1100–1103

Cognitive Radio: A Conceptual Future Radio and Spectrum Sensing Techniques - A Survey N. Rajanish1(&) and R. M. Banakar2(&) 1

D.S.C.E, Bangalore, India [email protected] 2 B.V.B.C.E.T, Hubli, India [email protected]

Abstract. Recent advancements in communication field have resulted in several multimedia applications that require higher data rates. With the available frequency space being limited, the competition for bandwidth allocation is increasingly getting tougher. It may not be possible to accommodate the needs of a device with higher rates by the use of static frequency allocation. Henceforth the above limitation creates a compelling need for us to come up with a new scheme which allows efficient exploitation of frequency spectrum (bandwidth). The Cognitive Radio with spectrum sensing capability offers a promising solution for the same. In this paper, an attempt has been made to explore cognitive radio concepts and various spectrum sensing techniques suitable for CR. Keywords: Cognitive radio
Spectrum sensing
Threshold
Energy detector
Compressive sensing

1 Introduction The revolutionary development in the digital wireless technology has surged the need for data communication with higher speed and low power. It has become the utmost priority of current industry. The need for higher resolution, irrespective of image or video, has led to increased utilization of the spectrum resulting in scarcity of available spectrum. A reconfigurable radio terminal, that is intelligent and adaptable, is the need of the hour for the next generation. By providing proper radio management services, the terminals can be made more adaptable and efficient enough to use spectrum resources across complex heterogeneous networks as well as spectrum constrained wireless networks. With the increased use of spectrum, the demand for wireless communication services is fulminating, exerting tremendous strain on the capacity of wireless channels. One of the parameters to be considered for offloading the strain on wireless communication network is managing radio frequency spectrum. While most of the RF spectrum is allocated, some of them either are over utilized or underutilized. The unused spectrum is called Spectrum holes. “Spectrum Hole” is a term defined in Cognitive radio in context with the spectral band. Spectrum hole or white space is the spectrum which is under used or not used defined in a region of space, time and © Springer Nature Switzerland AG 2020 S. C. Satapathy et al. (Eds.): ICETE 2019, LAIS 4, pp. 155–165, 2020. https://doi.org/10.1007/978-3-030-24318-0_19

156

N. Rajanish and R. M. Banakar

frequency. In terms of time it is explained as, within a geographical area of consideration, at a given time instant, it is a band of frequency unused or under used. Whereas in terms of frequency, it is the band of frequency which is not being used, can be used by the secondary user to operate without interfering the primary user’s operation. A brief review of the spectrum hole was presented by Tandra in his research paper in [1]. In the current scenario, the technique used to allocate spectrum is inefficient, leading to considerable amount of under-utilization of spectrum with the demand into consideration. RF frequency in the range of 30 MHz to 3 GHz is the spectrum in demand. A company named Shared Spectrum Company (SSC) near Washington DC conducted a survey on spectrum usage over a period of three and half days in the fall of 2009 across several places in and around Washington DC and it found that spectral or band occupancy of RF signal in cellular radio band was hardly 40%, which meant, the total bandwidth of nearly 806 to 902 MHz was dedicated and had 60% of bandwidth in idle state. This is the state of spectrum usage in almost all the countries, which is evident from various research data base. In order to make the efficient usage of the spectra, Cognitive Radio concept comes into existence [2]. The allocation of radio spectrum, RF spectrum usage and management is done through an international treaty among the nations and different organizations [3–7]. The work is organized as follows; Sect. 2 defines Cognitive Radio and gives a brief introduction to its parameters. Section 3 discusses Spectrum Sensing techniques, Sect. 4 briefs about the related work carried out by various researchers and finally in Sect. 5 Conclusions are presented.

2 Cognitive Radio “Cognitive Radio is an intelligent radio which keeps track of the RF parameters in its surroundings viz. modulation techniques, transmitting power, carrier frequency and adapts itself to provide a reliable communication and tries to make an efficient utilization of the available spectrum” [5, 7, 8]. In Cognitive radio technology, the users are classified as primary user (PU) and secondary user (SU). Primary users hold the highest priority and legal rights to operate on a band of spectrum allotted to them while the secondary users are at the least priority and operate over the unused spectrum [9]. The Cognitive radio needs to have certain functionalities like: • Flexibility & Agility: The ability to adapt to different modulation techniques and other radio parameters while under operation. CRs should also be able to reconfigure itself on wide band technology. • Spectrum sensing: Secondary users vying for spectrum should be able to detect the occupancy of spectrum by primary users. • Adaptability: The ability to adapt to the surrounding environment’s RF parameters in accordance to the feedback.

Cognitive Radio: A Conceptual Future Radio and Spectrum Sensing Techniques

157

The challenges that are encountered in implementing Cognitive Radio are, first, to have a universal receiver which would adapt to any band of frequencies received at SU, second, to differentiate between the noise and the signal at low energy level, third, allocation of radio resources. Apart from these, an effective sensing device to identify the spectrum hole, to reconfigure the network and agree upon various network protocols to setup physical link and functions of network layers alongside satisfying the constraints of a specific selected band [10].

3 Spectrum Sensing Realization of Cognitive Radio concept and its implementation worldwide will lead to efficient usage of spectra. Spectrum sensing is the most challenging and key computational modules in realizing Cognitive Radio. The detection of signals in presence of varying noise, channel impairments and complexity of observed spectrum from multiple devices is always less optimum. Various techniques and algorithms are being developed, tested and studied under various environmental conditions to estimate their behavior for effective and efficient implementation. A review of the spectrum sensing techniques and various tasks associated with it is given in the Table 1 [6] below. Techniques under consideration for spectrum sensing include [12–14]: (1) Energy Detectors: Received signal properties are unknown at the receiver (Blind Signals). (2) Matched filter: Secondary features of the transmitted signal are to be known prior at the receiver, usually employed in 802.11x protocol using device, viz. Bluetooth (Non Blind Signals). (3) Cyclostationary detectors: employ extraction of signal features for detection. (4) Cooperative Sensing: Signal from a single source received at multiple receivers spatially separated are combined and compared to know the existence of the transmitted signal. (5) Non Cooperative Sensing: Signal from a single source is time delayed and compared and verified for the existence. These methods are studied in order to ascertain their performance (static and dynamic), credibility, reliability, implementability and complexity with regard to the real world prototyping on various cognitive radio platforms [11, 15, 16]. Spectrum sensing forms the critical processing technique in cognitive radios. There are various parameters which have to be set in with extreme efficacy. Two of them are setting of threshold value and choosing the number of samples for an effective sensing operation. 3.1

Threshold Setting - k

The transmitted signal’s energy undergoes maximum variations due to interference signal viz., Multipath signals, Doppler Effect and Inter Symbol Interference added with noise.

158

N. Rajanish and R. M. Banakar Table 1. Spectrum sensing algorithms

Spectrum sensing techniques Approaches Employable algorithms

Internal sensing External sensing Beacon sensing Geolocation +Database

Challenges

Matched filter Energy detector Cyclostationary spectrum sensing Compressed sensing Radio identification based sensing Waveform based sensing

Cooperative sensing Centralized Distributed Non cooperative sensing Reactive/proactive sensing Multi dimension sensing

External

Hardware requirements Hidden primary user Problem Spread spectrum users Decision fusion Security

Std. comms. protocols employed

IEEE 802.11 IEEE 802.22 Bluetooth

It becomes difficult for the primary user to retain the spectrum with Cognitive technology [17]. Therefore a threshold is set to identify the primary user’s transmitted signal. Threshold setting is computation of a value k, selected with a target probability of detection Pd of the primary signal [18]. With an increased threshold value the probability of detection maximizes. Setting threshold value is a tradeoff between the performance matrices Pd, Pf and Pmd. For known ‘N’ (no. of samples) and ‘rw’ (Noise Variance), the optimal threshold value is calculated for a constant probability of false alarm Pf [23], e.g., Pf  0.1. The optimized Pf is given by k  Nð2r2w Þ Pf  Q pffiffiffiffi N ð2r2w Þ

!

 pffiffiffiffipffiffiffiffi kf ¼ Q1 ðPf Þ þ N N 2r2w

ð1Þ ð2Þ

The false alarm depends on number of samples N as well. Noise is characterized by its noise variance. In general, the normal distribution (Gaussian distribution) is taken for the analysis. Q function is also a distribution which indicates the tail distribution of normal distribution. For Gaussian distribution N (µ, r2) notation is used, where ‘µ’ and ‘r2’ are mean and variance respectively. The inverse Q−1(y) is used in digital communication to define a performance metric Q factor which is given by 20log10 (Q−1(y))

Cognitive Radio: A Conceptual Future Radio and Spectrum Sensing Techniques

159

dB. Q factor is expressed in dB. From the equation of ‘kf ’ it is evident that the threshold value depends on the Q factor, number of samples and the noise variance. The probability of detecting the signal in absence of message signal is termed as false alarm. The boundary limits of Pf is defined as 0 < Pf < 1. If Pf is 0.98, it indicates that there is a signal, although it may be more of noise. If Pf is 0.10, it still indicates that there is a signal, but now with less noisy background. When Pf (false alarm) is used as the performance metric in spectrum sensing algorithm during threshold detection (k) the efficiency of the sensing algorithm can be evaluated. The distance between the sensing node and the primary user also plays a major role even with an optimal threshold set. With an increased path loss, Pf or Pmd may go up. 3.2

Selection of Number of Samples of Input Signal

The objective of spectrum sensing is to achieve a target performance metric Pd to be high. This indicates that the system model applied as the design platform is more appropriate. In the previous section the threshold value k was selected as the optimum parameter to design the spectrum sensing algorithm. There is another way of increasing the efficiency of the sensing algorithm. This is by choosing the value of N. The Pd performance metric should not depend on k, so that the model can be designed which is mutually exclusive. To elaborate the meaning, efficiency of spectrum sensing: it is the ability of the model to detect the signal, in the region of various SNRs. One can differentiate two regions of model of operations, namely, low SNR and High SNR. SNR is signal to noise ratio. Low SNR means noise is more and signal strength is low. Any spectrum sensing algorithm, which gives better estimation of Pd in the low SNR region, is termed more efficient. Selecting the number of input samples to achieve a desired target performance metrics is also an optimal factor. The minimum required samples N is calculated as a function of SNR to achieve the target Pd and Pf. Excluding k from Pd and Pf, N is given by [19] h pffiffiffiffiffiffiffiffiffiffiffiffiffii2 N ¼ Q1 ðPf Þ  Q1 ðPd Þ 2c þ 1 c2

ð3Þ

where ‘c’ is signal to noise ratio. Even with Q−1(
) having decreasing property of function, a signal in a very low SNR region can be detected with higher values of input samples, with a known noise level. To achieve a required target performance metrics, the number of samples is in the order of C(c−2), that is, at low SNR, the energy detector needs more number of samples to detect the desired signal [33]. From the above analysis it is seen that if N is larger, then probability of detection Pd will be high. However as N is a product of ‘s’ sensing time and ‘fs’ sampling frequency, N = s * fs, there is an increase in sensing time by the detector as the number of samples increase. Time consumption with increase in samples forms the main drawback of spectrum sensing in energy detector. With low SNR, the numbers of samples required are more but there is a maximum allowable time of 2 s [19] for sensing and it doesn’t go well in real time scenarios. Hence clear tradeoff is seen selecting N as a design parameter in the spectrum sensing algorithm.

160

N. Rajanish and R. M. Banakar

4 Spectrum Sensing Approaches Spectrum sensing forms the main computational module in Cognitive Radio for communication setup with frequency being shared between primary user and secondary user. The transmitted signals can broadly be classified into Blind signals and Non Blind signals. Blind Signals are the one whose characteristic features are not know at the receiver whereas the non blind signal’s features viz. amplitude, phase and cyclic frequency are known to the receiver. Practically the non blind signals are implementable only at short distances and the detection probability Pd will be very high. With numerous researchers working on different techniques to bring out an effective and high speed spectrum sensing algorithm, few of the proposed and implemented work have been discussed below. The detection algorithms that are employed under blind signals are:4.1

Energy Detector

The most common algorithm to be implemented which needs no information of the received signal. It is also known as radiometry or periodogram [9]. It requires fewer computations to detect the message signal. The energy of the signal received is computed through the function r ðnÞ ¼ xðnÞ þ wðnÞ E ¼

XN n¼0

ð4Þ

r ð nÞ 2

ð5Þ

The computed energy is compared with a threshold set. The binary hypothesis for detecting a signal is given by

 where rðnÞ ¼

E  kjh1

ð6Þ

E hkjh0

ð7Þ

wðnÞ xðnÞ þ wðnÞ

h0 h1

If an energy sensing algorithm that can detect a signal at low signal to noise ratio i.e. at −20 dB in which the signal power is as low as −116 dBm and noise floor is of −96 dBm [19], then such algorithms can be termed better. Georgios et al. [20], address the issue of spectrum sensing using energy detection method by computing power spectral density of the received signal by implementing Welch periodogram algorithm and comparing the average value obtained with a precalculated threshold. The author performs several tests by varying different windows and number of input samples. According to their results the minimum number of samples required to detect a signal would be 1024 and more, below which the energy

Cognitive Radio: A Conceptual Future Radio and Spectrum Sensing Techniques

161

detector under performs [21–23]. Which means Energy detector requires more number of samples to detect a signal which in turn slows down the detecting process. 4.1.1 Eigen Value Based Spectrum Sensing Zeng and Liang [24], have worked on received signal’s covariance values through which the Eigen values of the matrix are computed using random matrix theory. With the Eigen values obtained, the presence of the signal is decided by employing either of the two sensing algorithms; Firstly, to calculate the ratio of Max. to Min. Eigen values (MME) or secondly, to calculate the ratio of average of the Eigen values to Minimum Eigen value (EME) obtained. The ratio is compared with a pre-set value (k) and the presence of signal is determined. The results obtained through this proposal are not quite optimum. The sensing algorithm performs well with fixed noise variance along with large input samples, whereas with varying noise variance, the algorithm fails to detect the signal accurately even with large input samples [8, 19, 25]. 4.1.2 Spectrum Sensing Through Differential Entropy Signals with highly non-correlation can be detected by computing Differential entropy [39] of the same. Differential entropy is calculated of the pdf fX(x) of the received signal which is given by Z1 fX log fX ðxÞ dx

hð X Þ ¼ 

ð8Þ

1

The estimated differential entropy of the message signal received is compared with pre calculated value ‘k’ with a constraint on probability of false alarm Pf. In [34], the author test verifies the same with SNR ranging between [0–50] dB in steps of 5 dB by calculating the generalized SNR through energy detection relation mentioned above. The differential entropy method performs better than the energy detection with increased samples and generalized SNR. 4.2

Nyquist and Sub-Nyquist Sampling Spectrum Sensing (Compressed Sensing)

Zhao et al. [26], address the issue of spectrum sensing by proposing a theory of compressed sensing executed in a time bound sequential flow to achieve higher accuracy in detection of Wideband signals (300 MHz–9.5 GHz). Wideband spectrum sensing (WBSS) can be classified into: 1. Nyquist wideband spectrum sensing 2. Sub-Nyquist wideband spectrum sensing In Nyquist WBSS, the received signal is sampled at a rate higher than nyquist rate. Hence sensing the whole band would introduce extreme time delay and also when the Nyquist rate for the actual spectrum is calculated, it introduces the usage of high-end wide band components, like wideband antenna, wideband RF spectrum, high speed

162

N. Rajanish and R. M. Banakar

analog to digital converters, which inevitably increases the cost of implementation [1, 10, 17, 40–42]. Whereas the Sub-Nyquist WBSS deals with under sampling of the received message signal. The spectrum sensing here is further classified into i. Compressive sensing based WBSS ii. Multichannel sub-nyquist WBSS Here, the authors [26] propose to divide the given wide band spectrum into ‘n’ subband spectrum so that the occupancy status by the primary user can be sensed by only sensing the sub-band. The sequential and periodic sensing of the sub-bands is undertaken so as to avoid overhead of sensing the whole band. They also propose a two stage change-point detection method to gather spectral occupancy variations and make quick decisions on spectral usage norms. They also introduce cooperative sensing along with compressed sensing to better identify the occupancy status of primary user. The results obtained through their technique are; low power signal sensing, accurate wideband detection, lesser delay in detection of signals, accuracy in detecting primary user occupancy, recovery of compressed signals [40–42]. 4.3

Co-operative Spectrum Sensing

Awasthi et al. [4], proposes Energy - efficient Technique in Cooperative spectrumsensing to verify the presence of a signal optimally with respect to time, decision threshold and number of secondary users. They also propose to reduce the consumed energy by decreasing the number of samples being processed. Cooperative sensing deals with fusion of inputs from ‘N’ number of secondary nodes and making a final decision based on them. Proper choosing of secondary nodes and voting scheme can reduce the energy consumption [2, 7, 27]. The Cooperative Spectrum sensing is achieved in two ways: 1. Centralized sensing 2. Distributed sensing. The cognitive devices spread around the primary user send the information to a central unit about the occupancy of the frequency. The central unit takes the decision and controls the traffic in centralized sensing [9]. In distributed sensing, the secondary users with cognitive capability spread around the primary user make their own decision to occupy spectrum which they consider to be free. This is more advantageous as it doesn’t require a central infrastructure. 4.4

Non - Cooperative Spectrum Sensing

Kamil and Yuan [28], proposes Non- Cooperative spectrum sensing. The algorithm employs conventional energy detector for detecting a signal. The signal received at the receiver antenna is delayed with predefined number of nodes, collated and processed at the energy detector. If there are ‘N’ number of verification nodes, then each one makes its own decision H0 (a hypothetical value to indicate the absence of signal) or H1 (a hypothetical value indicating the presence of a signal) and the results are compared with a predefined (threshold) value k. Each decision made is collaborated with the

Cognitive Radio: A Conceptual Future Radio and Spectrum Sensing Techniques

163

results of preceding decision device. Finally the decision about the existence of message signal is verified with either of two decision rules. One is ‘AND’ rule and the other is ‘OR’ rule [6, 28, 29]. In the former, iff, m == N, i.e., ‘m out of N’ where ‘N’ represents number of verification nodes and ‘m’ represents number of nodes which has ascertained H1, then the decision taken is, the signal exists else not exists. In OR rule, even if m = 1 which means that even if one out of N verification nodes ascertain H1, then the decision made is H1, i.e., the signal exists [30–32].

5 Conclusions Cognitive Radio forms the future of radio technology. Implementation of it would make an effective utilization of the spectrum and decongest the networks. This work provides an overview of Cognitive radio, its basic network architecture, spectrum sensing techniques and various algorithms implemented by researches across the globe. A wide spectrum sensing algorithms has been surveyed and their advantages and disadvantages have been discussed. Spectrum sensing forms one of the key functionality in CRs. Implementation of it in real time scenario would encounter many challenges to solve. A less complex, less computational algorithm with high speed and effective Cognitive Radio is the need of the hour.

References 1. Tandra R, Mishra SM, Sahai A (2009) What is a spectrum hole and what does it take to recognize one? Proc IEEE 97(5):824–848 2. Popescu A (2012) Cognitive radio networks. In: 2012 9th international conference on communications (COMM), Bucharest, pp 11–15 3. Hoyhtya M, Hekkala A, Karz M, Mammela A (2007) Spectrum awareness techniques and challenges for active spectrum sensing In: Cognitive wireless networks. Springer, pp 353– 372 4. Awasthi M, Kumar V, Nigam MJ (2017) Energy efficiency techniques in cooperative spectrum sensing: a survey. In: 3rd IEEE international conference on computational intelligence and Communication Technology (CICT), Ghaziabad, pp 1–6 5. Mitola J (2000) Cognitive radio: an integrated agent architecture for software defined radio. Ph.D. thesis, Bucharest, Royal Institute of Technology, Stockholm, Sweden 6. Steenkiste P, Sicker D, Minden G Raychaudhuri D (2009) Future directions in cognitive radio network Research. In: NSF workshop report 7. Akyildiz IF, Lee WY, Chowdhury KR (2009) CRAHNs: cognitive radio ad hoc networks. Ad Hoc Netw (Elsevier) 7(5):810–836 8. Haykin S (2005) Cognitive radio: brain-empowered wireless communications. IEEE J Sel Areas Commun 23(2):201–220 9. Yucek T, Arslan H (2009) A survey of spectrum sensing algorithms for cognitive radio applications. IEEE Commun Surv Tut 11(1):116–130 10. Steenkiste P, Sicker D, Minden G, Raychaudhuri D (2009) Future directions in cognitive radio network research. In: NSF workshop report, http://www.cs.cmu.edu/prs/NSF CRN Report Final.pdf. Accessed 9–10 Mar 2009

164

N. Rajanish and R. M. Banakar

11. Marinho J, Manterio E (2012) Cognitive radio: technology survey and future research directions. Wirel Netw 18(2):147–164 12. Stevenson C, Chouinard G, Lei ZW, Shellhammer SJ, Caldwell W (2009) The 1st cognitive radio wireless regional area network Std. IEEE Comm 47(1):130–138 13. Yucek T, Arslan H (2009) A survey of spectrum sensing algorithm for cognitive radio applications. IEEE Commun Surv Tutor II(1):116–130 14. Thomas RW, Friend DH, Dasilva LA, Mackenzie AB (2006) Cognitive networks: adaptation and learning to achieve end to-end performance objectives. IEEE Commun Mag 44(12):51–57 15. Joseph Mitola III (2001) Cognitive radio for flexible mobile multimedia communication. Mob Netw Appl 6:435–441 16. Lindu KE, Swamy MNS (2010) Wireless communication systems-RF subsystems to 4G enabling technologies. Cambridge University Press, Cambridge 17. Shanmugan KS, Breipohl AM Random signals (detection, estimation and data analysis). In: 3rd IEEE Intl. Conf. on Computational Intel. and Comms. Tech. Wiley, (July 1988) 18. Verma G, Sahu OP (2016) Intelligent selection of threshold in cognitive radio system. Telecommun. Syst. 63:11–15 19. Atapattu et al (2014) Energy detection for spectrum sensing in cognitive radio. Springer Briefs in Computer Science. Springer, Heidelberg, pp 11–26 20. Polymerou G, Kalialakis C, Sahalos JN (2009) Spectrum sensing for cognitive radio networks based on energy detectors. In: Conference paper 21. Hou YT, Nekovee M, Wyglinski AM (2009) Cognitive radio Communications and networks: principles and practice. Elsevier/Academic Press, Amsterdam 22. Huang XH, Du KL, Lai AKY, Cheng KKM (2001) A unified software radio arch. IEEE 3rd workshop on signal processing advances in wireless communications, Taiwan, pp 330–333 23. Jamin A, Mahonen P, Shelby Z (2001) Software radio implementability of wireless LANs. In: Del Re E ed Software radio: technology & services, London, Springer, pp 173–185 24. Zeng Y, Liang Y-C (2009) Eigen value based spectrum sensing algorithms for cognitive radio. IEEE Trans on Comm 57(6):1784–1793 25. Ghasemi A, Sousa ES (2005) Collaborative spectrum sensing for opportunistic access in fading environments. In: 1st IEEE international symposium on new frontiers in dynamic spectrum access networks, Baltimore, pp 131–136 26. Zhao J, Liu Q, Wang X, Mao S (2017) Scheduled sequential compressed spectrum sensing for wideband cognitive radios. IEEE Trans Mob Comput pp(99):1. https://doi.org/10.1109/ tmc.2017.2744621 27. Ahsant B, Viswanathan R, eds (2013) A review of cooperative spectrum sensing in cognitive radios. In: Mukhopadhyay S, Jayasundera K, Fuchs A Advancement in sensing technology, smart sensors, measurement & Instrumentation. Springer, Heidelberg, vol 1, pp 69–80 28. Kamil NH, Yuan X (2010) Detection proposal schemes for spectrum sensing in cognitive radio. Wirel Sens Netw Sci Res 2:365–372 29. Alexandru P Cognitive radio networks: elements and architectures. Ph.D. thesis Blekinge Institute of Technology, Karlskrona Sweden, Work in preparation 30. Ciftci S, Torlak M (2008) A comparison of energy detectability models for spectrum sensing. Proceedings of IEEE global telecommunications conference (GLOBECOM), New Orleans, pp 1–5, 30 31. Cabric D, Mishra SM, Brodersen RW (2004) Implementation issues in spectrum sensing for cognitive radios. In: Signals, systems and computers, Pacific Grove, CA, USA, pp 772–776 32. U.S. DoD (1997) Programmable Modular Communications Systems (PMCS) Guidance Document. US Department, Washington DC

Cognitive Radio: A Conceptual Future Radio and Spectrum Sensing Techniques

165

33. Cabric D, Tkachenko A, Brodersen RW (2006) Experimental study of spectrum sensing based on energy detection and network cooperation. In: Proceedings of international workshop on technology and policy for accessing spectrum (TAPAS), Boston, vol 5, Article No 12 34. Gurugopinath S, Muralishankar R, Shankar HN (2016) Spectrum sensing in the presence of cauchy noise through differential entropy. In: IEEE Conference paper (2016) 35. Xing Y, Chandramouli R, Mangold S (2006) Dynamic spectrum access in open spectrum wireless networks. IEEE J Sel Areas Comms 24(3):626–637 36. Van der Perre L et al (2009) Green software defined radios, series on integrated Ckts & Systems. IEEE J Sel Areas Comms 1:1–157 37. Urkowitz H (1967) Energy detection of unknown deterministic signals. Proc IEEE 55 (4):523–531 38. Mody AN, Chouinard G (2010) IEEE 802.22 wireless regional area networks, enabling rural broadband wireless access using cognitive radio tech. IEEE Conference Paper 39. Cover TM, Thomas JM (2005) Elements of information theory, 2nd edn. Wiley, Hoboken 40. Sun H, Nallanathan A, Wang CX, Chenm Y (2013) Wideband spectrum sensing for cognitive radio networks: a survey. IEEE Wireless Comms 20(2):78–81 41. Quan Z, Cui S, Sayed AH, Poor HV (2009) Optimal multiband joint detection for spectrum sensing in cognitive radio networks. IEEE Trans Sig Proc 57(3):1128–1140 42. Tian Z, Giannakis G (2007) Compressive sensing for wideband cognitive radios. In: Proceedings of IEEE international conference on acoustics, speech, and signal processing, Honolulu, HI, USA, pp 1357–1360

Non-max Suppression for Real-Time Human Localization in Long Wavelength Infrared Region Anuroop Mrutyunjay1(&), Pragnya Kondrakunta1, and Hemalatha Rallapalli2 1

FoGR Technologies Private Limited, Hyderabad 500 007, India [email protected] 2 Department of Electronics and Communication Engineering, Osmania University College of Engineering, Hyderabad 500 044, India

Abstract. The engenderment of thermal imaging techniques intended to provide military surveillance but, the then-nascent stage of this technology witnessed manual human detection. Significant research has been conducted in deep learning algorithms for accurate human detection, yet, so far it is only possible in captured images. In this research, we have explored YOLO, a state of the art algorithm for real-time object detection, in the context of Long Wave Infrared imaging. Exclusive methods for each - human detection and real-time object detection, hold the key to a more sophisticated approach. In pursuit of a unified system, this paper discusses complex localization algorithms for realtime human detection in a thermal feed. The efficacy of the proposed idea has been recorded and reported. Keywords: YOLO for FLIR
FLIR
Human detection
Infrared thermography
Thermal imaging
Convolutional neural network

1 Introduction Computer vision is a field that examines and interprets the ambient information from digital imagery. A crucial factor to achieve exceptional results is the use of wideranging universal data. However, the collection of visional data encounters a major setback in the case of poor/nil lighting and the presence of obscurants (such as smoke and fog). Thermal imaging is an approach that eliminates the limitations of conventional imaging modalities. The operation of a thermal imager is based on the principle that any matter above the absolute zero temperature emits heat radiation called the heat signature of the object. The possibility of two objects radiating an identical amount of heat is very rare. The thermal camera captures these infrared radiations to detect the minute temperature differences between objects. This temperature difference data is mapped to produce a thermogram. Thermal images are usually depicted in grayscale with black symbolizing low temperatures, while white corresponds to high temperatures and the extent of grey representing the temperature difference. © Springer Nature Switzerland AG 2020 S. C. Satapathy et al. (Eds.): ICETE 2019, LAIS 4, pp. 166–174, 2020. https://doi.org/10.1007/978-3-030-24318-0_20

Non-max Suppression for Real-Time Human Localization

167

The infrared band is parted into smaller regions due to the constraints of sensors to cater to the large frequency range. Just adjacent to the visible spectrum is the nearinfrared (NIR) region followed by Short-Wavelength Infrared (SWIR). Further away, positioned are the Mid-Wavelength Infrared (MWIR) and Long-Wavelength Infrared (LWIR). The Far-infrared marks the end of the Infrared region. The ability to capture data even in complete absence of illumination prompted the use of Long-Wavelength Infrared in this work.

2 Related Work Research [2] in 2005, aiming to identify pedestrians in visible light images, adopted linear Support Vector Machines. In their work, they proved the profound performance of the grids of Histogram Oriented Gradient (HOG) in comparison to the existent feature sets for human beings. The visible light images exhibited limited scope and implausible night-time detection. The need for imagery in unlit scenarios was widely accepted for its extended opportunities. Formerly, the thermal camera equipment was developed as intelligence for the military. Later, it was declassified and commercialized under strict regulations but was priced high. Access to LWIR tools for an average man was extremely challenging until the late 1990s. Over recent years, the cost has reduced, enabling more research and exploration in the field. In an attempt to improve human detection accuracy for unlighted settings, the research [3] deploys multiple Far Infrared cameras as well as visible light cameras, and exposes the data to spatial-temporal filtering and Min-Max fusion techniques. Another research [4] explores the possibility of the same with Near Infrared cameras. Nevertheless, corresponding the points of different cameras and their calibration, proves to be a tedious task not to mention the additional time taken for processing multiple copies of the image. Yet another approach is the Gaussian Mixture Model utilized in research [5], which eliminates multiple camera models and their associated calibration. In their work to scrutinize a thermal video, challenges included far field human detection and contour detection. Also, in the case of NIR cameras, respective illuminators had to be used which had restricted angle and distance covered. The research [6] discloses a system for the detection of remnant PFM-1 ‘Butterfly Mines’, which are notorious plastic mines, using long wavelength infrared (LWIR) images captured by an FLIR Vue Pro R mounted on an Unmanned Aerial Vehicle. The mine casing’s characteristic apparent temperature and shape, enable easy detection, identification and classification. The processing of raster data was in pix4Dmapper while the apprehension of the geospatial data used ArcMap 10. The detection of humans in visible light and infrared region has moved milestones. Notably, not most of the approaches have attempted the exertion of human identification in real-time with live thermal feed. Even otherwise, their performance was deemed to be insufficient.

168

A. Mrutyunjay et al.

3 Proposed Model The inadequacies put forth during the coalescence of extant infrared and deep learning research, promoted the orchestration of the ideas and implementations in this manuscript. A long wavelength infrared feed is analyzed and visually manipulated by the integration of deep learning algorithms trained in context to abundant thermal feed. The digital image sequences from an FLIR thermal imager are cautiously obtained and predicted for humans. Individuals are detected and their positions are precisely localized (in a bounding box) using the You Only Look Once [YOLO] technique. The bounding box coordinates and confidence scores corresponding to each detected human are realized and recorded. The YOLO algorithm is customized to befit the present technical scenario. The implementation methodology, in addition to its aftermath, has been discussed below.

4 Working Algorithm Common Image Classification problems rely on the accuracy rendered by the algorithm-in-context and, more specifically, the image data supplied for efficient backpropagation. However, the scope of localization problems is vast in comparison, and the data required for machine training is much more embellished. Besides multiclass digital signal data, pixel landmarks for distinct regions of interests are furnished during training, often known as bounding boxes. Bounding box values, precisely reveal the loci for varied regions of interests in one image. The four values Bx, By, Bh and Bw that constitute a bounding box aid the system in accurately predicting the position of an object. The values Bx and By disclose the pixel positioning coordinates of an object’s left extreme, while Bh and Bw conserve the area parameters for a rectangular bounding region, within which the object is inscribed. Representation of this data to a system can be challenging and is thus passed as a vector with a fixed length (N). 3 Pc 6 Bx 7 7 6 6 By 7 7 6 6 Bh 7 7 6 y¼6 7 6 Bw 7 6 C1 7 7 6 4 C2 5 C3 2

ð1Þ

The sliding window algorithm, even in the most recent times, does not lose its relevance in the field of modern object detection. The conceptualization of bounding box regions in sliding window exemplifies the technique of subsampling based localization. Given an input image, subsamples are processed for the detection of corresponding regions of interest thereby implicitly imparting localization to a traditional classification task. The network is formerly trained on, what are usually known as, tiles of digital images, representing the various classes of a multiclass data stream.

Non-max Suppression for Real-Time Human Localization

169

Stride and other subsampling parameters show relevance in case of window selection during classification. Processing extensive samples of data, in furtherance to subsampling, delivers adequate recognition performance but demands computationally expensive training and prediction. Also, the task of subsampling or cropping smaller regions of an image prior to detection consumes expansive memory and time thus rendering the algorithm slow and profligate in terms of memory. In the aforementioned scenario, the integration of convolutional networks and algorithms with bounding box based localization, tend to deliver the best of both worlds. The implementation of a sliding window equivalent, utilizing the classification mechanism of the renowned convolutional neural networks is appealing both in terms of computational inexpensiveness and detection time-optimization. The fully connected layers of a basic convolutional neural network architecture are modified and filters are appended to each layer of the existing configuration. The processed vectors from the convolutional and pooling layers are fed to the fully connected layers where a convolution operation is performed at each stage with preset filters. Thus, the output of the CNN is now a multi-dimensional vector of size K  K  N (Fig. 1).

Fig. 1. Convolutional implementation of sliding window algorithm

You Only Look Once or YOLO is a state-of-the-art algorithm for object recognition, based on a regression model to fit spatially separated bounding box data and their associated class probabilities. It is widely regarded as one of the most successful implementations of a convolutional net based sliding window algorithm. YOLO runs completely on a single unified neural network and thus can be optimized end-to-end in order to achieve high-yielding performance during object classification and localization. It is known to have outperformed all contemporary detection algorithms such as the DPM and the R-CNN by a wide margin when tested on gold standard datasets such as the Picasso dataset and the Microsoft COCO dataset. Since the architecture is completely based on a single network it is incredibly fast for detection and processes images at real-time speeds of up to 155 frames per second. The network discussed in this paper is a COCO pre-trained unified YOLO net. Multitudinous bounding boxes characterized by disparate areas and origin points are obtained as a result of the culmination of a highly trained YOLO net. This leads to ambiguity in the recognition of the trained classes of data. This discrepancy can be rationalized by accepting that the network is adept at recognizing the objects present in the feed but positions multiple probable bounding boxes around one object. In our effort to implement a regulating mechanism to curb the yield of such multitudinous bounding regions, we employ Non-max suppression, also known as Non-maxima

170

A. Mrutyunjay et al.

suppression. It has been stated that the network, along with the parameters of the bounding box, outputs the probability of prediction. Thus, for every bounding box obtained, we have a corresponding probability score that aids in identifying the region which inscribes the highest fraction of object data. Consider an algorithm outputs five bounding boxes for an object in an image. The bounding box that has the highest probability is marked with a bright colour, while the rest of the boxes are suppressed in colour and are masked from the user. Non-max suppression precludes the plausibility of multi-box confoundment in the output feed.

5 Dataset Collection and Training the Model on YOLO Neural networks are one of the primary points of emphasis when drawn to a supervised learning backdrop. Pattern recognition based on labelled digital signal data is best achieved using the neural network algorithm. Data forms the fundamental element for all supervised learning tasks and convolutional neural networks are no aberration. In the context of this research, the data used is digital images in a JPEG format captured using an FLIR Tau2 (13 mm f/1.0, 45-degree HFOV and 37-degree VFOV) (Fig. 2).

Fig. 2. Training the dataset

The images are subsamples of a long length video recording using the abovementioned thermal instrument. The input images are in a 640  512 resolution with most of them captured under generally clear sky conditions, both, during day and night. The thermal camera was mounted on an automobile in order to obtain copious input samples characterized by a rich feature set. Images from the video were sampled at a rate of two images per second. Labels coupled to image data can be construed in multiple formats. Annotations are regarded as the most endorsed approach to labelled data. They are, essentially, manuscripts of the bounding box parameters and class labels for an inscribed object.

Non-max Suppression for Real-Time Human Localization

171

Initially, all the collected data is to be annotated using the python Matplotlib library and the mouse click-release events which can be recorded using the innate mouse event listeners. Matplotlib returns the coordinates of the bounding regions, drawn manually using the mouse click and release phenomenon. The class labels for a particular object class is explicitly now mentioned. The position coordinates are then interpreted into machine parsable XML format. The dataset typically constitutes 10,228 images of people captured in the longwave infrared region. The dataset has been carefully scrutinized to scrupulously incorporate the images of people in a diverse assortment of positions, arrangement, attitudes and poses. Annotations have been expressed for the sum of images wherever necessary to depict humans. Multiple annotations can be identified for all the images where a crowd is evident. The dataset has been further segregated into train and test folders containing 6,137 and 4,091 images respectively. The algorithm is initially trained on a subsample of the train folder to attain hyper-parameter optimization followed by a full volume train procedure. A COCO pre-trained YOLO network is loaded and is dispensed with the label and training image directories. An initial prediction loss (108.90049021), and thereafter periodic losses, are recorded to verify convergence of the complex convolutional network. The training is uninterrupted until maximum loss convergence is realized after 376 epochs and a loss convergence graph versus time is plotted as depicted below. The weights and biases of the network delineating its learning and training policy are backed up to a checkpoint folder periodically for every 125 steps (Fig. 3).

Fig. 3. Error/loss rate vs time convergence graph

6 Results The motivation for our work is twofold: (1) To realize real-time object detection in a thermal feed. (2) To implement YOLO on non-conventional imagery such as the long wave infrared. To our knowledge, no performance results of using the YOLO algorithm on the long wavelength infrared region dataset (FLIR Thermal dataset) have been

172

A. Mrutyunjay et al.

published. However, a research conducted by the FLIR ADAS Team published efficiencies of a RefineDetect512 deep neural net yielding a moving average loss of 0.794 in the detection of humans. An all-time prediction confidence high of 93% has been achieved using the transfer learning techniques of YOLO in this research. Humans in both non-occlusive and occlusive, crowded environments are distinctly localized in well-defined bounding regions. A moving average loss rate of 0.352698 is attained (Figs. 4, 5, 6, 7 and Table 1).

Fig. 4. Input feed for human detection in solitary.

Fig. 6. Input feed for human detection in a crowd.

Fig. 5. Output feed from the algorithm encloses the human in a white bounding box.

Fig. 7. Output feed from the algorithm encloses the humans in multiple white bounding boxes.

Table 1. Comparison of moving average loss in RefineDet512 and our algorithm Algorithm Moving average loss RefineDet512 0.794 Our algorithm (YOLO) 0.352

Non-max Suppression for Real-Time Human Localization

173

7 Conclusion and Scope We have shown that real-time localization in the long wave infrared region can be acutely accomplished using the YOLO algorithm. The meticulous results obtained in this work can set a benchmark for similar experiments employing other profound deep learning algorithms. The aforementioned approach can be extended to detect manifold objects even in occlusive environments. The scope for this research is beyond horizons, provided, today’s technological zeal. Future cutting-edge applications involving the assimilation of this system with autonomous or manually controlled bots, rovers and drones can open doors to a wide range of security and signal processing explorations. The adept detection mechanism of the present model can be an addition to the surveillance systems, not only in civilian zones but also at the borders, thus, equipping our security forces in the detection of inter-border trespassing. Promotion of advanced defence research in this field reinforces the establishment of secure colonies for the civilians as well as the military.

References 1. Redmon J, Farhadi A (2018) YOLOv3: an incremental improvement. Computer vision and pattern recognition, arXiv:1804.02767 2. Dalal N, Triggs B (2005) Histograms of oriented gradients for human detection. In: Proceedings of the IEEE computer society conference on computer vision and pattern recognition, San Diego, CA, USA, 20–25 June 2005, pp 1–8 3. Chen Y, Han C (2008) Night-time pedestrian detection by visual-infrared video fusion. In: Proceedings of the 7th world congress on intelligent control and automation, Chongqing, China, 25–27 June 2008, pp 5079–5084 4. Yuan Y, Lu X, Chen X (2015) Multispectral pedestrian detection. Sig Process 110:94–100 5. Komagal E, Seenivasan V, Anand K, Anand raj CP (2014) Human detection in hours of darkness using Gaussian mixture model algorithm. Int J Inform Sci Tech 4:83–89 6. Smet T, Nikulin A, Baur J, Frazer, W (2018) Detection and Identification of Remnant PFM1 ‘Butterfly Mines’ with a UAV-based thermal-imaging protocol. Remote Sens 10. https:// doi.org/10.3390/rs10111672 7. Kim J et al (2017) Convolutional neural network-based human detection in nighttime images using visible light camera sensors. Sensors (Basel, Switzerland) 17(5):1065. https://doi.org/ 10.3390/s17051065 8. Scherer D, Muller A, Behnke S (2010) Evaluation of pooling operations in convolutional architectures for object recognition. In: 20th international conference on artificial neural networks (ICANN), Thessaloniki, Greece 9. Zhang S et al (2018) Single-shot refinement neural network for object detection, computer vision and pattern recognition, arXiv:1711.06897 10. Budzan S (2015) Human detection in thermal images using low-level features. In: Measurement automation monitoring, June 2015, vol 61, no 06 11. Ren S, He K, Girshick R, Sun J (2016) Faster R-CNN: towards real-time object detection with region proposal networks. In: Advances in neural information processing systems, vol 28 (NIPS 2015), arXiv:1506.01497v3

174

A. Mrutyunjay et al.

12. Dai J, Li Y, He K, Sun J (2016) R-FCN: object detection via region-based fully convolutional networks. In: Advances in neural information processing systems, vol 29 (NIPS 2016) arXiv:1605.06409v2 13. Liu W et al (2016) SSD: Single Shot MultiBox Detector. In: Leibe B, Matas J, Sebe N, Welling M (eds) Computer vision – ECCV 2016. ECCV 2016. Lecture Notes in Computer Science, vol 9905. Springer, Cham 14. Lin T-Y, Goyal P, Girshick R He, K, Dollár P (2018) Focal loss for dense object detection, arXiv:1708.02002v2 15. Lin T-Y et al (2017) Feature pyramid networks for object detection, arXiv:1612.03144v2

RF Energy Harvesting for Spectrum Management in Cognitive Radio Networks Suneetha, Harini(&), Yashaswini, and Hari Haran Department of Electronics and Communication Engineering, Vignana Bharathi Institute of Technology, Hyderabad, India [email protected]

Abstract. Cognitive radio is an intelligent and adaptive software radio that automatically detects the accessible channels in a wireless network. It is widely examined to perform the spectrum management. Spectrum management is a technique used to efficiently utilize the available radio spectrum. Due to everincreasing implementation in wireless communications, the classical static spectrum allocation stratagem fails to meet the new possible demands. In practice there remain many challenges and one of them is performing spectrum management in dynamic environment. To overcome this and manage the spectrum efficiently proper spectrum prediction and resource allocation to the secondary users (SUs) has to be done. In this paper prediction accuracy of available channels have been analyzed using Markov chain algorithms by implementing RF energy harvesting on SUs. Channel selection and Band utilization of SUs in Cognitive radio networks based on Dynamic user prediction (DUP) and spectrum prediction (SP) models were proposed. Keywords: Prediction accuracy


H2BMM
SP
DUP
Energy harvesting

1 Introduction Radio spectrum is the most valuable resource for the wireless communications but due to the limited amount of spectrum available, the spectrum efficiency is less as there is increase in the Quality of service requirements day by day, this spectrum scarcity can further be improved by using several methods, among which Cognitive Radio Networks is one of the best method that gives the promising solution [1]. A CRN is a form of wireless communication in which the transceiver can detect whether a particular channel is idle or busy and helps the user to immediately move to vacant channels avoiding the occupied ones. This CRN helps to adapt the effective utilization of spectrum in a crowded electromagnetic spectrum. The challenges that a CRN does are idle portion of the spectrum is to be sensed, best available channel is selected, coordination between the user and the channel is made and search for another channels if the said channel is busy. Spectrum Management is the process of regulating effective utilization of spectrum [2]. It manages the spectrum with ever growing new applications. Spectrum scarcity is the most predominant phenomenon in the spectrum management helps to find a best solution. The ability of Spectrum Management can be defined through four main © Springer Nature Switzerland AG 2020 S. C. Satapathy et al. (Eds.): ICETE 2019, LAIS 4, pp. 175–184, 2020. https://doi.org/10.1007/978-3-030-24318-0_21

176

Suneetha et al.

functions: Spectrum sensing, Spectrum decision, Spectrum sharing and Spectrum mobility. Spectrum sensing is the method used to identify the status of the channel whether the channel is idle or busy. In CRNs, it helps to monitor the frequency band periodically. Spectrum decision is the capability of selecting the best available channel for the users. In CRNs, the Cognitive Base station selects the best unoccupied channel and allocates to the users based on the availability. Spectrum sharing, for a large number of users there may be a chance of overlapping where the CRNs helps to avoid the conflicts and improves the better sharing of channels among users. Spectrum mobility, In general the users are dynamic and hence the CRNs will make a decision where the communication is to be continued in another part of the vacant spectrum. In this paper, how a channel status can be accurately detected if the number of users are more which is termed as spectrum prediction [2]. Spectrum prediction is the process of analyzing and predicting the behavior of channels. Bandwidth utilization is utilizing of available bandwidth of the channel occupied by the users. We proposed a radio frequency energy harvesting technique to reduce the amount of external power consumed by the user such that it improves the performance of the cognitive radio network.

2 Related Work In cognitive radio networks, there are two types of users, primary users (PUs) and secondary users (SUs). When there are less number of users or applications then the static spectrum allocation can be consider. In practical scenario there is a change in position of these users, this causes a need for dynamic spectrum allocation [8]. In a mobile environment, the users experience a high interference from other users. The spectrum prediction and bandwidth utilization helps in regulating spectrum management. 2.1

Spectrum Prediction

Many spectrum prediction techniques have been proposed to obtain a stable handoff. One of the technique is implementing algorithms using Markov model [4]. Markov Model Based In real applications, the actual state of the channel is hidden from the SUs due to noise and interference persisting in the network. So, the spectrum prediction algorithms are developed by referring hidden Markov model (HMM) [1]. In HMM we train the state transition probability and emission probability matrix. But the drawback of HMM is that it will not consider the dwell time of the channel. Hence to overcome this high order Markov model (HOMM) [7] and hidden bivariate Markov model (HBMM) [1] are considered to include the hidden and bivariate features. By combining them, high order hidden bivariate Markov model (H2BMM) is proposed which will also consider the movement of SUs. Hence we are implementing this H2BMM to achieve the required efficiency.

RF Energy Harvesting for Spectrum Management in Cognitive Radio Networks

2.2

177

Channel Selection

In practical there are more number of users who are dynamically moving but there may be a chance of overlapping. In such cases the user must be able to select the best channel suitable [6]. Analytic Hierarchy Process Based It reduces the complex decision and deals in perfect decision making. It splits the complex problem into factors with hierarchy making decisions in which each can be solved independently. It considers a set of assessments along with a set of alternative in which the best decision is made. In general not just one best suitable single criterion is known rather selects the best among different criteria. AHP generates a weight for each of the different criteria according to the decision maker. The higher the weight, the better is the achievement of the option. General implementation of AHP includes: 1. Vector of each criterion weights must be analyzed. 2. Matrix should be computed. 3. Ranking the priorities. This method is applicable only for single user. For more number of users to extend the AHP multiple analytical hierarchical processes are used.

3 Proposed Spectrum Management Strategy In this module, we initiate a spectrum management strategy to predict an efficient spectrum which is further utilized by users using the channel selection. H2BMM and advance H2BMM algorithms are proposed for predicting spectrum in CRN. A scenario with mobile users is taken which consists of multiple channels. As shown in the first scenario the network consists of a single user and five idle channels. So, the user has a chance to utilize any of the five channels. Hence the probability of a single channel to get utilized by the user is less. In the second scenario the user traffic increases but the number of channels available is constant. Therefore any of the users will definitely utilize any one of the channel, i.e. the probability of the channel being utilized by the user is more. From the scenario if the User1 wants to access channel A or B then the CBS gives the information whether the first sensed channel is in either idle or busy state using Markov model. Based on the prediction results obtained, CBS makes a feasible decision on the static or dynamic users. If there are multiple users who want to access the available spectrum sensed by the prediction can be accessed among those users and calculated the bandwidth utilized. Once RF Energy is harvested on secondary users which reflect on active users resulting in improvement of utilization and can be used further. In case if the user wants to access channel D or C the same process continues, even though if the channel is idle more power transmission is required to send the information. In such cases the energy which is harvested before can be utilized for the transmission such that the power can be compensated which further reduces the number of handoffs.

178

Suneetha et al.

Fig. 1. Scenario (a) one user has the chance to utilize any of the four channels (b) If there are more number of users, any of the users can utilize a particular channel hence the efficiency is high

4 Spectrum Prediction and Dynamic User Prediction In order to predict the future position of users and the status of users we define two cases spectrum prediction and Dynamic User Prediction. 4.1

Spectrum Prediction

In this section, High order hidden bivariate Markov model and Advance high order hidden bivariate Markov model are adopted to know the prediction accuracy of a channel [5]. In H2BMM to find the channel behavior hidden and underlying process are used. The states of the channels are also divided into multiple sub states to know the status of the dwell time. Dwell time is the duration of the channel remaining in the current. Due to the signal to noise interference, sensing results may or may not represent the actual state of channel [3]. Hence the state is hidden form users. Let X = {“0”, “1”} indicates the hidden state space, with “0” and “1” representing the idle and busy states of channel respectively. Correspondingly, let Y = {“0”, “1”} denote the observation state space, with “0” and “1” indicating that the sensed results is idle and busy respectively. The role of spectrum prediction is to predict future channel states from the previous states. The algorithms are implemented by generating the hidden state chain and observation state chain. The duration time for each state is the dwell time of the channel is indicated as either “idle” or “busy” which tells the status of the channel. The dwell/duration time is that how long the channel remains in the current state which can be predicted by introducing transient state probability. {Xn} is denoted as the hidden state chain and {Yn} is the observation chain which is obtained by Gaussian distribution. Transition matrix is the probability of the change that takes place between two states and emission matrix is theQoutput probability extracted after the transition matrix isQobtained. Here, k = {A, B, } where A = Transition matrix, B = Emission matrix, = Initial state probability.

RF Energy Harvesting for Spectrum Management in Cognitive Radio Networks

179

The training algorithm is implemented as:

Where {Xn} = Hidden process {Yn} = Observation process G = transition matrix P = initial probability distribution µ = vector mean of observed signal strengths R = Vector of observed signal strengths variances 4.2

Dynamic User Prediction

In practical the user is not static and the behavior of the channel changes with respect to time and space which sometimes leads to more handoff delay [6]. In order to handle such situation monitoring and predicting the status of the user is to be analyzed. Many of such predictions are made using several algorithms among which Markov model has highest prediction accuracy. With H2BMM the movement can be predicted. Here users note the required information along with the sensing result periodically and pass it to CBS. However, not always the situation is same. As the situation becomes complex we go for predicting the channel behavior that depend on several previous states (Fig. 2).

180

Suneetha et al.

Fig. 2. User1 is moving from one position to another from time T1 to T2

4.3

Channel Selection Algorithm

After the user mobility is predicted and the activity is defined the best channel must be predicted using analytical hierarchy process based weighted algorithm [10]. Analytical Hierarchy process Depending upon the performance criteria the channel selection problem is divided. At the beginning the decision makers is set to 4. A judgment matrix is defined for the elements having different factors and the future vector is defined in order to obtain weight vector of AHP. Each factor is reflected using the Saaty ninth scale which indicates the equal importance from integer 1 to 9 where the integers on the left are taken as actual judgment value and the integers on the right indicates the reciprocal value say for example for the integer 9 which is on the right side of the Saaty ninth scale is taken as 1/9 (for aij = 1/aij) according to the users preference. From which eigenvalue is calculated to the judgment value that is defined and hence the eigenvectors are calculated. Each element represents the weight vector. When this is subjected to decision algorithm based on relative weights of different factors gives multiple analytical hierarchy process (M-AHP) [11].

5 RF Energy Harvesting CRN enables to have less power consumption, such that it improves overall utilization and efficiency of spectrum [9]. When the base stations transmit their signals to the mobiles, in some areas at the boundary of cell site or at areas where there is more traffic etc., large amount of power is needed for transmission. Here, the user will be capable of using the energy (1) that which is harvested during the period where the transmission needs low power such that, secondary users (SUs) can utilize not only the available

RF Energy Harvesting for Spectrum Management in Cognitive Radio Networks

181

spectrum but also can predict whether the channel is idle by which energy is preserved when the channel is busy [12]. RF energy harvesting can be calculated using (2), sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi 2 Pn  i;m  H n:m i¼1 H RMSE ¼ n NRMSE ¼

RMSE     max H i;m  min H i;m

ð1Þ ð2Þ

Where RMSE = Root mean square error NRMSE = Normalized root mean square error n = Number of channels (n = 4) H i;m H i;m ¼ Actual target H 0i;m H 0i;m ¼ Predicted target m represents the mth frequency bin

6 Simulation Results In this paper, simulations are performed to assess the proposed spectrum management strategy in the particular aspect of prediction accuracy with and without RF energy harvesting and spectrum utilization. For simulation, the number of SUs are considered from 2 to 50 and the PUs are taken as 3 as shown in fig. The users randomly move in 500 by 500 m area. The available spectrum of 100 MHz is divided into four sub channels which is utilized by the users. For data transfer each SU utilizes a bandwidth of 5 MHz and assumed one slot for spectrum sensing, remaining slots for data transfer. We assume sensed results are collected by the base station from multiple users. In algorithm H2BMM, SUs perform spectrum prediction individually and deliver the local sensing outcomes to the base station. When an idle PU is found, the values in the training matrices are updated according to the new sensed results so that the new spectrum can be predicted. The advance H2BMM attains higher prediction accuracy as more number of sensed results is sent to the base station by SUs. From the scenario Fig. 1(b) we are interested to know the channel occupancy of channel D whose bandwidth is 100 MHz and each user has the capacity to use 5 MHz by which the overall channel can be occupied by 20 users. There may be more or less numbers of users according to their requirement. With the help of same situation if we want to know the prediction accuracy of channel D from Fig. 3, let there be are only four number of users then the obtained prediction accuracy of occupying that particular channel will be 66% and is increased to approximately 68% after the energy is harvested. If the number of users increased to 20 the prediction accuracy is almost the same as the H2BMM can sense only one particular channel at a time. Figure 4 is the obtained spectrum prediction for Adv. H2BMM before and after the energy is harvested here the prediction accuracy is less when there are less number of users and more for more number of users.

182

6.1

Suneetha et al.

H2BMM Before and After Energy Harvesting

Fig. 3. H2BMM before and after energy harvesting

. 6.2

Advance H2BMM Before and After Energy Harvesting

.

Fig. 4. Advance H2BMM before and after energy harvesting

RF Energy Harvesting for Spectrum Management in Cognitive Radio Networks

6.3

183

Bandwidth Utilization

Figure 5 shows the bandwidth efficiency of a CRN in various modes as Spectrum Prediction, Dynamic User Prediction, Joint Prediction and No Prediction used for channel allocation. Considering the users to be 50, channel bandwidth is 100 MHz then channel occupied by each user will be 2 MHz. From the above graph we can analyze the situation as, if there are 25 users then the channel occupied by them will be 40% for DUP and almost the same for SP incase if both the methods are combined then the bandwidth utilized is more as the CBS can predict both user mobility and spectrum availability. Thus, the bandwidth utilization is high for joint Dynamic User Prediction and spectrum prediction strategy.

Fig. 5. Bandwidth utilization of spectrum management modes for change in SUs

7 Conclusion In this paper, prediction based spectrum management strategy with RF Energy harvesting was implemented which results in high performance improvements on spectrum utilization, where the Markov model, user mobility, spectrum prediction and hybrid analytical hierarchy process are integrated. Based on these results cognitive base station is capable of advocating channels to users in a timely manner and the multiple analytical hierarchy process helps to reduce the computational complexity. As future scope evaluate the bandwidth utilization after energy saving.

184

Suneetha et al.

References 1. Zhao Y, Hong Z, Wang G, Huang J (2016) High-order hidden bivariate Markov model: a novel approach on spectrum prediction. In: Proceedings of the 25th IEEE international conference on computer communication and networks, August 2016, pp 1–7 2. Lin Z, Jiang X, Huang L, Yao Y (2009) A energy prediction based spectrum sensing approach for cognitive radio networks. In: Proceedings of international conference on wireless communications, networking and mobile computing, pp 1–4 3. Welch LR (2003) Hidden Markov models and the Baum-Welch algorithm. IEEE Inf Theory Soc Newsl. 53(4):10–13 4. Tragos EZ, Zeadally S, Fragkiadakis AG, Siris VA (2013) Spectrum assignment in cognitive radio networks: a comprehensive survey. IEEE Commun Surv Tuts 15(3):1108–1135 5. Nguyen T, Mark BL, Ephraim Y (2013) Spectrum sensing using a hidden bivariate Markov model. IEEE Trans Wirel Commun 12(9):4582–4591 6. Xing X, Jing T, Cheng W, Huo Y, Cheng X (2013) Spectrum prediction in cognitive radio networks. IEEE Trans Wirel Commun 20:90–96 7. Chen Z, Qiu RC (2010) Prediction of channel state for cognitive radio using higher-order hidden Markov model. In: Proceedings of the IEEE SoutheastCon conference, pp 276–282 8. Akbar I, Tranter WH (2007) Dynamic spectrum allocation in cognitive radio using hidden Markov models: Poisson distributed case. In: Proceedings of the IEEE SoutheastCon conference, pp 196–201 9. Azmat F, Chen Y (2016) Predictive modelling of RF energy for wireless powered communications. IEEE Commun Lett 20(1):173 10. Salgado C, López H, Rodriguez-Colina E (2015) Multivariable algorithm for dynamic channel selection in cognitive radio networks. Springer 11. Lahby M, Adib A (2013) Network selection mechanism by using M-AHP/GRA for heterogeneous networks. In: Proceedings of 6th IEEE joint IFIP wireless mobile networking conference, pp 1–6 12. Bhowmick A, Roy SD, Kundu S (2015) Performance of secondary user with combined RF and non-RF based energy-harvesting in cognitive radio network. In: 2015 IEEE international conference on advanced networks and telecommunications systems (ANTS)

Design and Development of a Broadband Planar Dipole Antenna S. D. Ahirwar1(&), D. Ramakrishna2, and V. M. Pandharipande2 1

Defence Electronics Research Laboratory (DLRL), Hyderabad 500005, India [email protected] 2 Department of ECE, University College of Engineering (A), Osmania University, Hyderabad 500007, India

Abstract. In this paper, design and development of a broadband planar dipole antenna is presented. Antenna is investigated by simulation, in frequency range of 100 MHz to 2200 MHz for its electrical characteristics. The antenna is electrically small at lower frequency band. A broadband impedance matching network is designed to match the antenna at lower frequencies. A comparative study for its impedance characteristics is carried out for the antenna without balun, with balun and with balun and impedance matching network. Antenna is omni directional in azimuth plane with sufficient elevation coverage. All the measured results are presented. Keywords: Broadband


Dipole
Matching network
Planar
Printed

1 Introduction Modern communication systems are being explored for diverse applications like telemetry, data, voice, video, multimedia, spectrum monitoring, law enforcement, direction finding etc. This requires broadband antennas covering wide frequency bandwidth and omni azimuthal antenna patterns for 360° coverage. For omni directional coverage, wire antennas in dipole or monopole form are the best suitable elements. These are basically resonant structures and offer bandwidth of the order of 2% to 5% [1]. Biconical and its variants are also omni directional antennas but they require full size (0.5k for bicone, 0.25k for monocone) in terms of wavelength of operating frequency [2]. Mounting and handling of such large structures is very difficult. At 100 MHz the required length and diameter for a biconical antenna will be of the order of 1500 mm and 750 mm respectively. Moreover pattern bandwidth is also limited for these antennas, which is less than 6:1. Design and analysis of different types of dipole and printed dipole antennas with integrated balun in various form factors is available in literature [3–9]. Antenna size in these designs is reduced by modification in radiating structure in this approach, limited size reduction is possible. A printed dipole antenna can also be considered as the planar form of biconical antenna. This type of antenna has the advantage of lesser weight and volume compared to conventional biconical antenna. The broadband operation in these type of antennas can be achieved by choosing lesser k/D ratio of radiating element. But there is not much advantage of this antenna element fatness on antenna size reduction as it requires the antenna length © Springer Nature Switzerland AG 2020 S. C. Satapathy et al. (Eds.): ICETE 2019, LAIS 4, pp. 185–193, 2020. https://doi.org/10.1007/978-3-030-24318-0_22

186

S. D. Ahirwar et al.

approximately 0.5k at lower frequency of operation. This 0.5k length requirement can be relaxed in case of electrically small antennas. Impedance of such type of antennas is highly reactive with lesser resistive component creating difficulty in impedance matching of the antenna. To realize compact and electrically small broadband wire antennas, a suitable impedance matching network is required which can match frequency dependent antenna impedance to 50 X system. Impedance matching could be accomplished either by loading the antenna along its length or by using tuning or matching networks employing commercially off-the-shelf L and C components [10]. However, the frequency response of these components is limited and matching network operating over multioctave bandwidth cannot be realized. A novel impedance matching network using custom built toroidal inductors has been designed and developed for impedance matching of the antenna over the required frequency bandwidth [11].

2 Antenna Design and Realization A planar dipole antenna was modeled in a FDTD based EM simulation tool. A dielectric material FR4 of 1.6 mm thickness is used as a support for printed radiating elements of dipole antenna. A triangular taper in microstrip line is used to provide the transition between antenna feed points of elements and coaxial feed cable. This taper provides the balun action and impedance matching for the antenna. The simulation model of the antenna with its dimensional details is shown in Fig. 1. 160

100

120

12

Fig. 1. Antenna simulation model

Design and Development of a Broadband Planar Dipole Antenna

187

Considering the size restrictions the antenna dimensions are restricted to 0.05k 0.03k at lowest frequency of operation (100 MHz). The length of triangular taper is taken as 0.04k. The larger widths of ground plane and strip conductor of microstrip line are taken as 12 mm and 3 mm respectively. Widths of these conductors at antenna feed point are reduced to 1 mm. The proposed antenna was realized on FR4 substrate by photolithography technology and fed by SMA (F) connector. Each of the two radiating elements are printed on opposite sides of the substrate. This facilitates to make the ground plane and conductor strip of microstrip line as an integral part of radiating elements. An impedance matching network is integrated at the feed point of the antenna to match the frequency dependent antenna impedance to 50 X system. Photograph of realized antenna is shown in Fig. 2.

Fig. 2. Photograph of proposed antenna

3 Results and Discussion Antenna is simulated for its impedance and radiation characteristics. The simulated VSWR is validated by measuring it using M/s Agilent E5071C Network Analyzer. The overlay of simulated and measured VSWR is shown in Fig. 3. As expected, a good impedance matching is observed in the frequency range of 650–2200 MHz. A step by step procedure is followed for impedance evaluation of the realized antenna. Return loss of the antenna is measured without balun; with balun and with balun and impedance matching network. Impedance profile of the antenna with balun is shown in Fig. 4. As can be seen from the this plot the impedance of the antenna is having very less resistive part with more reactance causing mismatch and high VSWR below 650 MHz. To match the antenna an impedance matching network is integrated in shunt with antenna at its feed point. The impedance profile of this network is shown

188

S. D. Ahirwar et al.

in Fig. 5. This impedance profile provides the required impedance to match the antenna impedance to 50 X system. The measured return loss comparison for different stages of the antenna is shown in Fig. 6. Measured return loss for the antenna without balun is more than 10 dB above 1 GHz. With balun the lower frequency of operation shifted to 650 MHz. Finally, with impedance matching network a good impedance match is achieved over the band covering from 100 MHz to 2200 MHz.

Fig. 3. Overlay of simulated and measured VSWR

Fig. 4. Impedance profile of the antenna with balun

Design and Development of a Broadband Planar Dipole Antenna

189

Fig. 5. Impedance profile of impedance matching network

Fig. 6. Comparison of return loss of antenna at different stages

The proposed antenna was evaluated for its radiation characteristics in outdoor test range and in anechoic chamber. For frequency band 100–1000 MHz outdoor antenna test range is used and for frequency above 1000 MHz antenna is evaluated in an anechoic chamber. Measured E-plane radiation patterns are shown in Fig. 7(a, b) over the frequency band. As evident from the measured results a sufficient elevation

190

S. D. Ahirwar et al.

coverage (>30°) is achieved. The azimuth plane radiation patterns are shown in Fig. 8 (a, b). The antenna exhibits the omni directional radiation characteristics with maximum omni deviation of ±3 dB, as can be seen from measured results. Measured antenna gain plot of the antenna is shown in Fig. 9. As evident from this plot the antenna gain varies from −15 to +4 dBi.

105

90

120

165

-30

195 210 255

240

225

(b)

270

285

1.2 G Hz

300

315

330

345

1.6 G Hz

0

180

-40

15

2.0 G Hz

30

150

-20

45

135

-10

60

dB

75

2.2 G Hz

(a) 100-1000MHz

1.2-2.2GHz

Fig. 7. (a, b) Measured elevation plane radiation patterns of proposed antenna

Design and Development of a Broadband Planar Dipole Antenna

(b) 1.2 G Hz

100-1000MHz 1.6 G Hz 345

2.0 G Hz 0

2.2 G Hz

15

330

30

315

45

300

60

285

75

270

-40

-30

-20

-10

dB

255

90

105

240

120

225

135 210

150 195

(a)

180

165

1.2-2.2GHz

Fig. 8. (a, b) Measured azimuth plane radiation patterns of proposed antenna

191

192

S. D. Ahirwar et al.

Fig. 9. Measured gain plot of the proposed antenna

4 Conclusion A broadband planar dipole antenna covering 100–2200 MHz band has been designed and developed. Antenna is electrically small, broadband and omni directional with sufficient elevation coverage. The impedance bandwidth of the antenna is first predicted with simulation software. A step by step study is carried out for impedance matching of the antenna experimentally. Simulated and measured results are presented. The antenna due to its wide bandwidth, compactness and light weight, has applications in airborne platform like UAV (Unmanned Aerial Vehicle), Aerostat and compact land mobile communication systems. This antenna serves as good candidates for direction finding, radio surveillance and communication applications. Acknowledgments. Authors would like to thank Dr Anil Kumar Singh, OS & Director, DLRL and Dr M. Chakravarthy, Sc-G & Group Director, DLRL for providing constant encouragement and support to carry out this research work.

References 1. Kraus JD, Marhefka RJ (2003) Antennas for all applications, 3rd edn. Tata McGRAWHILL, pp 181–183 2. Kraus JD, Marhefka RJ (2003) Antennas for all applications, 3rd edn. Tata McGRAWHILL, pp 380–385 3. Edward D, Rees D (1987) A broadband printed dipole with integrated balun. Microwave J, pp. 339–344 4. Chang K, Kim H, Hwang KS, Sim SH, Yoon SJ, Yoon YJ (2003) A wideband dual frequency printed dipole antenna using a parasitic element. In: 2003 IEEE Topical Conference on Wireless Communication Technology, pp 346–347

Design and Development of a Broadband Planar Dipole Antenna

193

5. Jaisson D (2006) Fast design of a printed dipole antenna with an integrated balun, IEE Proc.Microw. Antennas Propag., Vol. 153, No. 4, August 2006 6. Su C-M, Chen H-T, Wong K-L (2002) Printed dual-band dipole antenna with U slotted arms for 2.4/5.2 GHz WLAN operation. Electron. Lett. 38(22), 1308–1309 7. Chen ZN, Liu D, Nakano H, Qing X, Zwick T (2016) Handbook of antenna technologies. Springer Nature 8. Ahirwar SD, Sairam C, Khumanthem T, Kumar A (2008) An improved gain, reduced size, broadband helical elements dipole antenna. In: International conference on recent advances in microwave theory and applications, 21–24, Jaipur, India, pp 224–226 9. Ahirwar SD, Sairam C, Singh S, Khumanthem T (2011) Broadband dual linear antenna with omni directional coverage. In: IEEE applied electromagnetics conference (AEMC) & IEEE Indian antenna week, Kolkata, India, 18–22 December 10. Ludwig R, Bretchko P (2001) RF circuit design theory and applications. Pearson Education Asia publisher, London 11. “Dong” DeMaw MF FerroMagnetic core design and application handbook, pp 140–141

Actuation System Simulation and Validation in Hardware in Loop Simulation (HILS) for an Aerospace Vehicle M. V. K. S. Prasad1(&), Sreehari Rao Patri2, and Jagannath Nayak3 1

Research Centre Imarat (RCI), DRDO, Hyderabad 500069, Telangana, India [email protected] 2 National Institute of Technology, Warangal 506004, Telangana, India [email protected] 3 Centre for High Energy Systems and Sciences, DRDO, Hyderabad 500069, Telangana, India [email protected]

Abstract. An Aerospace vehicle is developed as a result of integration of many subsystems with each subsystem marked to perform a particular function. The heart of the Aerospace vehicle is the control and guidance computer known as On-board computer. The rate gyroscopes which sense the aerospace vehicle rates and the accelerometers which sense accelerations are embedded in sensor system package. The Actuation system of the aerospace vehicle helps in steering the vehicle to the required direction. The Aerospace vehicle’s guidance algorithm calculates the correction required to move the vehicle to the intended trajectory. This error is converted into control deflection commands by the Autopilot. These deflection commands are given to the hardware actuation system so that the vehicle course is changed as per design. In this paper simulation of actuator model and Hardware-In-Loop-Simulation (HILS) of hardware actuation system is presented. Keywords: Aerospace vehicle
On-board computer Autopilot
Hardware-In-Loop-Simulation


Actuation system


1 Introduction HILS is carried out to validate flight hardware and mission software of various Aerospace vehicle systems [1]. Before reaching the stage of HILS, there are so many steps. First a Non Real Time model (NRT) is developed [2] which is an integration of aerospace vehicle model, control, guidance and navigation. This NRT model is developed in a single computer. Execution of NRT model is carried out by providing kinematic data, aerodynamic data and other mode shapes data of aerospace vehicle. The results of NRT model are validated against the expected behaviour of the Aerospace vehicle. Each model in the NRT program is developed separately and ported to one computer system. The Aerospace vehicle (known as plant) model in one computer, the control in one computer, the guidance in one computer and the navigation in © Springer Nature Switzerland AG 2020 S. C. Satapathy et al. (Eds.): ICETE 2019, LAIS 4, pp. 194–200, 2020. https://doi.org/10.1007/978-3-030-24318-0_23

Actuation System Simulation and Validation in Hardware in Loop Simulation (HILS)

195

another computer. Subsequently the NRT model is converted into Real Time (RT) model by introducing the Real Time execution using RT Linux [11] for achieving required periodicity. The periodicity is sometimes as high as 500 µs. So the Actuator (which is considered as second order) model is developed and tested. The model development is discussed in [II]. The mythology of porting each model of NRT into one computer system and bringing in Real Time with mission interfaces is called PC-IN-LOOP simulation [4]. Subsequently the actual Aerospace vehicle hardware is introduced and validated along with mission software which is widely known as Hardware-in-loop simulation.

2 Model Development As explained in [I] each aerospace vehicle subsystem is validated in HILS thoroughly. In order to carry out validation meticulously a pre-defined scheme is followed. First of all a mathematical model representing the Hardware Actuation system is developed (As given in Eq. 1). The mathematical model is an approximation of Real Hardware system. The parameters which define this actuator model are frequency of operation (xn = 2.0 * pi * fn) of the actuator and the damping (n) applied to the actuator. fn is the frequency of the operation of the actuator and n is the damping factor which indicates sluggish and faster response of the actuation system. In this paper an electro mechanical actuation system [8] of a particular class is considered. For this actuation system fn = 18 Hz and n = 0.7. This actuation system model is developed in C Language in Linux operating system [7]. The actuator model considered is second order [9] considering the fact that terms higher than second order contribute very negligible effect on the response [12]. The second order model [6] of the actuation system is represented mathematically as output x2n ¼ 2 input s þ 2nxn s þ x2n

ð1Þ

This model is executed for different values of n for checking the system response. A third order model is considered with a dominant pole approximation as given in Eq. 2 below. output 5 ¼ input ðs + 5Þðs2 þ s + 1Þ

ð2Þ

Here a second order under damped system (n = 0.5) with two poles at −0.5 + 0.866j and −0.5 − 0.866i is considered. This is made third order by adding a pole at −5. This pole is 10 times away from the two second order system dominant poles. The response of the second order system and third order system as given in Eq. 2 is seen with unit step input. The time response (output) is shown in Fig. 1. From the figure it is clear that the contribution of third pole to the overall system response is

196

M. V. K. S. Prasad et al.

negligible. However a third pole nearer to the two dominant poles will cause some difference in the response when dominant pole theory is not considered.

Fig. 1. Plot of second and third order system response for unit step input

3 Actuator Model Implementation in Aerospace Vehicle Simulation As explained in [II] Actuator model is developed as second order model as shown in Fig. 2. The aerospace vehicle model comprises of three translational accelerations (ax, ay, az) and three rotational accelerations (pdot, qdot, rdot) [3]. These parameters are calculated based on vehicle thrust, drag, forces acting on the vehicle body. The rotational accelerations are converted into pitch, yaw, roll (q, r, p) rates and are given to the aerospace vehicle mission computer (comprising of control, guidance and navigation). The guidance in the mission computer calculates the error between current position and expected position of the aerospace vehicle from time to time. This error is input to the control module of the aerospace vehicle. The control module calculates the required rates in order to nullify this error. Then the rates required (commanded rates) are calculated as the difference between demanded rates and rates sensed from the gyroscopes. The autopilot calculates the deflection commands to the actuators in order to generate the commanded rates and thus nullifying the error. These deflection commands are taken into the plant model and given to actuator model to simulate actuator lag and nonlinearities. The output deflections of actuator model are used in the plant rates and accelerations to calculate rates and accelerations for next iteration. In this manner an actuation system is mathematically modelled to carry out aerospace vehicle simulation.

Actuation System Simulation and Validation in Hardware in Loop Simulation (HILS)

197

Fig. 2. Block diagram of simulation setup with actuator model

4 Hardware Actuation System Simulation The Actuator model is validated for aerospace vehicle simulation as explained in [III]. After analysis of simulation results, the actuator model is replaced with Hardware actuation system as shown in Fig. 3. Here an electro mechanical actuation system [10] with four actuators is connected in the setup.

Fig. 3. Block diagram of HILS setup with hardware actuation system

In this scenario the deflection commands generated by aerospace vehicle control module excite the Hardware actuation system. The feedbacks of the hardware actuators are given to aerospace vehicle plant model. In this way the vehicle rates and accelerations are calculated for the next iteration. The feedbacks of the hardware actuation system provide exact lag, dead band, backlash and nonlinearities of actuation system. This type of Hardware-In-Loop-simulation of an aerospace vehicle is nearer to Real mission compared to simulation with actuator model. The control, guidance [5] designer should use these HILS results for finalising their designs for the aerospace vehicle mission. The exact behaviour of the aerospace vehicle with actuation system is evident by carrying out HILS with Hardware Actuation system. The gain and phase margins can

198

M. V. K. S. Prasad et al.

be exactly calculated only by carrying out HILS rather than mathematical models in loop. This helps in retuning control and guidance to avoid growing or sustaining oscillatory behaviour of the system. The stability margins of the system with different modes getting excited can also be calculated with the help of HILS results.

5 Validation of Results The four deflection commands for a typical Aerospace vehicle during hardware Actuator-In-Loop HILS are shown in Fig. 4. The deflection commands generated by the control module of mission computer are shown in blue colour whereas the feedbacks from the hardware actuation system are shown in red colour. For an actuator model it is observed that the model input and output match except for lag and

Fig. 4. Plot of actuator commands and feedbacks

Fig. 5. Time expanded plot of actuator commands and feedbacks

Actuation System Simulation and Validation in Hardware in Loop Simulation (HILS)

199

nonlinearities created by second order model. But in the case of hardware actuation system along with lag and nonlinearities, the effect of dead band and backlash are also considered inherently. Due to the mechanical motion and dead band effect the difference in deflections amplitude is also visible in Fig. 5 where the deflections are expanded in a limited time region. This behaviour is to be considered during design of control for any aerospace vehicle.

6 Conclusion From the results which are showing d command and d feedback with Hardware it is very clear that the Hardware behavior is not the exact replica of the model. The backlash effect, dead band effect and other nonlinearities are the factors which contribute to this difference. So while going for any live aerospace vehicle mission HILS is to be carried out mandatorily with Hardware actuation system while validating the entire vehicle. This exercise clearly brings out the margins (gain and phase) available which dictate vehicle performance during mission. The control and guidance design is to be freezed only after thorough analysis of HILS results. Acknowledgements. Authors express their gratitude to Director RCI for giving excellent support and ideas during the course of this work. Authors are thankful to Mrs. Amrita Singh, Scientist Directorate of Hardware In Loop Simulation (DHILS) for the excellent support in preparing this paper. Authors are thankful to Programme Area Defence members and employees of DHILS for their valuable guidance and support.

References 1. Srinivasa Rao B, Satnami S, Prasad MVKS, Sobhan Kumar L (2004) Hardware in loop Simulation for Exo-atmospheric interceptor Missiles, ICAA (International Conference on Advanced Avionics)-2014, Hyderabad 2. Gangadhar M, Singh A, Prasad MVKS, Sobhan Kumar L: Real time simulation system for aerospace interceptors. International conference on computing and communication technologies (ICCCT-2014), Hyderabad 3. Gangadhar M, Singh A, Prasad MVKS (2014) Distributed real time modelling of an aerospace vehicle. In: International conference on computing and communication technologies (ICCCT-2014), Hyderabad 4. Rao BS, Satnami S, Prasad MVKS (2016) Modeling and Simulation of IIR Seeker for an Aerospace Vehicle, INDICON-2016, Bengaluru 5. Gangadhar M, Singh A, Prasad MVKS (2017) Hard Real Time Distributed Aerospace System, Jointly published by Dr. APJ Abdul Kalam Missile Complex, DRDO and Institute of Defence Scientists & Technologists in 2017, Bengaluru 6. Prasad MVKS, Gangadhar M, Singh A (2018) Rate gyroscope sensor simulation and validation in Hardware in Loop Simulation (HILS) for an Aerospace vehicle, ICITE 2018, Osmania University, Hyderabad 7. Krishna CM, Shin KG (1997) Real Time Systems

200

M. V. K. S. Prasad et al.

8. Chun T-Y, Hur G-Y, Chol K-J, Woo H-W, Kang D-S, Kim J-C (2007) Design of the aeroload simulator for the test of a small sized electro-mechanical actuator. In: 2007 international conference on control, automation and systems, South Korea 9. Chin H (2009) Feedback Control System Design 10. Schneiders MGE, Makarovic J, van de Molengraft MJG, Steinbuch M (2005) Design considerations for electromechanical actuation in Precision motion systems 11. (2001) The RTLinux® embedded enterprise versus the Linux“RT” patches: FSMLabs White Paper 12. Roland S (2001) Burns: Advanced Control Engineering

A Real Time Low Cost Water Quality Progress Recording System Using Arduino Uno Board Mohammad Mohiddin(&), Kedarnath Bodapally, Sravani Siramdas, Shainaz, and Sai Kumar Sriramula Department of Electronics and Communication Engineering, Guru Nanak Institute of Technology, Ibrahimpatnam, Ranga Reddy 501506, Telangana, India [email protected], [email protected], [email protected], [email protected], [email protected]

Abstract. There are millions of people who die because of contaminated water. There is a necessity for development of Water quality monitoring system. This system states whether the water is apparent for drinking. Hence, we propound a monitoring system equipped with PH sensor, the Turbidity Sensor and water temperature sensor with Arduino Uno as main board. The physical and chemical properties like pH level, temperature, turbidity, TDS are being monitored using different respective sensors. The values measured by the sensors are being processed by the core controller. The quality of water is monitored continuously and the data is sent to the excel sheet. This system ensures whether the water stored in reservoirs after purification of water is fit for drinking or not. It displays a message on the LCD screen about the portability of water. Our project focus at providing hygiene drinking water at low cost. Keywords: pH sensor TDS. Arduino model


Turbidity sensor
Temperature sensor


1 Introduction Water quality monitoring is essential to keep the people healthy and sustainable. At the present times, various challenges are being faced in real time due to global warming and growing population. Therefore, it is important to develop better methods to monitor the real time water quality parameters [1]. In the world 20% of people does not have safe drinking water. The reasons for this are no proper monitoring of the water and no proper purification of water is done during rainy season because water is usually more turbid in rainy season. The most commonly monitored parameters of water quality are temperature, pH, turbidity, conductivity. Thus, the water quality monitoring can be achieved by collecting samples of required parameters manually and sending them to laboratory for detecting and analyzing. The general methods for detecting and

© Springer Nature Switzerland AG 2020 S. C. Satapathy et al. (Eds.): ICETE 2019, LAIS 4, pp. 201–209, 2020. https://doi.org/10.1007/978-3-030-24318-0_24

202

M. Mohiddin et al.

analysing are time taking processes and it also has certain limitations. In process to overcome this difficulty sensors are to be developed. Sensor is an ideal detecting device which converts power information into electrical signals [2]. A good quality water is an important factor to prevent humans from water-borne diseases and helps in improving quality of life [3]. 1.1

Literature Survey

The station which measures the quality of water is in the metropolis USA. This project measures the pH, Turbidity and Temperature using respective sensors and microcontoller as arduino uno. These factors are monitored because turbidity give the suspended solid particles, pH states whether water is acidic or basic temperature is monitored because these parameters may change with temperature [4].

2 Design and Experımental Detaıls 2.1

Architecture

The block diagram is illustrated in Fig. 1 which gives monitoring of water and it also includes about the sensors used in the project and their connections. In this system the properties of water that are being supervised is Temperature, Tds, Ph value and turbidity [5, 6]. The readings from all the sensors are received by arduino uno through serial communication and thus these values are sent to excel sheet using plx-daq software tool. The information about the portability of water for drinking purpose is displayed on LCD screen.

Fig. 1. Block diagram of a water quality monitoring system using Arduino.

A Real Time Low Cost Water Quality Progress Recording System

2.2

203

Arduino UNO Microcontroller Board

Arduino UNO microcontroller board as a platform for both hardware and software. The Arduino UNO consists of ATmega328p microcontrollers. The arduino hardware program is written in a wired language (syntax and library), such as C++ with minimal modification and a combined processing environment. It allows communication between computers through programming. It receives the input signal from the sensor, and then generates the output voltage displayed by the numbers displayed by the digital display [7, 8]. 2.3

Temperature Sensor

Figure 2 illustrates the circuit connections of Ds18b20. It has 3 (three) pins. The Vcc, the Data pin (also called DQ line), the Ground. The Vcc is given to 5v in Arduino and the ground pin same ground pin of Arduino. Since the sensor used is the digital one therefore the data pin is connected to any one of the digital pins of Arduino (here connected to 2nd pin). Specifications: Range: −55 to 125° centigrade. +/−0.5-degree accuracy from −10 to 85° centigrade and converts heat to voltage and gives values in degree centigrade.

Fig. 2. Circuit diagram of temperature sensor with Arduino Uno board

2.4

PH Sensor

PH sensor is a combination of ph probe and module. Figure 3 illustrates the connections of ph sensor with Arduino and Ph probe is connected to the module. The ph module also consist of a 9 V DC supply and the output pins are to connected to Arduino pins. One output pin is given connection to the analog pins of Arduino ie A0 and other to the gnd pin of Arduino.

204

M. Mohiddin et al.

Specifications: The supply voltage is 9v(DC) and working temperature ranges from 10 to 50 °C. Converts H+/OH- concentration between two solutions to electrical signals and these signals are converted to pH values by module.

Fig. 3. Circuit diagram of PH sensor using Arduino Uno board

2.5

Turbidity Sensor

Turbidity sensor can made using an IR sensor. A turbidity sensor has 3(three) pins viz: Vcc pin, Ground pin, Data pin. The 5v pin of Arduino is given to the vcc pin and the Ground pin is same in Arduino and Data pin to the analog pins of Arduino as shown in Fig. 4. The turbidity units are given as NTU which is abbreviated as Nephelometric Turbidity Units. Specifications: supply voltage: 3.3 V to 5 V. Output values between 0 to 500. Conversions: v = the value of sensor * (5.0/1023.0); turbidity = v * 100/4.

Fig. 4. Circuit diagram of turbidity sensor with Arduino Uno board.

A Real Time Low Cost Water Quality Progress Recording System

2.6

205

TDS Sensor

Figure 5 illustrates Tds (total dissolved salts) sensor. It can be making use of conductivity sensor. It has a module and two large exposed pads(probes). The Probes are connected to the module and the module has 3(three) pins that goes to the Arduino i.e. Vcc, Ground and the Data pin. The Vcc is connected to the 5v pin of Arduino and the Ground to the gnd pin and the Data pin is given to any of the analog pins of Arduino (here connected to the pin no A2 of Arduino). Specifications: supply voltage: 3.3 V to 5 V, Operation Range: −40 to 85 °C, Accuracy: +/1 °C.

Fig. 5. Circuit diagram of TDS sensor with Arduino uno board.

2.7

Circuit Design and Flow Chart

The Fig. 6 shows the complete circuit connections of different sensors with Arduino uno. The temperature sensor, ph sensor, turbidity sensor and tds sensor data pins are connected to the pin numbers d2, A0, A1 and A2 respectively of Arduino uno. The Vcc pins of all the sensors are given to the 5v pin and all the ground pins are given to the gnd pin of Arduino uno. The pins of lcd are connected such as Enable to pin no 3, rs to pin no 4, d4 to pin no 8, d5 to pin no 9, d6 to pin no 10 and d7 to pin no 11. Figure 7 shows the hardware implementation of the monitoring of water quality at the low cost using Arduino Uno and sensors with a LCD display. The display screen used to display the quality of water. The flow chart of the system in Fig. 9 illustrates data from the sensors provided to Arduino UNO board. Process the data of PH, TEMPERATURE, TDS and TURBIDITY using the data sheets with different slopes, we acquire the data of PH, TEMPERATURE, CLOUDLINESS and CONDUCTIVITY of water. This data now connected to laptop using PLX-DAQ. Since the data monitored is obtained in excel sheet which can be further used. Arduino UNO connected to the sensors and throug the PC the data is directly sent into Excel (Fig. 8).

206

M. Mohiddin et al.

Fig. 6. Complete circuit diagram of the real time water quality monitoring system using Arduino.

Fig. 7. The experimental set up for water quality monitoring.

A Real Time Low Cost Water Quality Progress Recording System

207

Fig. 8. Flow Chart for obtaining data from the sensors provided to Arduino UNO board

3 Results and Discussion Figure 9 shows the graphical representation of data gathered through one hour monitoring of different solutions namely Acid, Distilled Water and Base. The variation of values for different solutions is tabulated in Table 1. Since the ph value of an acid solution lies between 0 to 7 and base from 7 to 14. From the Table 1 the lemon water got the ph around 4.8 which states it is acid in nature and the sodium hydroxide got the ph value around 8.4 which states that it is base. The ph value obtained for distilled water is 7.8. We know that conductivity is more in acid than in base and distilled water. From Table 1 lesser the value obtained the more is conductivity. Table 2 shows the permissible levels for distilled water. From Table 1 the ph of distilled water is 7.8, turbidity obtained is 87 and tds obtained is 0.7. Since this project is meant for qualitative analysis but not for quantitative analysis the units of turbidity and tds can be converted to their respective units.

208

M. Mohiddin et al.

Fig. 9. Results of parameters monitored for acid, distilled wate rand base. Table 1. Compares the values obtained for acid, base and distilled water Parameter Temperature pH value Turbidity TDS

Acid 28–27 4.8 68 0.9

Basic 29 8.4 27 2

Distilled water 28–34 7.8 87 0.7

Table 2. Permissible ranges of distilled water Ph value Turbidity TDS 7 to 8 30 dBiC and Axial Ratio 30 dB, the ARs measured was 3.5, 1.0, 1.45 dB at 1.57, 1.59, 1.61 GHz frequencies respectively as shown in Fig. 5(b–d). Since no provision to rotate the feed positioner continually, the axial ration measurement was done as explained. SNR was observed as 40 dB in both GPS and GLONASS Bands during satellite link check. The realized active antenna was qualified for all environmental conditions required for airborne applications.

Fig. 5. (a) Fabricated active antenna (b) Measured axial ratio in 1.57 GHz (c) Measured axial ratio in 1.59 GHz (d) Measured axial ratio in 1.61 GHz

264

K. Sambasiva Rao et al.

Fig. 5. (continued)

3 Conclusion A miniature CP active antenna has been developed and demonstrated its performance. The performance of the antenna was achieved with 30 dBiC Gain and >40 dB SNR throughout the band of operation. The antenna was successfully used for GPS+GLONASS application in air bone platform. Acknowledgments. The authors would like to acknowledge the encouragement of Shri B.H.V.S NarayanaMurthy Director, RCI Hyderabad for granting permission to publish this work.

An Airborne Miniature Circularly Polarized Active Antenna

265

References 1. Garg R, Bhartia P, Bahl I, Ittipiboon A (2001) Microstrip antenna design handbook. Artech House, Norwood 2. Hong J-S, Lancaster MJ (2001) Microstrip filters for RF/Microwave applications. Wiley, Hoboken 3. Jahagirdar DR (2012) A GPS+GLONASS active antenna for extremely high temperature aerospace applications. In: Antenna and propagation society 2012. IEEE 4. Nasimuddin, Karim MF, Alphones A (2016) A low profile dual-band circularly polarized GPS antenna. In: 2016 Asia Pacific microwave conference 5. Rao BR, Kunysz W, Fante R, McDonald K (2013) GPS/GNSS Antennas. Artech House, Norwood 6. Bao XL, Ammann MJ (2008) Dual-frequency dual circularly polarized patch antenna with wide beam width. Electron Lett 44(21):1233–1234 7. Nasimuddin, Anjani Y, Alphones A (2015) Wide beam circularly polarized asymmetric micro strip antenna. IEEE Trans Antennas Propag 63(8):3764–3768 8. Bauer R, Schuss J (1987) Axial ratio of balanced and unbalanced fed circularly polarized patch radiator arrays. In: 1987 antennas and propagation society international symposium, vol 25. IEEE

Efficient Obstacle Detection and Guidance System for the Blind (Haptic Shoe) Rajani Akula(&), Bhagavatula Ramya Sai, Kokku Jaswitha, Molugu Sanjay Kumar, and Veeramreddy Yamini JNTUH College of Engineering, Hyderabad, India [email protected], [email protected], [email protected], [email protected], [email protected]

Abstract. Aiming for the safety of blind people who are facing many difficulties in today’s busy world, a design method for the detection of obstruction in the path of a blind person and conveying directions to him/her based on the location that has to be reached is proposed. The purpose of this paper is to detect obstructions using a distance sensor and a vibrator and to convey directions to the person using a voice recognition module. This sensor setup is placed in the shoe of the blind person such that it can detect any obstruction in his path and warn him using the vibrator when the distance is less than the safe distance. Keywords: Haptic shoe


Voice recognition module

1 Introduction Out of the 285 million visually impaired people in the world about 39 million are totally blind and close to 246 million having low or poor vision. The average age of around 65% of visually impaired people and about 82% of blind people is over 50 years. The purpose of this work is to ensure the safety of blind people by providing them with a wearable obstacle detection device which detects any obstructions in their path and warns them when the distance between the person and the obstruction is less than the safe distance. In most developing countries, the popular Guide Dog technique is ruled out as they are not permitted in public places. Comparably, electronic sensor canes are difficult to use in congested or packed traffic places. Canes though generally safe in free places might cause difficulty in terms of orientation in new places. Some modern systems provide audio as feedback but the audio could distract the person as visually impaired people depend on their sense of hearing to a large extent. To ensure the safety of a person in a non-obtrusive and non-distracting manner, using technology will be quite helpful. For this purpose, using shoes is a good idea as shoes are the most natural extension of a human body and they also point in the direction one intends to walk. In the past few years, researchers developed some technical systems to support people navigate through city streets. These systems generally stand on the wrist or are consolidated into glasses or are put on as a vest. However, recently some young scientists has come up with a navigational device for blind people which is built into a shoe namely Le Chal, which means” Take Me There” in Hindi [1]. It navigates the © Springer Nature Switzerland AG 2020 S. C. Satapathy et al. (Eds.): ICETE 2019, LAIS 4, pp. 266–271, 2020. https://doi.org/10.1007/978-3-030-24318-0_32

Efficient Obstacle Detection and Guidance System for the Blind

267

users to their intended geographical locations and also avoids obstacles by detecting them and generating a warning signal.

2 Methodology The major work is to provide information of obstacle detection which is achieved by using ultrasonic distance sensor which detects any obstacles in the Line of Sight. This sensor is placed in the front of the shoe and the remaining setup is placed at the back of the shoe such that the system does not disturb the blind person. This sensor generates a 8 cycle signal burst signal of the same duration as the duration for which the trig pin is set to high (1000 ms in this project) which travels at the speed of sound and bounces back to the sensor when it hits an obstacle. Hence this time gap can be used to derive the distance of the obstacle and warns the user when the distance is less then the safe distance i.e. 90 cm.

Speed of sound = 340 m/s = 3.4 cm/s Speed =distance/speed Hence distance = speed * duration = 0.034 cm/s * duration Additionally the voice recognition module is used to convey directions. Firstly the instructions are stored in the module one by one where each of the instructions to be given is repeated 5 times by the same speaker. The instructions given in this project were left, right and straight. Once the instructions are stored in the memory of the voice recognition module in hexadecimal format, each pin in the arduino is selected for each instruction and these pins are connected to coin vibration motors respectively. When the speaker voices the instruction, the voice signal is converted to hexadecimal format and is then compared with the original instruction hexadecimal form to identify the instruction. Then the corresponding arduino pin is set to high which in turn starts the vibration motor.

byte b= Serial.read(); if (b>=0x11 && b Vc1 then state of the Leg a is 1, If Vx < Vc1 or Vx > Vc2 then state of the Leg a is 0, If Vx < Vc2 then state of the Leg a is −1. The generated pulses with the above logic are used for switching leg ‘a’ of the NPC inverter. In a similar way switching logic for leg b and leg c are obtained by using above logic by taking two more sine waves Vy and Vz repectively, instead of Vx. where Vx, Vy and Vz are balanced sine waves.

Analysis of CMV Reduction Methods

5.2

377

Phase Opposite Disposition SPWM (POD SPWM)

In this Technique, it uses two carrier waves and one sine wave as shown in Fig. 7 but here two carrier waves are out of phase to each other. The switching logic per phase is same as explained in PD SPWM.

Fig. 6. Reference and carrier signals for PD

Fig. 7. Reference and carrier signals for POD

5.3

Analysis of CMV for PD and POD Methods

Figures 8 and 9 shows the Reference sine waves, carrier waves, switching states and CMV Vsn for PD and POD methods respectively, where peak value of sine wave is 0.88, instantaneous angle is 1000 and duration considered is one carrier cycle. In this paper Vdc for SPWM technique is taken as 677 V. It is observed that peak to peak voltage of CMV is Vdc/2, i.e. 338.5 V for PD method whereas it is Vdc/3 i.e. 225.6 V for POD method. Voltage change at each state for both the methods is Vdc/6 i.e. 112.8 V. Hence CMV is reduced for POD method compared to PD method.

378

R. Linga Swamy and R. Somanatham

Fig. 8. Sine waves, carrier waves, switching states and CMV Vsn for PD method

5.4

Fig. 9. Sine waves, carrier waves, switching states and CMV Vsn for POD method

Conventional Space Vector PWM (CSVPWM)

The space-vector diagram of a three-level inverter is shown in Fig. 4 [4]. A 3-level inverter consists of 27 switching states. In SVPWM, the reference vector is synthesized by using three nearest vectors. In Fig. 4 the reference vector is in Hexagon 1 and triangle 1, it is synthesized by using nearest vectors V1(1, −1, −1), V7(1, 0, 0), V7(0, −1, −1) and V13(1, 0, −1), thereby reducing the voltage deviation from reference vector. V7 is the middle vector and it is having two redundant states. Similarly all middle vectors will have two redundant states. In this technique, the three-level inverter space vector modulation analysis is done in terms of two-level space vector modulation. This is done by creating a new reference vector Vb which is obtained by subtracting VP (Pivot vector) from Vref (Eq. 4), where Vref is the reference space vector of three-level inverter, Vb is the reference space vector of imaginary two level inverter. In the Fig. 4, Pivot vector VP is V7. Vb \b ¼ Vref \a  VP

ð4Þ

Where a is the angle between reference vector Vref and pivot vector V7, b is the angle between new reference vector Vb and x-axis of Hexagon 1. The equations for dwell times for each vector in a triangle are given below: T1 ¼

Vb sinð60 bÞ Vb sin(bÞ Ts ; T2 ¼ Ts ; Tz ¼ Ts - (T1 + T2 Þ  0:5Vdc sinð60 Þ 0:5Vdc sin(60 Þ

Where Vdc is the input DC voltage, Ts is the sampling time.

ð5Þ

Analysis of CMV Reduction Methods

5.5

379

Partial CMV Reduction SVPWM (PRSVM)

The algorithm of this PWM is same as CSVPWM except that it avoids the redundant states. As seen from Fig. 4 that there are 27 possible switching states and there are 19 unique voltage vectors. As explained in Sect. 5.4, that there are 6 redundant vectors in middle vectors, they are (0, −1, −1), (1, 1, 0), (−1, 0, −1), (0, 1, 1), (−1, −1, 0), (1, 0, 1). There are also two redundant zero vectors (1, 1, 1) and (−1, −1, −1). All the redundant states are having the CMV |Vsn|  Vdc/6 and other vectors are having | Vsn|  Vdc/6. Hence in this PWM method, CMV reduction is done by using conventional space vector modulation algorithm but selecting only non-redundant voltage vectors. 5.6

Analysis of CMV for CSVPWM and PRSVM Methods

Figures 10 and 11 shows the switching states and CMV Vsn for CSVPWM and PRSVM methods respectively. The reference vector is in Hexagon 1 and triangle 1 as shown in Fig. 4. In this paper Vdc for SVM technique is taken as 586 V as explained in Sect. 6. It is observed that peak to peak voltage of CMV is Vdc/2, i.e. 293 V for CSVPWM method whereas it is Vdc/3 i.e. 195.3 V for PRSVM method. Hence CMV is reduced for CSVPWM compared to SPWM methods. CMV is lesser for PRSVM method compared to remaining three methods.

Fig. 10. switching states and CMV for CSVPWM

Fig. 11. switching states and CMV for PRSVM

380

R. Linga Swamy and R. Somanatham

6 Experimental Results The hardware analysis of common mode voltage reduction of 415 V, 5.5 kW, 1450RPM, Induction motor driven by 3-level NPC inverter based on PD SPWM, POD SPWM, Conventional SVM and partial CMV elimination SVM techniques is implemented. Figure 12 shows the Experimental setup. Figure 13 shows the Line voltage waveform for all the methods. Fig. 12. Experimental Setup Figure 14 shows the CMV for all the methods. All the PWM methods are implemented in DSP28335 [8]. The Switching frequency taken is 3000 Hz.

Fig. 13. Line voltage waveforms for (a) PD (b) POD (c) CSVPWM (d) PRSVM

For Experiment, to obtain rated voltage and rated frequency for modulation index ma = 1, the input dc voltage required for NPC inverter with Sinusoidal PWM is Vdc ¼ pffiffiffiffiffiffiffiffiffiffiffi pffiffiffiffiffiffiffiffiffiffiffi Vll1  2  ð2=3Þ ¼ 415  2  ð2=3Þ ¼ 677 V and with Space Vector PWM it is

Analysis of CMV Reduction Methods

381

Fig. 14. CMV waveforms for (a) PD (b) POD (c) CSVPWM (d) PRSVM

pffiffiffi pffiffiffi Vdc ¼ Vll1  2 ¼ 415  2 ¼ 586 V. Where Vll1 is fundamental Line voltage of the Induction motor. For Sinusoidal PWM the input DC Voltage required is 677 V for 415 V Induction motor whereas for SVM the input DC Voltage required is 586 V, which is 15% less. Hence SVM technique has the advantage of 15% more DC bus utilization.

Table 2. Results obtained from experiment Parameters Input dc CMV (RMS) Current THD (no load) Current THD (load)

PD 677 110 8.29 6.6

POD 677 80 10.5 9.6

CSVPWM 586 75 8.9 5.03

PRSVM 586 68 8.93 5.4

From Experimental results (Table 2), It can be observed that CMV is reduced for POD compared to PD and CMV is also reduced for PRSVM compared to all other methods. It can be also observed that Current THD is less for PD than POD and Current THD is nearer for CSVPWM and PRSVM.

7 Conclusion Analysis and Hardware Implementation of a three-level Neutral-Point Clamped Voltage Source Inverter using sinusoidal pulse width modulation techniques (Phase Disposition (PD), Phase Opposite Disposition (POD)) and Space vector modulation

382

R. Linga Swamy and R. Somanatham

techniques (conventional SVM, partial reduction CMV SVM) are done. The Experimentation was done on 5.5 kW three-phase induction motor drive. Voltage and Current waveforms are observed in Power Analyzer. Performance parameters in terms of Current THD’s and CMV are tabulated. The current THD and CMV obtained using experimentation results were compared at no-load and Half-load at rated voltage, which revealed that THD for PD SPWM technique is smaller than POD technique. Whereas CMV is reduced for POD technique compare to PD technique. Partial CMV reduction SVM is observed to have lesser CMV than all the methods. It was also observed that the THD of partial CMV reduction SVM scheme is nearer to CSVPWM scheme. With reduced CMV, Bearing currents are reduced and Bearing failures are reduced. Hence lifetime and performance of the motor will be improved.

References 1. Gupta AK, Khambadkone AM (2007) A space vector modulation scheme to reduce common mode voltage for cascaded multilevel inverters. IEEE Trans Power Electron 22(5):1672–1681 2. Loh PC, Holmes DG, Fukuta Y, Lipo TA (2003) Reduced common-mode modulation strategies for cascaded multilevel inverters. IEEE Trans Ind Appl 39(5):1386–1395 3. Zhang H, Von Jouanne A, Dai S, Wallace AK, Wang F (2000) Multilevel inverter modulation schemes to eliminate common-mode voltages. IEEE Trans Ind Appl 36(6):1645–1653 4. Lee YH, Kim RY, Hyun DS (1996) A novel PWM scheme for a three-level voltage source inverter with GTO thyristors. IEEE Trans Ind Appl 32(2):260–268 5. Das S, Narayanan G (2011) Space vector based analysis and comparision of sinusoidal pulsewidth Modulation Schemes for a Three-Level Inverter. In: Proceedings of the 5th national power electronics conference (2011) 6. Zhou J, Li Z (2009) Research on multi-carrier PWM modulation strategies of three-level inverter. In: Power and energy engineering conference, APPEEC 2009, Asia-Pacific 7. Kim H, Lee H, Sul S (2001) A new PWM strategy for common mode voltage reduction in neutral-point-clamped inverter-fed AC motor drives. IEEE Trans. Ind. Applicant. 37:1840– 1845 8. TMS320x2833x, 2823x Enhanced Pulse Width Modulator (ePWM) Module Reference Guide by Texas Instruments

Blaze Averter – A Cooking Sustenant for Visually Challenged People Sai Venkata Alekhya Koppula1(&), Suma Bindu Inumarthy2(&), Devi Radha Sri Krishnaveni Korla1(&), Pavani Mukkamala2(&), and Padma Vasavi Kalluru2 1

Department of Electrical and Electronics Engineering, Shri Vishnu Engineering College for Women, Bhimavaram, India [email protected], [email protected] 2 Department of Electronics and Communication Engineering, Shri Vishnu Engineering College for Women, Bhimavaram, India [email protected], [email protected], [email protected] Abstract. Simple daily activities like cooking, walking etc., are major problems for visually challenged people because of their problem with ‘vision’. Though challenged by vision, they would like to lead their life independently. However, tasks like cooking food on their own may be hazardous for them as touching hot utensils or touching a lit stove burner may burn their skin. This project ‘BLAZE AVERTER’ aims at helping the visually challenged people in cooking their food without burning their body. A proximity temperature sensor would be used to sense the temperature of the surrounding and a flame sensor would be used to detect any flame in the kitchen. A threshold is set which distinguishes high temperature and presence of flame. Any change in set threshold will be identified and transmitted to the speaker to alert the user and protects them from getting burns. Keywords: Flame sensor
Temperature sensor Cooking
Assistive technology


Visually challenged


1 Introduction Disability is a physical limitation in one’s own life but this disability won’t prevent them from their interested aspects like cooking. Elderly people gradually lose their vision due to ageing. are expert in cooking, They may be unaware of the hot utensils behind them while cooking. Not only elderly people, women in different age groups in India are more likely to be interested in cooking though they are visually challenged. So there is a strong necessity of a device BLAZE AVERTER in the cooking environment to detect hot utensils behind the user to avoid them from getting burnt, mainly for visually challenged. In India, out of 121 crores of total population, the male and female population are 51% and 49% respectively. Out of total population, 2.68 crore persons are disabled which is about 2.21% of total population. Among the disabled population, 56% (1.5 crores) are male and 44% (1.18 crores) are female. In the case of total population, 69% are from rural areas while the remaining 31% are residing in urban areas. Majority of the disabled population (69%) are living in rural areas (1.68 © Springer Nature Switzerland AG 2020 S. C. Satapathy et al. (Eds.): ICETE 2019, LAIS 4, pp. 383–389, 2020. https://doi.org/10.1007/978-3-030-24318-0_46

384

S. V. A. Koppula et al.

crore disabled persons in rural areas and 0.81 crores in urban areas). Out of disabled population, visually challenged women are about 23,94,403 lakhs. About 12 lakhs visually challenged women are in the age group of above 60 years which constituents about 50% of visually challenged women. In India, women are more likely to be interested in cooking. So, the products available in the market for visually challenged people to provide them ease of cooking is low. So, there is a need to increase the market size of products for visually challenged people. Many products are available in the market to provide ease of cooking for visually challenged people. But the main problem is that available products help them in handling the ingredients carefully but they can’t protect them from getting burnt while cooking. To avoid this problem, the available product in the market is PROTECTION GLOVE. This protection glove helps them only to avoid burns on their palms but not on their body. To overcome this problem our product BLAZE AVERTER helps them to avoid burns because of its feasible operation with temperature and flame sensors to alert the user through a speaker. The main prospective of blaze averter is to provide: ease of cooking, assurance of safety environment in the kitchen, to instill self confidence in them to achieve their dreams and goals and to create a self-sustainable cooking environment for visually challenged people. The rest of the paper is outlined as follows: Sect. 2 describes the architecture of Blaze Averter, Sect. 3 describes the working of the device Results are provided in the Sect. 4 and finally Sect. 5 concludes the paper.

2 Architecture of Blaze Averter The system level diagram of Blaze Averter is shown in Fig. 1. As shown in Fig. 1, two arms extend out of the base to place the flame sensors. A side arm is fixed to one of the extremes of the top surface of the base by means of a stepper motor. To make the device compact the stepper motor moves the side arm to the horizontal position when the device is idle. It is moved towards right when the device is under working condition. Temperature sensors are placed at the bottom of the side arm to detect the hot vessels kept on the kitchen base. An On/OFF button is placed on the left side and the speaker that announces the audio messages is also placed at the bottom of the Blaze Averter base.

Fig. 1. System level diagram of Blaze Averter

Blaze Averter – A Cooking Sustenant for Visually Challenged People

385

As shown in system level diagram, Blaze Averter consists of an array of temperature sensors and flame sensors which exactly coincides with the 4 burners of the stove. These sensors detect the temperature of the utensils placed in any of the 4 burners of the stove and also detects whether the flame is on or off. The extended arm consists of temperature sensors which detects the surrounding utensils temperature. When the flame is on or any of the utensil temperature exceeds the threshold, the speaker will give a voice message which is previously stored in SD Card through an SD Card Reader (Fig. 2).

Fig. 2. Block diagram of Blaze Averter

The Blaze Averter is built around a microcontroller for controlling the I/O peripherals. The array of temperature sensors takes the input from the hot vessels and the signals from temperature sensor are given to the analog ports of the microcontroller. The flame sensors detect the flame when the gas stove burner is turned ON and the signals from the flame sensors are also given to the microcontroller. A push button is used to control the operation of servo motor. Several audio messages which give the information about the burner status and presence of hot vessels are stored in SD card. The SD card us interfaced with the microcontroller by making use of Serial Peripheral Interface. The audio messages from the SD card are heard through the loud speaker which is amplified by making use of audio amplifier. A servo motor is controlled by the microcontroller to move the side arm of the Blaze averter in the horizontal and vertical directions. The details of each subsystem in the block diagram is given below: Arduino MEGA 2560: The Arduino Mega is a Microcontroller. It has 54 digital input/output pins, 16 analog inputs, 16 MHz crystal oscillator, a USB connection, a power jack and a reset button. Its main features include analog comparator, inbuilt RTC, advanced timer and JTAG support for programming, debugging and troubleshooting.

386

S. V. A. Koppula et al.

Temperature sensor: This is a pre-wired and waterproofed version of the DS18B20 sensor made with a PTFE (Poly tetrafluoroethylene) wire cable. It can be handled very easily to measure temperatures at long distances or in wet conditions. Usable with 3.0– 5.0 V systems. There is no effect of signal degradation as the sensor signal is digital. It can detect wide temperatures in the range of −55 °C–125 °C. It is a K- type Thermocouple used to detect the temperature of the hot utensils. Flame sensor: It has a built in potentiometer for sensitivity control and the working voltage of flame sensor is 3.3 V–5 V, Generally used for fire alarm purposes. It detects the presence of flame on the burner and passes the information to the micro controller. Servomotor: A servomotor is a rotating actuator. It has metal gears with an angle of 180 rotational degree. It serves the movement of side arms which consists of a temperature sensor to detect the presence of temperature of the hot utensils placed beside the stove to avoid burns for the user. SD CARD: SD Card is a nonvolatile memory card format used in portable devices. In this SD card, we store the voice messages of different burners which are related to temperature and presence on the flame on a particular burner. Audio Amplifier: It is a power amplifier used to amplify low power electronic signals into high power electronic signals. The voice messages which are stored in the SD card will get amplified so that the user will be able to hear the voice messages clearly. Speaker: It is a device that converts electrical impulses into sound. The user can be alerted according to speaker’s instructions.

3 Implementation The implementation of the Blaze Averter is divided into two sections: • Software Implementation • Hardware Implementation The software implementation details of Blaze Averter are shown in Fig. 3. All the GPIOs are initialized and the readings from all the sensors is set to zero. The microcontroller reads the data from the flame sensor array and temperature sensor array. If the value of flame sensor placed on the first burner of the stove exceeds the set threshold then the voice message “Alert-First burner is ON”. Similarly whenever, the threshold value for any of the flame sensors exceeds the set threshold the corresponding audio message is spelt out of the loud speaker. If the value of any of the temperature sensors exceeds the set threshold then the audio message “Be alert- Hot Vessel” comes out of the loud speaker (Table 1).

Blaze Averter – A Cooking Sustenant for Visually Challenged People

387

Fig. 3. Flow chart of Blaze Averter Table 1. Hardware details of Blaze Averter S. No 1 2 3 4 5 6

Name of the component Micro controller Thermocouple Flame sensor Audio amplifier Speaker Servomotor

7

SD card reader module

Specifications

Range

Quantity

ArduinoMega2560 DS18B20 Xcluma PAM8610 8 ohm 25 watts MG995

– −55 °C–125 °C 80 cm – – Metal gear type, rotational degree = 180 –

1 2 4 1 1 1

–

1

The circuit diagram for the hardware implementation of the Blaze Averter is shown in Fig. 4.

Fig. 4. Circuit diagram for hardware implementation

388

S. V. A. Koppula et al.

During the operation, if the set threshold value of temperature sensors are exceeded and the state of Flame sensor is detected and a voice message will come through a Speaker which was amplified with the help of Audio Amplifier. This whole process will be done with the help of a Microcontroller which was initially embedded by Software Programming. After the completion of the task performed by the user, the extended side arms of the device can be folded with the help of Servomotor by using Pushbutton.

4 Results The Prototype of Blaze Averter is shown in Fig. 5. For a four-burner stove, if burner1 is in ON condition, then the flame sensor corresponding to that burner detects the presence of the flame and alerts the user through the speaker in the form of voice messages as ‘burner1 is ON’. Similarily, for four-burner, if more than one burner is in ON condition, then it alerts the user through voice messages with corresponding burner number. If the temperature exceeds the set threshold (30 °C) on particular burner, then the user can be alerted through the voice messages with the particular burner number” burner1 is hot, Be cautious”. So, visually challenged people can cook on their own.

Fig. 5. Prototype of Blaze Averter

Flame sensor detects the presence of flame at an angle of 90° at a distance of 80 cm for the corresponding four burners of a four-burner stove. Similarly, temperature sensor detects the temperature of hot utensils at a distance of 20 cm for the four burners and alerts the user through voice messages. Working Video Link: https://youtu.be/YQACdJefKzs https://youtu.be/gCXK9cA4Iow

Blaze Averter – A Cooking Sustenant for Visually Challenged People

389

5 Conclusion “BLAZE AVERTER” will assists and assures safety for the visually challenged people while cooking. So, our aim to provide a self sustainable environment for the visually challenged people was achieved. Hence, visually challenged people can also make their cant’s into can’s and dreams into reality like Christine Ha, finalist of Masterchef competition.

References 1. https://www.arduino.cc/ 2. http://www.xcluma.com/flame-sensor-infrared-receiver 3. https://www.digikey.com/catalog/en/partgroup/temperature-sensor-waterproof-ds18b20/ 57171 4. https://www.easyandworkproject.xyz/2016/02/10-watts-8-ohms-class 5. https://www.sitepoint.com/how-many-users-need-accessible-websites/

Symmetrical and Asymmetrical Fault Response of DC Link in AC-DC Interface HVDC System Maanasa Devi Atluri(&), N. V. L. H. Madhuri Ramineedi, and Revathi Devarajula EEE Department, Vardhaman College of Engineering, Hyderabad, India [email protected], [email protected], [email protected]

Abstract. With the increase in population, consumption of electricity has been increased tremendously in the past few decades and requirement to meet this demand has become a concern of many countries. The best solution for transmitting power to remote areas is the use of HVDC link, which offers reliable and economical transmission of power for long distances and also for connecting multiple AC systems with DC link. In the modern world, protection of DC system is still a challenge as the protection units are still in the development stage. This paper focuses on the design of HVDC system [1] which connects two 100 kV AC systems and also the response of DC line for the various Symmetrical and Asymmetrical faults occurring on AC side of the system [3]. The help of software simulation tool PSCAD version 4.6.2 has been taken for this analysis and further studies. Keywords: HVDC modeling
100 kV 400 kms system DC response
Clearance time


AC faults


1 Introduction Bulk power transmission for long distances above 500 kms is complex and costly for AC lines and frequency control is a difficult task in this process. HVDC transmission is the solution associated to this problem. Connection of two or more AC systems even at different frequencies is possible in HVDC link. Also for the integration of renewable energy sources to AC grid is possible with HVDC system [1]. The protection system of the DC link is still in the developing stage and the proper analysis of faults should be done in order to get better idea about the behavior of the DC link to the changes occurring in ac system. The main motto of this paper is to assess the response of DC Link [2] and the receiving end for AC system, for the symmetric and asymmetric faults on sending end of AC system.

© Springer Nature Switzerland AG 2020 S. C. Satapathy et al. (Eds.): ICETE 2019, LAIS 4, pp. 390–397, 2020. https://doi.org/10.1007/978-3-030-24318-0_47

Symmetrical and Asymmetrical Fault Response of DC Link

391

2 HVDC Design Parameters While designing a HVDC system, certain parameters need to be taken into consideration. Those parameters shown in schematic Fig. 1 given below are modeled suitably using PSCAD simulation software and are described:

Fig. 1. Schematic of HVDC system

2.1

AC Source

AC systems of 50 Hz frequency are considered on sending and receiving ends which are of 100 kV, 150 MVA and 1.5 kA ratings. 2.2

Rectifier/Inverter Unit

Thyristor based rectifier and inverter units are designed with switching frequencies of 1 kHz. Triangular pulses are given on rectifier end whereas; rectangular pulses are preferred on inverter end. 2.3

Transformers

The grounded star delta transformer of 100 kV/500 kV step up on sending end and delta star transformer of 500 kV/100 kV step down on receiving end are used. 2.4

DC Capacitor

To make sure that steady state stability problems to not to arise, the DC capacitor is designed not to be too small or too large. A 50 µF capacitor bank can be installed on DC Link. Along with these, a cable length of 400 kms has taken into account for DC Link with an impedance equivalent to 50 kms on AC systems [4].

392

M. D. Atluri et al.

3 Fault Analysis and Results After simulating HVDC system with AC-DC Interface, the result of the normally working AC and DC currents in the system are shown in the Figs. 2 and 3 After which various faults are simulated on AC side of the system and the corresponding DC response is studied. The currents Ia, Ib and Ic are of peak values 1.5 kA on sending end and the currents I1, I2 and I3 are of peak values 150 A on receiving end [5]. DC current Id has a steady state value of 1.5 kA and DC voltage Ed has a steady value of 150 kV, both with an initial switching disturbances lasting for 40 ms. Main : Graphs 1.5

Ia

Ib

Ic

I2

I3

I1

1.0 0.5 0.0 -0.5 -1.0 -1.5 0.15 0.10 0.05 0.00 -0.05 -0.10 -0.15 se c

0.06

0.08

0.10

0.12

0.14

0.16

Fig. 2. Three phase line currents on sending and receiving ends

Main : Graphs 2.00 1.75 1.50

Id

1.25 1.00 0.75 0.50 0.25 0.00 200.00 175.00 150.00 125.00 100.00

Ed

75.00 50.00 25.00 0.00 sec

0.00

0.10

0.20

0.30

0.40

0.50

Fig. 3. DC line current and voltage during normal operation of system

Symmetrical and Asymmetrical Fault Response of DC Link

3.1

393

Single Line to Ground Fault

A single line to ground fault is the most frequent type of fault (around 75%) and would occur between any line and ground. The simulation is done for fault occurring on line C at 120 ms and lasting for 50 ms. The automatic clearance is done within 50 ms [5]. The effect of LG fault on AC and DC systems is shown in Figs. 4 and 5. The disturbance is observed for 75 ms with the drop of voltage and current by 45% in this period in DC line.

Main : Graphs 2.0 1.5 1.0 0.5 0.0 -0.5 -1.0 -1.5 -2.0 -2.5 0.20 0.15 0.10 0.05 0.00

Ia

Ib

Ic

I2

I3

I1

-0.05 -0.10 -0.15 -0.20 sec

0.08

0.10

0.12

0.14

0.16

0.18

0.20

0.22

Fig. 4. Three phase line currents on sending and receiving ends during LG fault

Main : Graphs 2.00 1.75 1.50

Id

1.25 1.00 0.75 0.50 0.25 0.00 200.00 175.00 150.00 125.00 100.00

Ed

75.00 50.00 25.00 0.00 sec

0.00

0.10

0.20

0.30

0.40

Fig. 5. DC line current and voltage during LG fault

0.50

394

M. D. Atluri et al.

3.2

Double Line to Ground Fault

This fault causes more disturbances compared to LG fault. Phases A and B are in fault along with neutral. The simulation is done for this fault occurring at 120 ms and clearance in 50 ms. The effect of LLG fault on AC and DC systems is shown in Figs. 6 and 7. The disturbance is observed for 90 ms with the drop of voltage and current by 46.67% in this period in DC line.

Main : Graphs 2.0 1.5 1.0

Ia

Ib

Ic

I2

I3

I1

0.5 0.0 -0.5 -1.0 -1.5 -2.0 0.20 0.15 0.10 0.05 0.00 -0.05 -0.10 -0.15 -0.20 sec

0.10

0.12

0.14

0.16

0.18

0.20

0.22

Fig. 6. Three phase line currents on sending and receiving ends during LLG fault

Main : Graphs 2.00 1.75 1.50

Id

1.25 1.00 0.75 0.50 0.25 0.00 200.00 175.00 150.00 125.00 100.00

Ed

75.00 50.00 25.00 0.00 se c

0.00

0.10

0.20

0.30

0.40

Fig. 7. DC line current and voltage during LLG fault

0.50

Symmetrical and Asymmetrical Fault Response of DC Link

3.3

395

Triple Line to Ground Fault

This is the rarest but severe fault caused in the AC system. Generally during repairs, the person may tie all three lines and neutral together with supply off. After repair, supply may be given suddenly without untying the lines; this would cause heavy short circuit currents and severe fault in the system [3]. Simulation result shows the fault occurring at 120 ms and clearance in 50 ms. The effect of LLLG fault on AC and DC systems is shown in Figs. 8 and 9. The disturbance is observed for 90 ms with the drop of voltage and current by 83.33% in this period in DC line.

Main : Graphs 2.0 1.5 1.0

Ia

Ib

Ic

I2

I3

I1

0.5 0.0 -0.5 -1.0 -1.5 -2.0 0.25 0.20 0.15 0.10 0.05 0.00 -0.05 -0.10 -0.15 -0.20 -0.25 sec

0.10

0.12

0.14

0.16

0.18

0.20

0.22

Fig. 8. Three phase line currents on sending and receiving ends during LLLG fault

Main : Graphs 2.00 1.75 1.50

Id

1.25 1.00 0.75 0.50 0.25 0.00 200.00 175.00 150.00 125.00 100.00

Ed

75.00 50.00 25.00 0.00 se c

0.00

0.10

0.20

0.30

0.40

0.50

Fig. 9. DC line current and voltage during LLLG fault

3.4

Line to Line Fault

This fault may occur due to the short circuit between two lines mainly because of birds or any other external metallic elements. Lines A and B are shorted and is simulated to

396

M. D. Atluri et al.

occur at 120 ms and clearance in 50 ms. The effect of LL fault on AC and DC systems is shown in Figs. 10 and 11. The disturbance is observed for 90 ms with the drop of voltage and current by 46.67% in this period in DC line which is similar to that of LLG fault.

Main : Graphs 2.0 1.5 1.0

Ia

Ib

Ic

I2

I3

I1

0.5 0.0 -0.5 -1.0 -1.5 -2.0 0.25 0.20 0.15 0.10 0.05 0.00 -0.05 -0.10 -0.15 -0.20 -0.25 sec

0.10

0.12

0.14

0.16

0.18

0.20

0.22

Fig. 10. Three phase line currents on sending and receiving ends during LL fault

Main : Graphs 2.00 1.75 1.50

Id

1.25 1.00 0.75 0.50 0.25 0.00 200.00 175.00 150.00 125.00 100.00

Ed

75.00 50.00 25.00 0.00 sec

0.00

0.10

0.20

0.30

0.40

Fig. 11. DC line current and voltage during LL fault

0.50

Symmetrical and Asymmetrical Fault Response of DC Link

397

4 Observations and Discussion From the above results it is observed that the DC line current and voltage drop abruptly during AC faults. This may be caused due to the fact that the DC capacitor could discharge current to the fault area. This impact may even vary with the size of DC capacitor. The same DC capacitor is the reason for quick recovery of system and steady state stability of DC Line [6]. The impact of AC faults on DC line is shown in Table 1. Table 1. Impact of various faults on DC Line Fault type

Clearance time in DC line LG fault 75 ms LLG fault 90 ms LLLG fault 90 ms LL fault 90 ms

Percentage drop of voltage and current in DC line 45% 46.67% 83.33% 46.67%

5 Conclusion In this paper, faults in AC system are analyzed and the response of DC system is observed for various faults on AC side. Also the clearance time and transient occurrence and the percentage drop in DC current and voltage have been studied [6]. This work can be further extended in future for fault analysis of DC line with different lengths, different sizes of capacitors and even for the MTDC system. Furthermore, fault mitigation methodologies can be studied and applied for the extended versions of this work.

References 1. Wu J, Zhang S, Xu D (2013) Modeling and control of multi-terminal HVDC with offshore wind farm integration and DC chopper based protection strategies. IEEE. ISSN 978-1-47990224-8/13 2. Ma T (2017) Modeling and simulation of HVDC transmission based on PSCAD. In: Advances in computer science research (ACSR), volume 76 7th international conference on education, management, information and mechanical engineering (EMIM 2017) 3. May TW, Yeap YM, Ukil A (2016) Comparative evaluation of power loss in HVAC and HVDC transmission systems. IEEE. https://doi.org/10.1109/tencon.2016 4. Liu P, Che R, Xu Y, Zhang H (2015) Detailed modeling and simulation of ±500 kV HVDC transmission system using PSCAD/EMTDC. In: 2015 IEEE PES Asia-Pacific power and energy engineering conference (APPEEC) 5. Sheng CQ (2016) Simulation of HVDC control and transmission lines protection based on PSCAD/EMTDC. In: 4th international conference on machinery, materials and computing technology (ICMMCT 2016), Atlantis press 6. Karthikeyan M, Yeap YM, Ukil A (2014) Simulation and analysis of faults in high voltage DC (HVDC) power transmission. In: IECON 2014 - 40th annual conference of the IEEE industrial electronics society

Condition Assessment of Composite Insulator Removed from Service B. Sravanthi1(&), K. A. Aravind2, Pradeep Nirgude2, M. Manjula1, and V. Kamaraju3 1

Department of EEE, Osmania University, Hyderabad, India [email protected] 2 UHVL CPRI, Hyderabad, India 3 JNTU, Kakinada, India

Abstract. Composite Insulators are widely used because of their advantages, such as high contamination resistance, improved vandalism and high surface hydrophobicity. In recent years composite insulators usage in different contaminated areas has increased prominently. The contaminants affect the behavior of the composite Insulators. This paper presents 25 kV composite insulators condition assessment, removed after field exposure of 8 years from South Central Railway Region, India. Three different pollutant samples along with virgin sample are used for condition assessment. Various laboratory tests were conducted on the polluted samples; tests include ESDD, NSDD and clean fog tests. Simulation was executed for different models through coulomb 3D software, based on their relative permittivity as obtained from the polluted samples. Electric field distribution is found to be similar both in laboratory tests and simulation results. And all results have shown that the polluted insulators were observed to be in good condition after being exposed to 8 years in service. Keywords: Composite insulators
Contaminants
ESDD Railways
Clean fog test and electric field distribution


NSDD


1 Introduction Electrical insulator is a device which resists the flow of electric current and separate electrical conductors without allowing current through among them. Composite Insulators in recent years play an important role in maintaining the reliability of power system [1, 2]. Insulators performance in contaminated environment is one of the guiding factors in the field of insulation coordination of high voltage transmission lines [3]. A major concern for electric power utilities as they face socio-economic impact of power system outages due to high voltage Insulators pollution [4, 5]. Therefore, recently more attention has been paid to this area because of its complexity and importance. Hence, several models, norms and test guidelines have been proposed related to pollution flash over of insulators to avoid the consequences of power system outages [6]. Electrical discharge resulted from contaminated insulator is considered to be one of the most important problems [7, 8]. In general when insulator surface is © Springer Nature Switzerland AG 2020 S. C. Satapathy et al. (Eds.): ICETE 2019, LAIS 4, pp. 398–405, 2020. https://doi.org/10.1007/978-3-030-24318-0_48

Condition Assessment of Composite Insulator Removed from Service

399

exposed to humid by rain, fog or dew, then electrical discharge occurs resulted from the polluted content [9, 10]. The continuous ageing and service of in service insulator may lead to significant deterioration with their pollution withstand characteristics and result in insulator flashover. Composite insulators properties tend to change with time as they are generally installed in outdoor and can be affected by the condition of weather, mechanical loads, surrounding environment and electrical discharges in the form of corona or arcing. Such a reduction in the mechanical and electrical properties is termed as ageing; this reduction may lead to a reduction in working performance of the contaminated insulators followed by insulation failure. And the failure leads to leakage current that will result in severe flashover of contaminated samples. When composite insulators continuously subjected to different extreme levels of contamination then they tend to lose one of their most important properties of hydrophobicity. Therefore, periodic assessments of these insulators are to be performed. Experiences with composite Insulators are significantly less compared with porcelain one, hence providing the appropriate method to evaluate the performance of composite insulators is a needful enquiry by power utilities [11]. Indian Railways replaced most of the age old porcelain insulators with composite insulators. According to RDSO Lucknow, nearly 28% of Indian Railway lines across the country reported failures of porcelain insulators subjected to vandalism and heavy pollution. Since the demand of composite insulators have increased in recent years, a performance check is required for further improvement in design characteristics of composite insulators to meet the future pollution scenario of power sector [12]. This paper focuses on the condition assessment of 25 kV composite insulators removed from service from three different locations of South Central Railway region. Contaminated insulators for research study were selected from critical points of railway lines. The contaminated insulators used for study have had eight years of service. Insulators include, cement samples from Thandur line, coal sample’s from Manugur line and marine sample from Samarlakota line. Various tests were performed on the polluted samples and compared with the new sample to assess the behavior and pollution severity of in service samples.

2 Various Tests Performed This paper reports the assessment of contaminated samples by undertaking of various tests that were carried out at CPRI, UHVL Hyderabad. Investigations were performed on aged composite insulators after removal from their service by ESDD and NSDD methods to determine their pollution levels. Electrical performances of contaminated samples are observed. Under operating conditions a layer of contamination is deposited on the insulating surface. This layer when combined with moisture becomes conducting and a leakage current is flow through it [13]. This phenomena results in heat generation and evaporation of the water leading towards the formation of dry bands and the potential distribution of the insulator surface gets distorted. As a consequence it initiates the surface discharges and accelerates the ageing of insulator, and can lead to an undesirable flashover. Electric Field distribution was calculated on polluted samples

400

B. Sravanthi et al.

by approximately modeling 25 kV insulator by Coulomb 3D software. The electric stress distribution on 25 kV insulators depends on geometry of the insulator and capacitive field distribution along the surface of insulator. A relative performance of similar type composite insulators having same creepage and arcing distance, when subjected to different contaminants in field are studied and compared with virgin sample [14].

Fig. 1. Field aged samples and virgin sample.

2.1

Determination of Pollution Severity

The contaminants on composite insulators surface are composed to both water soluble and non-soluble materials. The soluble component consists of different types of salinity expressed as equivalent salt deposit density (ESDD) and non-soluble part of the contaminant expressed as non-soluble deposit density (NSDD). According to IEC 60507 the ESDD and NSDD methods were measured. And as per IEC 60815 the range of pollution classified as light, medium, high & very high [6, 7]. Observing the contaminants distribution on the insulator, the contaminants were found to be unevenly distributed on the surface of field aged samples shown in Fig. 1. The ESDD and NSDD results were shown in Tables 1 and 2. Results represent that there was no excess content of contaminants deposited on these samples even after 8 years of field exposure [15]. ESDD = (SA  V)/A

mg/cm2

ð1Þ

Where SA = (5.7 * r20)1.03 r20 = Volume Conductivity of Polluted water at 20 °C V = Distilled water volume A = Insulator’s washed surface area NSDD = Wf  Wi =A mg=cm2

ð2Þ

Condition Assessment of Composite Insulator Removed from Service

401

Where Wf is the weight of contaminant with filter paper under dry conditions in milligrams Wi is the weight of filter paper under dry condition in milligrams A is the Insulator washed surface area in cm2

Table 1. ESDD measurement ESDD Pollution range Contamination ESDD (mg/cm2) larger shed (mg/cm2) smaller shed Coal 0.045 0.557 Medium Cement 0.205 0.217 High Marine 0.069 0.102 High

Table 2. NSDD Measurement NSDD Pollution range Contamination NSDD (mg/cm2) larger shed (mg/cm2) smaller shed Coal 0.000303 0.005686 Light Cement 0.002685 0.005036 Light Marine 0.00276 0.00575 Light

2.2

Clean Fog Test

Electrical tests are performed in order to assess the capability of external electrical insulation to withstand the contamination [16–18]. Clean Fog test is performed as per IEC 60507 standard in an artificial test chamber at UHVL (Ultra High Voltage Laboratory), CPRI (Central Power Research Institute), Hyderabad. From the surface of insulator the leakage current is continuously recorded. This current represents the level of pollution severity and the clean fog results are comparable [19]. The experimental test set up with test sample is shown in Fig. 2. The Insulator is suspended vertically inside the artificial chamber. A plastic tent is surrounded around the test object, to limit the volume of the clean fog test chamber. 25 kVrms of the test voltage is applied. Digital ammeter is used to record this leakage current. Inside the chamber a continuous steam of de-ionized water is arranged, as it generates fog. The insulator samples are set to expose towards steam generation and the leakage current values are recorded. The magnitude of leakage current through the test samples is monitored continuously and the value is recorded for a total period of 90 min [20].

402

B. Sravanthi et al.

Fig. 2. Test setup for clean fog test

Fig. 3. Leakage current of all samples.

Clean Fog Tests Results The leakage current of contaminated samples is compared with virgin sample shown in Fig. 3. The maximum allowable safe leakage current is 0.25 mA [21]. In this research work, it is observed that the test insulator exposed to marine contamination has more leakage current as shown in Table 3.

Table 3. Leakage current values Sample ID S2-Coal S3-Cement S6-Marine

2.3

Leakage current (µA) 53 17 180

Electric Field Distribution

To study the behavior 25 kV composite insulator exposed under various pollutants, Coulomb 3D software was used. The basic design of polymeric insulator consists of fiber reinforced plastic core (FRP) covered with silicone rubber weather sheds equipped with metal end fittings having a relative permittivity (er) of 6, 4.5 and 1 respectively. The magnitude of the current flowing through the insulator relies upon its area and conductivity of the polluted conductive contaminants. For investigation, a uniform distributed polluted layer of 1 mm thickness was assumed on upper surface of insulator strings, and this was repeated for various polluted contaminants. The disks are numbered from bottom to top as line to ground respectively shown in Fig. 4d. The lower metal fitting is energized with 25 kV and top is connected to ground. The insulator division has alternative longer and smaller shed with a total creepage distance of 1050 mm and a dry-arcing distance of 385 mm Table 4.

Condition Assessment of Composite Insulator Removed from Service

403

Table 4. Maximum E-Field across the surface of insulator Insulator with contaminants Voltage applied in KV Max E-field in KV/mm Cement 25 0.795 Coal 25 0.629 Marine 25 5.02 Virgin 25 0.103

Fig. 4. a, b, c represent Efield curves & d represent Coulomb sample

Initially coal contaminants are applied across the 1 mm surface with uniform distribution along all the sheds of insulators. The Relative permittivity of 2.91 is applied and simulated in Fig. 4b. Similarly relative permittivity of 4.5 and 80 is applied to understand the behavior of cement and marine water pollutants in Fig. 4a and c respectively.

404

B. Sravanthi et al.

The computed electric field distribution results in Fig. 4a to c with 1 mm thickness of different (such as coal, cement and marine) surface contamination differ with each other based on their contamination level. Through simulation it was noted that the electric field stress increases as long as the contaminant severity increases, all graphs along the insulator surface represents U shape electric field. It was observed that electric field stress is more at metal end fittings and lower field in middle. The surface exposed to marine contamination is observed with highest electric field compared with other samples as shown in Table 4. It was also observed that the electric field strength along surface of the insulator increase with higher dielectric permittivity strength of the insulating material.

3 Conclusion From this research work, it is observed that the in-service insulators were still in relatively good condition after eight years of exposure to pollutants and there was no major degradation observed for the contaminated samples. It was noticed that composite insulators under SCR region are affected more by marine and cement contaminants deposited on the aged samples. All the tests report the effect of marine contaminants deposited on the surface of insulator would result in the increase of leakage current and reduce the surface resistivity. The electric field analysis on the surface showed that the effect of even a light pollutant layer on the test sample is observed to linearise the potential distribution along its length. However, the maximum electric field strengths on the sample occur at the surface with a small radius of curvature. The theoretical simulations were approximately similar viva field observations. It is concluded that there can be a possibility of reporting a flashover on the marine contaminated samples, and it may affect the reliability of the contaminated SC railway Insulators. Therefore, to improve the reliability, it is recommended to change the design of the insulators by increasing the creepage distance of the insulators to be used near cement contaminated regions. Further research work would be focused on marine and cement contaminated insulators with the aim to reduce leakage current and improved changes in creepage length of the samples. Acknowledgement. Authors thank the staff of CPRI for their technical support while conducting tests. We are grateful to South Central Railway for providing contaminated in service samples. We thank CPRI and Osmania University College of Engineering for permitting to publish this paper.

References 1. Looms JST (1990) Insulators for high voltages. IEE series (1990) 2. Aouabed F, Bayadi A (2010) Conductivity effect on the flashover voltage of polluted polymeric insulator under AC voltage. UPEC (2010) August–September, Setif University 3. Suflis SA, Topalis FV et al (2003) Study of the dielectric behaviour of non-uniformly polluted insulators. In: 13th International symposium on high voltage engineering, Netherlands. Foster I, Kesselman C (1999) The grid: blueprint for a new computing infrastructure. Morgan Kaufmann, San Francisco

Condition Assessment of Composite Insulator Removed from Service

405

4. Farzaneh M, Chisholm WA (2009) Insulators for icing and polluted environments. Wiley, Hoboken 5. Gopal S, Rao YN (2002) Flashover phenomena of polluted insulators. IEE Proc C: Gener Transm Distrib 131(July (4)). Foster I, Kesselman C, Nick J, Tuecke S (2002) The physiology of the grid: an open grid services architecture for distributed systems integration. Technical report, Global Grid Forum (2002) 6. IEC 60507 (1991) Artificial Pollution Tests on High Voltage Insulators to be used on AC systems, second edition 7. Rsuli H, Gomes C et al (2015) Surface arcing of Insulators due to bentonite contamination. J Electrostat 76:73–77 8. Dahabi B (2000) Flash over dynamic model of polluted insulators under AC voltage. IEEE Trans Electr Insul 7:283–289 9. Gencoglu MT, Cebeci M (2009) Investigation of pollution flashover on high voltage insulators using artificial neural network. Expert Syst Appl 36:7338–7345 10. Rizk FAM (1981) Models for pollution flashover. Electra 78:71–103 11. Rezaei M, Shariati M, Jabbari S (2011) Assesment of in service composite insulators in very harsh coastal environment of Iran: laboratory & field testing. CIRED June 2011 12. Technical Specification–Research Designs and Standard Organisation (RDSO), Lucknow 13. Hosseini SMH, Tavakoli MMM (2017) Investigation of the influence of hydrophobicity and dry band on the electric field and potential distributions in silicon rubber insulator. In: Lecture Notes in Engineering and Computer Science: Proceedings of The International Multiconference of Engineers and Computer Scientist 2017, 15–17 March 2017, Hong Kong, pp 1074–1080 14. Working Group 2.21,481 CIGRE, Guide for the Assessment of Composite Insulators in the Laboratory after removal from service 15. Sravanthi B et al Assesment of field aged composite insulators. In: The 20th ISH, Argentina, 27 August–01 September 2017 16. Composite Insulator Status program: Field inspection of Composite Line Insulator. STRI Guide 3 (2005) 17. Technical specifications of Indian Railways for Silicone Composite Insulators for 25 kV AC 50 Hz single phase overhead traction lines (No. TI/SPC/OHE/INSCOM/, p 21) 18. High Voltage test techniques – Part 1: General definitions and Test requirements (IEC 60060-1, 3rd ED 2010) 19. Sidthik S (2015) Evaluation and prediction of contamination level in insulators based on the leakage current characteristics using neural network 20. Rezei M (2013) Evaluation of actual field ageing on silicone rubber insulator under coastal environment. Life Sci. J 21. Amin M, Amin S et al (2009) Monitoring of leakage current for composite insulators and electrical devices. Adv Material Sci 21:75–89

Power Quality Improvement of Weak Hybrid PEMFC and SCIG Grid Using UPQC G. Mallesham(&) and C. H. Siva Kumar(&) Department of Electrical Engineering, University College of Engineering, Osmania Universtiy, Hyderabad, India [email protected], [email protected]

Abstract. In an advanced electrical power system, a wide range of loads used by industrial, domestic, irrigation, traction loads, commercial applications, advanced machinery of different electrical drives, control systems, and proliferation of different electronic gadgets day to day life, road, rail transport systems etc., have a greater impact on power quality. Another power quality issue is due to lesser power generating capacity of existing conventional resources and the combined operation of non-conventional energy resources into the power systems as hybrid electrical power generating systems. In this paper to decrease the gap between electrical power generation and demand added proton exchange membrane fuel cell as a hybrid electrical power system and addressed different power quality issues arise in power systems due to different loading conditions, impact of the tower shadowing effect of wind energy system in a weak grid system. A power electronics device: UPQC build up with instantaneous power theory to address these power quality issues within international standards. Keywords: Hybrid non-conventional energy system
Proton exchange membrane fuel cell
Electrical wind energy system Squirrel cage induction generator
Weak grid




1 Introduction Out of all major global issues, the more impacted issues on the globe are: re-duction of carbon emissions and reducing the gap between the production of electrical energy and electrical demand. One of significant contribute for the carbon emission is the dependency of conventional electrical power generating systems: thermal power station needs to lowdown the carbon emissions across the globe [1]. The alternative solution is dependency on non-conventional energy resources, like wind, solar, fuel cells, geothermal etc. All though these sources totally cannot be replaced with the conventional sources, as the capacity of the power generation is less, intermittent in operation, cost per unit is high. But these can be operated as co-generation or a hybrid energy system by operating parallel with the existing conventional energy source [2]. During the past decade there is a significant increase in penetration of wind energy based electrical generating systems and fuel cell based electrical energy systems in to the power system [3]. A major challenging task for the engineers is combined operation of these diversified characterized sources as a hybrid system to meet the power quality © Springer Nature Switzerland AG 2020 S. C. Satapathy et al. (Eds.): ICETE 2019, LAIS 4, pp. 406–413, 2020. https://doi.org/10.1007/978-3-030-24318-0_49

Power Quality Improvement of Weak Hybrid PEMFC and SCIG Grid Using UPQC

407

requirements. One of the today’s power quality issues of uncertain natured wind energy system is considered in this work: tower showing effect of wind system in a weak grid system [4]. Further, modelled two non-conventional energy resources: proton exchange membrane fuel cell (PEMFC) and wind electrical systems with squirrel cage induction generator (SCIG) as a hybrid non-conventional energy sources for the study. With the recent developments in power electronics technologies, controls methods, power theories and their applications to power systems have become a handy solution to many power system problems arise in power systems. Different power quality issues arise in distributed generating systems, power quality standards and mitigating techniques using different custom power devices have been discussed in [5]. From the literature review it is observed that majority of the researchers used instantaneous reactive power theory to enhance the power quality [6, 7], symmetrical components theory [8]. In this paper modelled an instantaneous power theory based custom power device: unified power conditioner (UPQC). The center of interest of the work is to enhance power quality in a hybrid nonconventional energy resources: PEMFC and SCIG wind power generating systems are connected in a weak grid using instantaneous power theory based custom power device: unified power conditioner (UPQC) within the international standards. The work in this paper is distributed in the following sections: Matlab simulation models of SCIG based wind energy system: wind turbine and SCIG, weak grid, PEMFC, and instantaneous power theory based custom power device unified power quality compensator are presented in Sect. 2. Details of the system considered for implementing different power quality issues in an environmentally friendly clean hybrid non-conventional energy resources to boost the quality of electrical power is conferred in Sect. 3. Finally concluded the work in Sect. 4.

2 Mathematical Models for Simulation Work Almost all real-world problems of engineering, the state-of-art method used by the designers before development and control the desired designs is by computer simulation. In this paper, MATLAB software is used to build up the require mathematical models of wind turbine system, wind electrical generator: SCIG, weak grid system, PEMFC, and instantaneous theory based UPQC. The following subsequent sections describes these models in detail. The incoming wind energy strikes the turbine blades which in turn rotates the high speed SCIG shaft through the gear box to generate the electrical power. 2.1

Wind Turbine

The dependency of generation of electrical energy by wind turbines have become the one of the major sources of non-conventional energy sources, because of its inexhaustible in nature, environmentally friendly, free and abundantly available in nature, great source of energy to remote places etc. A mathematical model of horizontal axis wind turbine considered for the study which is available in Matlab/Simulink

408

G. Mallesham and C. H. Siva Kumar

SimPowerSystems c libraries [9]. For a known wind ow, the dimensionless power coefficient Cp could be expressed as: Cp ¼ f ðk; bÞ ffi

16 ¼ 0:593 27

ð1Þ

where, - Cp power coefficient, - tip speed ratio, turning the angle of attack of wind turbine blades - b. The researchers shown the impact of tower shadow effect of wind energy system on the weak grid system is more [4, 10] and is expressed as: Vd ¼ Vh ð1 þ vm þ vtower Þ

ð2Þ

where, Vh: incoming wind speed: density of air, Vd: wind speed (disturbed), vm: variation in wind shear, and vtower: wind turbine disturbance due to tower shadowing. The following ratings are used for the study. Mechanical output of the wind turbine (MW) 21.6e6, pitch angle control gain (Kp) - 5, base wind speed - 14 m/s etc. 2.2

Wind Electrical Generator (Squirrel Cage Induction Generator)

SCIG is a wind turbine system is a fixed speed with the features: robustness, low price, mechanically simple, little maintenance, wider range of applications etc. The detailed modeling of self-excited induction generator is discussed in [11]. For the squirrel cage induction generator, library models of Matlab/Simulink SimPowerSystems c used for the simulation [10]. Total power: 21.6e6 watts, 690 Volts, Pair of poles: 3, 3 - transformer: 690 V/33 kV, 630 kVA. used for simulation studies. 2.3

Weak Grid

The concept of weak grid come into picture in different ways: with the presence of wind system and without presence of wind system with the basic meaning of not maintaining the voltage levels constant (undamaged oscillations) [4]. If r  20 then the grid is considered as a weak electrical grid system [12]. The short circuit power SSC  120 MVA. And it is calculated using the following equation: r¼

2.4

SSC ’ 5:5 PWF

Fuel Cell System: Proton Exchange Membrane Fuel Cell

PEMFC is one among the blooming non-conventional energy resources with the following salient features: modular designs, noise less, vibrations less, more power density per unit volume, less land area is required etc. [13]. For this work, modelled two parallel connected 50 kW, 625 V of DC proton exchange membrane fuels with the

Power Quality Improvement of Weak Hybrid PEMFC and SCIG Grid Using UPQC

409

following non-theoretical specifications: The stack power/cell is 50 kW, the internal resistance of the fuel cell on ohms: 0.65604, utilization of hydrogen and oxygen are: 99:25%, and 58:67% respectively. [14]. The mathematical expressions for open circuit voltage represented as: 8  44:43 RT
> 1=2 > þ In P 1:22 þ ð T  298 Þ P o > H2 2 > zF zF > > > >  < T  100 C  1 ð4Þ En ¼ 44:43 RT PH2 P1=2 o2 > > > þ In 1:22 þ ð T  298 Þ > > zF zF PH2 > > > : T ðtemperatureÞ [ 100 C where, - gas constant (R): 8.3145 J/(mol K), - Faraday constant (F): 96485 A s/mol, oxygen percentage in the oxidant - (%), PH2 - partial pressure of hydrogen, PO2 ; PH2 O and are oxygen and water vapor inside fuel cell stack. 2.5

Custom Power Device: Unified Power Quality Compensator

With the evolution of power electronics-based controller devices: custom power devices, used for enhancement of power quality in distribution systems. In this work, a instantaneous power theory based custom power device: Unified Power Quality Compensator (UPQC) is modelled to improve the electrical power quality at PCC. It consists of a series controller: injects the difference of PCC and reference voltages to compensates voltage fluctuations. With the shunt controller counteract the reactive power demand and maintain the DC link voltage remains constant. Both the series and shunt controllers are connected back to back with a common dc-link capacitor as shown in Fig. 1.

Fig. 1. UPQC: block diagram

410

G. Mallesham and C. H. Siva Kumar

The ab0 transformed or the Clarke transformed 3 – / instantaneous voltages va; vb, vc and currents ia; ib and ic. Voltage and current quantities are considered to determine Instantaneous active power (p) and reactive power (q) as: pab ¼ va ia þ vb ib

ð5Þ

qab ¼ va ia  vb ib

ð6Þ

3 Simulation Studies For simulation studies built a weak grid power system with a hybrid sources is shown in Fig. 2. Where BB (i), Zij and ZTR represents bus bar number, impedance between the buses i & j and impedance of the line respectively. At BB6 UPQC is connected in the system. Notation SW is used to represent the switch to connect the load. The rating of the wind energy system and PEMFC system are: 21.6 MW and 100 kW connected to the system through the trans-formers of ratings: 690 V/33 kV, 630 kVA and 100 MVA 690/33 kV respectively.

Fig. 2. Hybrid power system model: PEMFC and SCIG based wind energy system

The following simulation studies have been carried to increase quality of power of hybrid weak grid system. – Case 1: Power quality issue due to balanced sags. – Case 2: Power quality issue due to unbalanced swells. 3.1

Case 1: Power Quality Issue Due to Balanced Sags

Variation in wind speed: 13 m/s to 15 m/s for the entire simulation period. Average wind speed 14 m/s. It is clear that from the Fig. 3 the power quality issue aroused in wind energy systems: active power and reactive power fluctuations due to wind system Fig. 3(c) and (d) and their impact on weak grid Fig. 3(a), (b) and PEMFC terminals Fig. 3(e), (f) with magnitude of variations of oscillations in active power: −2.6 MW to −1.6 MW and reactive power 2.2 MVAr to 8.2 MVAr with fluctuating PCC voltages

Power Quality Improvement of Weak Hybrid PEMFC and SCIG Grid Using UPQC

411

between 2:64 104 V (peak-peak) and 2.68  104 V (peak-peak) unacceptable limitations of IEC standard IEC61000-4-15. It further, clearly demonstrates that power fluctuations are effectively invalidated by the UPQC Shunt Controller (SC) with in IEC standard IEC61000-4-15. In Fig. 4: represents: (a) Phase - a- Voltage at PCC (b) grid Phase -a - voltage (c) SCIG Phase - a - voltage (d) PEMFC Phase - a - during the simulation period. Currents drawn from grid, SCIG and PEMFC in presence of UPQC and without presence of UPQC in the circuit are shown in Fig. 5(I) and (II) respectively.

Fig. 3. Powers at the terminals (a) grid: P (b) grid: Q (c) SCIG: P (d) SCIG: Q (e) PEMFC: P (f) PEMFC: Q

Fig. 4. (a) phase ‘a’ voltage at PCC (b) grid phase ‘a’ VOLTAGE (c) SCIG phase ‘a’ voltage (d) PEMFC phase ‘a’

3.2

Case 2: Power Quality Issue Due to Unbalanced Swells

Both active and reactive power fluctuations due to the impact of tower shadowing effect is nullified similar to case 1. Figure 6(a) represents voltage on 33 kV side of the threephase transformer. As it clearly indicates the maintaining the constant voltage during the simulation period effectively by UPQC for different faults conditions created on other side of the transformer. Simulation operations: t = 0.0 s: series controller put into operation, t = 3.5 s: at BB2 15 MW +j 6.31 MVAr, t = 4.0 s: at BB4 2 MW - j2.4

412

G. Mallesham and C. H. Siva Kumar

MVAr, t = 4.6–4.8 s: at BB5 9.2 MW + j 1.85 MVAr, t = 5.2 s: 3- phase fault with Rf - 0.001 X and Rg - 0.001 X on secondary side 630 kVA, 33 kV/690 V transformer, t = 5.2–5.4 s: lL L with Rf 9 X and Rg - 0.001 X on transformer secondary side, t = 5.6–5.8 s: L - G with Rg - 0.001 X on transformer secondary side.

Fig. 5. (I)3 - phase currents (with compensator): (a) grid (b) SCIG (c) PEMFC. (II)3 - phase currents (without compensator): (a) grid (b) SCIG (c) PEMFC.

Fig. 6. Three phase voltages: (a) 3 - phase fault (b) 3 - phase fault with Rg: 0.001 X (c) L - L fault with Rf: 9 X and Rg: 0.001 X (d) L - G fault with Rg: 0.001 X.

Power Quality Improvement of Weak Hybrid PEMFC and SCIG Grid Using UPQC

413

4 Conclusion In this work, to increase the overall generating capacity modelled a non-conventional energy-based hybrid power system consisting of PEMFC and SCIG based wind electrical energy system connected to weak grid. Further, modelled a instantaneous power theory based series and shunt controllers of UPQC: to enhance the power quality issues aroused due to balanced load variations, un-balanced conditions and the impact of tower shadowing effect of wind energy system in a weak grid system. The outcomes of the simulation studies demonstrated the efficacy of the conditioner to enhance the power quality of the hybrid generating system within IEC 61000-4-15 and IEEE 1159-1995 standards.

References 1. Jones R, Haley B, Kwok G, Hargreaves J, Williams J (2018) Electrification and the future of electricity markets: Transitioning to a low-carbon energy system. IEEE Power Energy Mag 16(4):79–89 2. Guo S, Liu Q, Sun J, Jin H (2018) A review on the utilization of hybrid renewable energy. Renew Sustain Energy Rev 91:1121–1147 3. Karlsson K, Petrovic S, Hernando DA (2018) Global outlook on energy technology development. DTU international energy report 2018, p 21 4. Fan L, Miao Z (2018) An explanation of oscillations due to wind power plants weak grid interconnection. IEEE Trans Sustain Energy 9(1):488–490 5. Hossain E, Tr MR, Padmanaban S, Ay S, Khan I (2018) Analysis and mitigation of power quality issues in distributed generation systems using custom power devices. IEEE Access 6:16816–16833 6. Akagi H, Kanazawa Y, Nabae A (1984) Instantaneous reactive power compensators comprising switching devices without energy storage components. IEEE Trans Ind Appl 3:625–630 7. Akagi H, Nabae A, Atoh S (1986) Control strategy of active power filters using multiple voltage-source PWM converters. IEEE Trans Ind Appl 3:460–465 8. Korobeynikov BA, Shestack R, Ishchenko AI, Ishchenko D (2018) Symmetrical components filters for power system protection based on converters with rotating magnetic field. In: 2018 IEEE/PES transmission and distribution conference and exposition (T&D). IEEE, pp 1–5 9. Medved D (2012) Modeling of power systems using of matlab/simpowersystem. Elektroenergetika 5(2) 10. Sintra H, Mendes V, Melício R (2014) Modeling and simulation of wind shear and tower shadow on wind turbines. Procedia Technol 17:471–477 11. Heier S (2014) Grid integration of wind energy: onshore and o shore conversion systems. Wiley, Hoboken 12. Cai LJ, Erlich I (2015) Doubly fed induction generator controller design for the stable operation in weak grids. IEEE Trans Sustain Energy 6(3):1078–1084 13. Dicks A, Rand DAJ (2018) Fuel cell systems explained. Wiley, Hoboken 14. Spiegel C (2011) PEM fuel cell modeling and simulation using MATLAB. Academic Press, Cambridge 15. Devassy S, Singh B (2018) Design and performance analysis of three-phase solar PV integrated upqc. IEEE Trans Ind Appl 54(1):73–81

Comparison of State Estimation Process on Transmission and Distribution Systems M. S. N. G. Sarada Devi(&) and G. Yesuratnam Electrical Department, Osmania University, Hyderabad, Telangana, India [email protected], [email protected]

Abstract. This paper focuses on State Estimation (SE) process on transmission network and distribution network. State estimation obtains laudable data from the raw data supplied by measurement meters which are distributed in the network for available network model. This information is used in on-line monitoring and analyses of the network/system. This paper provides detailed comparison of state estimation process on transmission systems and on distribution systems in 9 different points. This paper also focuses on how measurement Jacobian matrix is changing in Distribution System State Estimation (DSSE). Keywords: Power system state estimation
Transmission System State Estimation and Distribution System State Estimation
Measurement Jacobian

1 Introduction System operating conditions monitoring is a continuous process for secure and uninterrupted operation of power systems to meet the consumer’s demand. In this connection different measurement meters (analog and logic/digital) are placed in different places in the entire system. The raw data given by these meters through RTU is transmitted to the energy management system. Power System State Estimation (PSSE) is one of the EMS functions which have been known as basis of EMS. SE is the process of obtaining system variables for given network model and measurements acquired from the system. Logic measurements are used to determine the system configuration and analog measurements are used to obtain the state variables. Detailed information about Transmission System State Estimation (TSSE) (like TSSE observability methods, available state estimator methods and bad data process) is given in [1–24]. Coming to distribution systems, load flow program is used for planning purpose. But now- a-days distribution automation also requires estimate of the system state. Detailed information about Distribution System State Estimation (DSSE) is given in [25–64]. However, to the best of our knowledge no paper is available on comparison of SE process on transmission and distribution systems. This paper focuses on this gap and provides the survey of state estimation process on transmission systems and on distribution systems. This paper also explains measurement Jacobian matrix in both the systems. © Springer Nature Switzerland AG 2020 S. C. Satapathy et al. (Eds.): ICETE 2019, LAIS 4, pp. 414–423, 2020. https://doi.org/10.1007/978-3-030-24318-0_50

Comparison of State Estimation Process on Transmission and Distribution Systems

415

2 Significance of Power System State Estimation The creditable data (each bus voltage magnitude and phase angle) obtained from state estimation is utilized by other EMS functions like contingency analysis, optimal power flow and security enhancement, limit checking program and economic dispatch etc. It is shown in Fig. 1.

Fig. 1. Significance of PSSE

3 Power Systems State Estimation Functions The power system state estimation (simply State Estimation) process has following functions: 1. 2. 3. 4.

Network Topology Processor (NTP) Network Observability Analysis (NOA) SE solution algorithm Bad data processing

The CB’s on/off information is given to NTP to form one line diagram of the network. Available measurements set and network configuration is considered in NOA. The Measurements set is added by pseudo measurements if network is unobservable with the available measurements. In SE solution algorithm, state variables are obtained. Any bad data in measurements is detected, identified and eliminated in bad data processing. Each function is explained clearly in [1].

416

M. S. N. G. Sarada Devi and G. Yesuratnam

Figure 2 shows block diagram of power system state estimation process.

Fig. 2. Block diagram of PSSE process

4 SE Process on Transmission Networks Transmission networks are more meshed typed and over-determined (number of available measurements are more than number of state variables). A. SE process on transmission networks: 1. Transmission Network Topology process: Based on the status data about CB and switches stored in database, one line diagram of the system is configured. 2. Observability Analysis: Algorithm for observability analysis can be either topological or numerical method. The measurement Jacobian matrix in TSSE is given in Eq. (1). Numerical method is widely used method for network observability analysis of the system. 2

@ ðPinjÞ @ ðPinjÞ @h @V

6 @ ðP flÞ 6 6 @h 6 6 @ ðQinjÞ H¼6 6 @h 6 @ ðQ flÞ 6 @h 6 4 @ ðVmagÞ @h

:

3 7

@ ðP flÞ 7 7 @V 7 @ ðQinjÞ 7 7 @V 7 @ ðQ flÞ 7 7 @V 7 @ ðVmagÞ 5 @V

ð1Þ

Comparison of State Estimation Process on Transmission and Distribution Systems

417

3. SE solution algorithm: A power system state estimator based on the Weighted Least Square (WLS) method is the most known. Detailed explanation of WLS method is given in [3]. 4. Bad data processing in TSSE: In bad data analysis, usually chi-square test for detecting bad data and largest normalized residual test for identifying bad data are performed (Table 1). Table 1. SE process on transmission networks Function NTP NOA SE solution algorithm (widely used) Bad data process Parameter and structural error processing

Widely used method Based on changes in Ybus Numerical method WLS–bus voltages as state variables Chi-square test for detection largest normalized residual test for identification Based on changes in Ybus

5 SE Process on Distribution Networks A. Challenges in DSSE: Fundamentally the distribution network characteristics hold opposing views to that of transmission network (and it is tabulated in Table 2), DSSE facing following challenges. 1. 2. 3. 4. 5. 6.

Conventional methods cannot be used directly Requires assumptions and modifications either in configuration or methods. Measuring meters are very few. Depending on pseudo measurements Low accuracy of pseudo measurements. Difference in the time references of real-time data.

B. Measurement Data in DSSE: As number of meter points is much lower in distribution network, most of the measurements data used in DSSE are pseudo measurements data. Real-Time Data: a. Voltage magnitude and phase angle at feeder bus and real and reactive power of the feeder. b. Customer smart meters report demand data.

418

M. S. N. G. Sarada Devi and G. Yesuratnam

Pseudo Data: a. Pseudo power injection measurements at feeder buses. b. Customer billing data. c. Typical historical load profiles. C. SE process on distribution networks: 1. Distribution Network Topology Process: Object oriented approach is used in NTP. The changes in the system state are identified by autonomous network operation with SE and automatically update the system model. 2. Observability Analysis Algorithm for observability analysis can be either topological or numerical method. 3. SE solution algorithms in DSSE Weighted Least Square (WLS) method with branch-currents as state variables is widely used method. 4. Bad data processing in DSSE In bad data analysis, usually chi-square test for detecting bad data and largest normalized residual test for identifying bad data are performed.

Table 2. Characteristics of distribution network Construction Redundancy Available Measurements Load Phase R/X ratio

Transmission N/W Meshed type No. of meter points are more V, Pinj, Qinj at buses, Pfl, Qfl of branches Balanced type Low

Distribution N/W Radial type No. of meter points are much lower Power or current injections at feeders, very few branch currents Unbalance type High

The measurement Jacobian matrix in DSSE is given in Eq. (2). Numerical method is widely used method for network observability analysis of the system (Table 3).

Comparison of State Estimation Process on Transmission and Distribution Systems

2 6 6 6 6 6 H¼6 6 6 6 6 4

@ ððIinjÞeq of PinjÞ @Ir @ ððI flÞeq of PflÞ @Ir @ ððIinjÞeq of QinjÞ @Ir @ ððI flÞeq of Q flÞ @Ir @ ðVmagÞ @Ir @ ðIflÞ @Ir

@ ððIinjÞeq of PinjÞ @Im @ ððI flÞeq of PflÞ @Im @ ððIinjÞeq of QinjÞ @Im @ ððIflÞeq of Q flÞ @Im @ ðVmagÞ @Im @ ðIflÞ @Im

419

3 7 7 7 7 7 7 7 7 7 7 5

ð2Þ

*Note: eq of means equivalent of

Table 3. SE process on distribution networks Function NTP NOA SE solution algorithm (widely used) Bad data process Parameter and structural error processing

Widely used method Object oriented approach [37] Numerical method WLS–branch currents as state variables Chi-square test for detection largest normalized residual test for identification Based on changes in Ybus

6 Comparison of TSSE and DSSE Transmission System State Estimation and Distribution System State Estimation are compared in nine points and those are listed in Table 4. From the Table 4 it is clear that, accuracy of the DSSE is mainly depends on available historical/forecasted data of demand/load. If pseudo measurements accuracy increases, accuracy of state estimator also increases. In TSSE, direct measurement partial differentiator (∂x) are used to form measurement Jacobian matrix where as in DSSE, considered measurements are first converted into equivalent current measurements. Later these measurements partial differentiator (∂x)eq are used to form measurement Jacobian matrix. In DSSE, branch-currents chosen as state variables to increase the estimator accuracy and reduce the calculations complexity.

420

M. S. N. G. Sarada Devi and G. Yesuratnam Table 4. Comparison of TSSE and DSSE

TSSE (i) State variables: X = [v, ø] in polar form X = [Vr, Vi] in rectangular form In TSSE nodal voltage based state vector is considered in polar form to reduce the calculations complexity (ii) Pseudo measurements: All measurements except critical measurements. Some of the available measurements are also comes under pseudo measurements (iii) Types of measurements: Voltage magnitude, real/reactive power injections/flows (iv) Measurements used in Jacobian matrix: V, Pinj, Qinj, Pfl, Qfl Voltage magnitude, real/reactive power injections/flows. These measurements are used to form H

(v) H in TSSE: Mostly used form is polar form. It is given in Eq. (1) (vi) Only un available data are calculated (vii) Weighted Least Square method with nodal voltages as state vector in polar form is mostly used to obtain the state variables (viii) Observability analysis and bad data process can be done easily because available measurements are more. Mostly used method is numerical method (ix) TSSE is more accurate

DSSE (i) State variables: (a) in branch current based method: X = [Ir, Im] in rectangular form X = [i, ø] in polar form (b) in nodal voltage based method: X = [v, ø] in polar form X = [Vr, Vi] in rectangular form (ii) Pseudo measurements: These are obtained from the historical information of the powers drawn by the loads. Pseudo measurements do not exist physically (iii) Types of measurements: Voltage magnitude, real/reactive power injection at substation, very few branch currents and real/reactive power of the loads (iv) Measurements used in Jacobian matrix: Ireal, Iequ All power measurements are converted into equivalent current measurements and those are used along with available substation voltage equivalent current and few current measurements to form H (v) H in DSSE: Mostly used form is rectangular form. It is given in Eq. (2) (vi) Unavailable and available data (load data) are calculated (vii) Weighted Least Square method with branch currents as state vector in rectangular form is mostly used to obtain the state variables (viii) Observability analysis and bad data process is difficult because available measurements are very few. Mostly used method is numerical method (ix) DSSE is less accurate

7 Conclusions State estimation provides creditable data from raw data supplied by the measurement devices which are distributed in the network for available network model. This information is used in on-line monitoring and analyses of the network/system. This paper provides detailed comparison of state estimation process on transmission systems

Comparison of State Estimation Process on Transmission and Distribution Systems

421

and on distribution systems. This paper also focuses on how measurement Jacobian matrix is changing in Distribution System State Estimation (DSSE). Here all power measurements are converted into equivalent current measurements and those are used along with available substation voltage equivalent current and few branch current measurements to form the measurement Jacobian matrix H.

References 1. Schweppe F, Rom D (1970) Power system static-state estimation: parts I, II, & III. IEEE Trans Power Apparatus Syst 89(1):125–130 2. Costa I, Leao J (1993) Identification of topology errors in power system state estimation. IEEE Trans Power Syst 8(4):1531–1538 3. Wu FF (1990) Power system state estimation: a survey. Int J Electr Power Energy Syst 12 (2):80–87 4. Monticelli A (1999) State Estimation in Power Systems: A Generalized Approach. Kluwer, Norwell 5. Abur A, Expsito AG (2004) Power System State Estimation: Theory and Implementation. Marcel Dekker, New York 6. Filho M, Leite da Silva A, Falcao D (1990) Bibliography on power system state estimation (1968–1989). IEEE Trans Power Syst 5(3):950–961 7. Monticelli A (2000) Electric power system state estimation. Proc IEEE 88(2):262–282 8. Zarco P, Gomez-Exposito A (2000) Power system parameter estimation: a survey. IEEE Trans Power Syst 15(1):216–222 9. Do Coutto Filho MB, De Souza JCS (2009) Forecasting-aided state estimation: parts I & II. IEEE Trans Power Syst 24(4):1667–1677 10. Arefi A, Haghifam M-R (2011) Smart Power Grids. Springer, Cham, pp 439–478 11. Huang Y-F, Werner S, Huang J, Kashyap N, Gupta V (2012) State estimation in electric power grids: meeting new challenges presented by the requirements of the future grid. IEEE Signal Process Mag 29(5):33–43 12. Liu W-H, Wu F, Lun S-M (1992) Estimation of parameter errors from measurement residuals in state estimation [power systems]. IEEE Trans Power Syst 7(1):81–89 13. Liu W-H, Lim S-L (1995) Parameter error identification and estimation in power system state estimation. IEEE Trans Power Syst 10(1):200–209 14. Alsac O, Vempati N, Stott B, Monticelli A (1998) Generalized state estimation. IEEE Trans Power Syst 13(3):1069–1075 15. Castillo E, Conejo A, Pruneda R, Solares C (2005) State estimation observability based on the null space of the measurement Jacobian matrix. IEEE Trans Power Syst 20(3):1656– 1658 16. Nian-De X, Shi-Ying W, Er-Keng Y (1982) A new approach for detection and identification of multiple bad data in power system state estimation. IEEE Trans Power Apparatus Syst 101(2):454–462 17. Monticelli A, Garcia A (1983) Reliable bad data processing for realtime state estimation. IEEE Trans Power Apparatus & Syst 102(5):1126–1139 18. Koglin H-J, Neisius T, Beiler G, Schmitt K (1990) Bad data detection and identification. Int J Electr Power Energy Syst 12(2):94–103 19. Chen J, Abur A (2006) Placement of PMUs to enable bad data detection in state estimation. IEEE Trans Power Syst 21(4):1608–1615 20. Singh D, Misra R, Singh V, Pandey R (2010) Bad data pre-filter for state estimation. Int J Electr Power Energy Syst 32(10):1165–1174

422

M. S. N. G. Sarada Devi and G. Yesuratnam

21. Garcia A, Monticelli A, Abreu P (1979) Fast decoupled state estimation and bad data processing. IEEE Trans Power Apparatus Syst 98(5):1645–1652 22. Falco D, Bezerra U (1991) Power system operating state forecasting for security analysis applications. Int J Electr Power Energy Syst 13(6):330–336 23. Shih K-R, Huang S-J (2002) Application of a robust algorithm for dynamic state estimation of a power system. IEEE Trans Power Syst 17(1):141–147 24. Blood EA, Krogh B, Ilic, M (2008) Electric power system static state estimation through Kalman filtering and load forecasting. In: 2008 IEEE power and energy society general meeting, July 2008, pp 1–6 25. Singh R, Pal B, Jabr R (2009) Choice of estimator for distribution system state estimation. IET Gener Transm Distrib 3(7):666–678 26. Roytelman I, Shahidehpour SM (1993) State estimation for electric power distribution systems in quasi real-time conditions. IEEE Trans Power Deliv 8(4):2009–2015 27. Baran M, Kelley A (1995) A branch-current-based state estimation method for distribution systems. IEEE Trans Power Syst 10(1):483–491 28. Lu C, Teng J, Liu W-H (1995) Distribution system state estimation. IEEE Trans Power Syst 10(1):229–240 29. Meliopoulos AS, Zhang F (1996) Multiphase power flow and state estimation for power distribution systems. IEEE Trans Power Syst 11(2):939–946 30. Lin WM, Teng JH (1996) Distribution fast decoupled state estimation by measurement pairing. IEE Proc Gener Transm Distrib 143(1):43–48 31. Lin W-M, Teng J-H, Chen S-J (2001) A highly efficient algorithm in treating current measurements for the branch-current-based distribution state estimation. IEEE Trans Power Deliv 16(3):433–439 32. Baran M, Kelley A (1994) State estimation for real-time monitoring of distribution systems. IEEE Trans Power Syst 9(3):1601–1609 33. Ghosh A, Lubkeman D, Jones R (1997) Load modeling for distribution circuit state estimation. IEEE Trans Power Deliv 12(2):999–1005 34. Singh R, Pal B, Jabr R (2010) Distribution system state estimation through gaussian mixture model of the load as pseudo-measurement. IET Gener Transm Distrib 4(1):50–59 35. Thukaram D, Jerome J, Surapong C (2000) A robust threephase state estimation algorithm for distribution networks. Electr Power Syst Res 55(3):191–200 36. Teng JH (2002) Using voltage measurements to improve the results of branch-current-based state estimators for distribution systems. IEE Proc Gener Transm Distrib 149:667–672 37. da Silva AL, Do Coutto Filho MB, De Queiroz JF (1983) State forecasting in electric power systems. IEE Proc Gener Transm Distrib 130(5):237–244 38. Meliopoulos APS, Cokkinides G, Huang R, Farantatos E, Choi S, Lee Y, Yu X (2011) Smart grid technologies for autonomous operation and control. IEEE Trans Smart Grid 2(1):1–10 39. Pau M, Pegoraro P, Sulis S (2013) Efficient branch-current-based distribution system state estimation including synchronized measurements. IEEE Trans Instrum Meas 62(9):2419– 2429 40. Korres G (2011) A distributed multiarea state estimation. IEEE Trans Power Syst 26(1):73–84 41. Kekatos V, Giannakis G (2013) Distributed robust power system state estimation. IEEE Trans Power Syst 28(2):1617–1626 42. Dzafic I, Gilles M, Jabr R, Pal B, Henselmeyer S (2013) Real time estimation of loads in radial and unsymmetrical three-phase distribution networks. IEEE Trans Power Syst 28 (4):4839–4848 43. Gómez-Expósito A, de la Villa Jaén A, Gómez-Quiles C, Rousseaux P, Cutsem TV (2011) A taxonomy of multi-area state estimation methods. Electr Power Syst Res 81(4):1060–1069

Comparison of State Estimation Process on Transmission and Distribution Systems

423

44. Gomez-Exposito A, Abur A, de la Villa Jaen A, Gomez-Quiles C (2011) A multilevel state estimation paradigm for smart grids. Proc IEEE 99(6):952–976 45. Abbasi AR, Seifi AR (2013) A new coordinated approach to state estimation in integrated power systems. Int J Electr Power Energy Syst 45(1):152–158 46. Hayes B, Hernando-Gil I, Collin A, Harrison G, Djokić S (2014) Optimal power flow for maximizing network benefits from demandside management. IEEE Trans Power Syst 29 (4):1739–1747 47. Meliopoulos A, Polymeneas E, Tan Z, Huang R, Zhao D (2013) Advanced distribution management system. IEEE Trans Smart Grid 4(4):2109–2117 48. Vargas A, Samper M (2012) Real-time monitoring and economic dispatch of smart distribution grids: High performance algorithms for DMS applications. IEEE Trans Smart Grid 3(2):866–877 49. Haughton D, Heydt G (2013) A linear state estimation formulation for smart distribution systems. IEEE Trans Power Syst 28(2):1187–1195 50. Song I, Yun S-Y, Kwon S, Kwak N (2013) Design of smart distribution management system for obtaining real-time security analysis and predictive operation in Korea. IEEE Trans Smart Grid 4(1):375–382 51. Deshmukh S, Natarajan B, Pahwa A (2014) State estimation and voltage/VAR control in distribution network with intermittent measurements. IEEE Trans Smart Grid 5(1):200–209 52. Baran M, McDermott T (2009) Distribution system state estimation using AMI data. In: 2009 IEEE/PES power systems conference and exposition, PSCE 2009, March 2009, pp 1–3 53. Arritt RF, Dugan R, Uluski R, Weaver T (2012) Investigation load estimation methods with the use of AMI metering for distribution system analysis. In: 2012 IEEE rural electric power conference (REPC), April 2012, pp B3-1–B3-9 54. Lopez G, Moreno J, Amaras H, Salazar F (2014) Paving the road toward smart grids through large-scale advanced metering infrastructures. Electr Power Syst Res 120:194–205 55. Alam S, Natarajan B, Pahwa A (2014) Distribution grid state estimation from compressed measurements. IEEE Trans Smart Grid 5(4):1631–1642 56. Sun H, Gao F, Strunz K, Zhang B (2013) Analog-digital power system state estimation based on information theory: parts I & II. IEEE Trans Smart Grid 4(3):1640–1655 57. Hippert H, Pedreira C, Souza R (2001) Neural networks for shortterm load forecasting: a review and evaluation. IEEE Trans Power Syst 16(1):44–55 58. Manitsas E, Singh R, Pal B, Strbac G (2012) Distribution system state estimation using an artificial neural network approach for pseudo measurement modeling. IEEE Trans Power Syst 27(4):1888–1896 59. Wu J, He Y, Jenkins N (2013) A robust state estimator for medium voltage distribution networks. IEEE Trans Power Syst 28(2):1008–1016 60. Hayes B, Prodanovic M (2014) Short-term operational planning and state estimation in power distribution networks. In: CIRED 2014 workshop, Rome, Italy, June 2014 61. Hayes B, Gruber J, Prodanovic M (2014) A closed-loop state estimation tool for MV network monitoring and operation. IEEE Trans Smart Grid 6(4):2116–2125 62. Choi S, Kim B, Cokkinides G, Meliopoulos APS (2011) Feasibility study: autonomous state estimation in distribution systems. IEEE Trans Power Syst 26(4):2109–2117 63. Singh R, Manitsas E, Pal B, Strbac G (2010) A recursive bayesian approach for identification of network configuration changes in distribution system state estimation. IEEE Trans Power Syst 25(3):1329–1336 64. Zhi Z, Min L, Shiyu X (2012) A survey on state estimation algorithm of distribution grid. In: 2012 IEEE Proceedings of the spring congress on engineering and technology (S-CET), May 2012

E-Mobility in India: Plug-in HEVs and EVs G. Sree Lakshmi(&), Vimala Devi(&), G. Divya(&), and G. Sravani(&) EEE Department, CVRCE, Hyderabad, India [email protected], [email protected], [email protected], [email protected]

Abstract. The future of the transportation is Hybrid Electric Vehicles. So it can be said that a huge wave of electric mobility is coming up but the knowledge of its time is not known yet. This paper briefs about different types of Electric Vehicles along with battery usage and majorly used batteries. The classification of different types of charging stations based on the capacity is given. This paper also gives the history of Electric Vehicles and in which countries the development is taking hike. India which is known to be a developing country is also taking initiative to bring EHV developed technology into the country and the government of India has also set plan flowchart to achieve those goals. The target and the companies which are ready to release the EHV vehicle and latest development are studied in this paper. In present EHV market Lithium ion battery is being used as its life is more. Keywords: Hybrid Electric Vehicles (HEV) Battery Charging Stations (BCS)


Electric Vehicles (EV)


1 Introduction In general hybrid car means which use more than on fuel for starting. With emerging technology, the name hybrid vehicle is used by the vehicles which are using combine gas- fueled internal combustion engine unit with battery driven. The hybrid vehicles are also called as green vehicles as they help us in green driving [1–3]. The first invention of hybrid vehicles started in the first half of 20th century in part the brain child of a Viennese coach builder named Jacob Lohner but it was too noisy and smelling bad. In 1990 Lohner-Porsche Elektromobil started but it was converted into purely hybrid vehicle by Ferdinand Porsche adding a combustion engine to recharge the battery which operates motor [4–6]. Charging of elektromobil’s was main draw back so electric motor with gasoline powered engine was developed by Porsche. The speed of this vehicle is 61.2 km per hour and this is a four wheel vehicle. This elektromobil is not considered as a green vehicle because it did not show any advantages of increasing fuel efficiency. 300 elektromobil were sold by the company named Porsche SE. As gasoline is less costly, therefore the government did not fund for the project so the development stopped but as the fuel depletion started and fuel cost is increasing © Springer Nature Switzerland AG 2020 S. C. Satapathy et al. (Eds.): ICETE 2019, LAIS 4, pp. 424–434, 2020. https://doi.org/10.1007/978-3-030-24318-0_51

E-Mobility in India: Plug-in HEVs and Evs

425

vehicle companies are getting attracted towards hybrid vehicles from late 2000. But now it is becoming a core segment of the automotive market of the future [7–9]. In 1997 again when executive vice president Akihiro Wadi made a statement regarding challenge to develop energy efficient company Toyota started its work. Toyota has done all surface work from then along with Prius to develop a fuel efficient hybrid vehicles which was introduced in Japan in the year 1997 and then to outside Japan in the year 2001. Over 1 million Priuses was sold worldwide. Secondly, Company named Honda introduced Accord Hybrid into the market. The car that started the whole Hybrid range in the world is Prius [10, 11].

2 Types of E-Mobility A. Hybrid Electric Vehicles (HV) A Hybrid electric vehicle uses both combustion engine and electric motor. But main source of power is gasoline only. Initially during starting, the Vehicle draws power from electric motor. When speed of motors increased then hybrid electric vehicle switches from electric motor generated power to gasoline power. If hybrid vehicle needs further more power then both the engines are used to supply power. Electric motor uses regenerative braking to grasp energy and store in battery which is used by motor to generate power and drive wheels. In few HEV, petrol is used to drive vehicles but consists of battery also. Hybrid vehicles can generate power and store in the batteries using petrol engine. B. Plug-in Hybrid Electric Vehicles (PHEV) These vehicles are also HV only but they have an option to plug into the power grid by cables. So the main source of power for PHEVs is electricity which is captured from grid. The stored energy in battery is used by the vehicle till the battery reaches the saturation point (SPB). After the (SPB) has reached internal combustion engine is used as generator by using a technique called regenerative braking to start the electric motor by petrol because stored energy in battery will not be sufficient all the time. Examples for HV are Ubiquitous Toyota Prius, Mitsubishi outlander PEHV, Sporty BMW i8 (Fig. 1).

Fig. 1. Plug-in hybrid vehicle using fuel and electricity

426

G. Sree Lakshmi et al.

C. Electric Vehicles (EV) Electric vehicle uses only electricity which is stored in the battery but does not have option to fuel with petrol and diesel. This technology is increasing as government is giving subsidies for improving market of electric vehicles. EV also reduces emission of harmful gases becoming environmentally friendly. The electric fuel to drive electric motor which further moves the wheel of EV is taken from electric grid. Few examples of Electric vehicle are Nissan leaf, tesla manufacturers and Renault zoe. Tesla is best known as manufacture of EV (Fig. 2).

Fig. 2. Electric vehicle with internal equipment

D. Fuel Cell Electric Vehicles (FCEVs) FCEVs also drive its wheel using electricity but stores energy as Hydrogen in fuel cell. Hydrogen and air combines and generate electricity and the byproduct will only be water. It takes only few minutes to fill the fuel cells by hydrogen which is advantageous (Fig. 3).

Fig. 3. Internal structure of FCEV

E-Mobility in India: Plug-in HEVs and Evs

427

3 Market of HEV in the World Year Wise Sale of EHV: From the year 1999 till date year wise sale of hybrid vehicles are as follows: 2012: From the year 1999 Norway ranks second with a hybrid market share of 6.9% of new car sales followed by the Netherlands with 3.7%, France and Sweden, both with 2.3%. 2014: Honda Motor Company Limited with cumulative global sales of more than 1.35 million hybrids as of June. The Hyundai Group with cumulative global sales of 200,000 hybrids as of March 2014. 2015: Ford Motor Corporation with over 424,000 hybrids sold in the United States till June. 2016: Global Lexus hybrid sales achieved the 1-million-unit milestone in March. Hybrid market share accounted for 38% of new standard passenger car sales, and 25.7% of new passenger vehicle sales including kei cars. Japan ranked as the market leader with more than 5 million hybrids sold, followed by the United States with cumulative sales of over 4 million units since 1999, and Europe with about 1.5 million hybrids delivered since 2000. Japan also has the world’s highest hybrid market penetration. 2017: Global sales are led by the Toyota Motor Company with more than 10 million Lexus. As of January 2017, over 12 million hybrid electric vehicles have been sold worldwide since their inception in 1997. As of January 2017, worldwide hybrid sales are led by the Toyota Prius, with cumulative sales of almost 4 million units. The Prius had sold more than 6 million hybrids up to January 2017. Global Lexus hybrid sales achieved the 1-million-unit milestone in March 2016. The conventional Prius is the all-time best-selling hybrid car in both Japan and the U.S., with sales of over 1.8 million in Japan and 1.75 million in the United States. EHV Cars in Different Countries Japan: Japan ranks as the world’s market leader with more than 5 million hybrids sold since 1997 in the year representing around 45% of cumulative global hybrid sales since their inception. After 18 years since their introduction in the Japanese market, annual hybrid sales surpassed the 1 million mark for the first time in 2014. With cumulative sales of over 4 million hybrids through December 2014, Japan surpassed the United States as the world’s largest hybrid market was also the first time that all eight major Japanese manufacturers offered hybrid vehicles in their lineup. U.S Market: After Japan the U.S market is 2nd largest seller of hybrid vehicles and it is about 36% of total worlds selling rate. After gradual increase it’s market selling rate increased up to 11 million by 2016. In U.S especially California is best seller of hybrid vehicles then New York and Florida. In U.S the largest selling company of 48% of EHV is Toyota Prius

428

G. Sree Lakshmi et al.

European Market: Europe, however, as of December 2016, ranked as the eighth largest plug-in market in the world and the fifth largest in Europe. This market has sold EHV about 13.6% of total worlds share. Toyota Prius Hybrid is the bestselling electric hybrid vehicles in Europe. Germany: One million electric vehicles on the road by 2020 is the bold aim of Germany’s “National Electro mobility Development Plan”. Federal Government has invested in the region of EUR 1.5 billion in electric mobility development. Australia: More than half of the people said they would seriously consider a hybrid vehicle. Hybrids and electric cars are not sold at a recognized rate, only 1.2% of the market sales are shown. In a survey 85% of Australians said that EVs could be a positive economic opportunity for Australia, while another 70% believe the world is in the middle of a transportation transformation that only occurs one every 100 years. Australia was “more than 10 times” behind the rest of the world in EV adoption. The adoption of plug-in electric vehicles in Australia is not demanded in market due to lack of government policies or incentives for adoption and deployment of low or zero emission vehicles.

4 India in EHV Adoption and Car Manufacturing Companies Complete acceptance of electric and hybrid cars is not done but the mindset of customers is changing slowly yet. But mostly Indians rely on CNG and LPG when alternate sources are considered. A big challenge for the growth of EV market in India is due to lack of lithium deposits. For this there is need to set up a battery manufacturing plant in India and the companies must look for other options to power such vehicles. The Maini Reva was the first affordable electric car ever produced in India by Chetan Maini. Soon, Mahindra acquired his electric firm and with the help of Maini himself, built the first e2O 2-seater car. Post that, Mahindra went a step ahead to launch the 4-seater version called the e2O Plus, with an all-electric range of 80 km, and easy charging. Toyota Prius well known and successful hybrid vehicle worldwide. The same technology which is used in the Toyota Prius is used in Camry executive sedan and became successful in India by selling 80%. Honda accord is not gained many customers so for the time being it has stopped its production in India but now it is competing with the Toyota by the new name Honda accord hybrid. Volvo XC90 T8 is named as safest car and now it has added hybrid version. The main aim of this company is to make each car electric. BMW i8: This electric car is powered by a 1.5-l, 3-cylinder engine that generates 320 Nm torque and 231 PS power to the rear tires while lithium-ion batteries power the front wheels. The battery can be charged in just 4 h. The car can go from 0 to 100 kmph in under 4.4 s.

E-Mobility in India: Plug-in HEVs and Evs

429

Parallel hybrid vehicles are the recent development in market. It is taking market place because of its performance and less complex system. It is suitable for long distance and engine and battery are connected to motor, so energy is not much wasted for conversion. A hybrid vehicle uses regenerative braking (Fig. 4).

Fig. 4. Toyota Prius

Government Encouragement in India for EHV: The Indian government understood that the electric vehicles are future, so it has set a goal to make all vehicles as electric cars by 2030 by a plan named National Electric Mobility Mission Plan 2020 and has released Rs 3.05 crores for research and development. The government has initiated 60% funds on the research and development (R&D) cost for developing indigenous low-cost electric technology that will help power two, three wheelers and commercial vehicles to reduce pollution. The automotive industry in India is further expected to increase the share in India’s GDP from 25% from 2022 to 15% with production of Electric Vehicles. Central Government Goals Set to Improve the EHV Life in India • The overall electric vehicle market for storage in India is likely to be 4.7 GW in 2022. • Over 50% of the market in 2022 will be driven by e-rickshaw batteries.200 charging stations are proposed to be set up in Delhi, Jaipur & Chandigarh. • Delhi government launched a subsidy scheme of INR 30,000 for the E-Rickshaws in 2016. Government is targeting of 6–7 Million electric and hybrid vehicle on road by 2020. • Smart charging company, new motion announced to invest INR 1000 crore in India on charging infra development. HEV and EV in India • Mahindra e2oPlus: Is the only manufacturer in India who is selling the electric cars. Mahindra e2oPlus with four doors has been launched in October 2016 in India. This is using a battery of 72 V lithium-ion battery. The battery which is charged fully can be used for running the car up to 9 h and battery can be charged within an hour. Price: ` 5.46 lakh - ` 8.46 lakh.

430

G. Sree Lakshmi et al.

• VOLVO cars are planning to increase their vehicles in India. XC90 will occupy the last quarter in 2019 in Bangalore. In 2019 it is planning to launch a fully electric car. • In survey done almost 1.4 million EV cars will enter in Indian market by 2030 whereas 30%of 2 wheels will be electric vehicles by 2030. • Gore’s electric scooter, which was built up by the Indian startup company names Okinawa will run up to 100–120 km when it is charged 1 time. As Indian market mostly has sails in 2 wheels the concentration is mainly on this type of vehicles.

5 Types of Batteries Electric-vehicle batteries differ from other batteries because they are designed to give power over sustained periods of time. Batteries for electric vehicles should be having high power-to-weight ratio, specific energy and energy density weight of the vehicle which will improve its performance. The main component which makes hybrid electric vehicles efficient is batteries. The batteries in these vehicles are large and occupy large space and are manufactured by few companies in Japan like Panasonic and Sanyo, chiefly. Many hybrids vehicles which are developed now are using lead acid, nickel metal hydride (NiMH) batteries, lithium-ion (Li-ion). Lead acid battery: Lead acid battery is not been used much as it is most dangerous, heavy and less efficient even though it is having advantages like less cost. Nickel-metal hydride: Drawbacks of lead acid battery made the use of nickel-metal hydride battery which is less dangerous than lead acid battery, but mining process is hazardous. It also offers reasonable specific energy and specific power capabilities. These batteries have long life. But due to high cost, high self-discharge and heat generation at high temperatures lithium-ion battery gained importance. Lithium-ion battery: Next mainly used batteries are these only because of high powerto-weight ratio, high energy efficiency, and good high-temperature performance, low self-discharge, rechargeable, less toxic and high energy density to weight ratio. These batteries are rechargeable. The major research is going on in this area. Lithium-ion batteries by adding capacitors to help lithium ion battery to increase its storage capacity. Hydrogen Powered Batteries: In this type of batteries the hydrogen is used continuously to charge a battery which consists inbuilt hydrogen fuel cell. Hydrogen which is stored in battery is mixed with the oxygen gas in air and runs it through a proton exchange membrane and releases electric current throughout the way. The only waste obtained by this process is water. Therefore it is a clean process and one more advantage of this process is the vehicle acts as a normal vehicle. The weight of the vehicle also decreases as tank size is reduced for storing highly pressurized hydrogen gas.

E-Mobility in India: Plug-in HEVs and Evs

431

In US market customers are shifting towards hydrogen and parallel hydrogen fuel stations are being built in areas because they need to be filled for every 300 miles. Japan is planning to build 160 fueling stations my march 2021 in Ireland region. California 50 stations by 2020. Lithium ion battery is shifting towards hydrogen fuel cell battery because hydrogen fuel cell batteries can store 236 times more amount of energy when compared per kilogram of lithium ion battery.

6 Different Techniques to Charge Batteries There are different techniques to charge a battery in different vehicles they are Regenerative braking, Gasoline engine, plug-In battery. Regenerative Braking: Even though brake is applied, the motor still rotates despite vehicle speed is decreasing. So regenerative braking will use the energy wasted to run the motor like generator and generate electricity. This electricity is stored in battery for running the Vehicle (Fig. 5).

Fig. 5. Regenerative braking

Gas Engine: Gas engine is used as generator to convert mechanical energy into electrical energy which is used to recharge a battery. Plug-in Battery: Addition to regenerative and gas engine, the battery can be charged by plugging it into an electric socket. Standard charging outlets are of 120 V which takes overnight time to charge, but if fast charging is needed the 240 V charging points can be used (Fig. 6).

432

G. Sree Lakshmi et al.

Fig. 6. Gasoline charging

7 Charging Equipment and Charging Stations Electric vehicle charging stations are divided into 3 types: Level 1: These are like A.C outlets with 120 V, 20 A current and maximum 2000 watts. These chargers take 10 to 20 h to charge battery completely (Fig. 7).

Fig. 7. Level 1 charging cables

Level 2: Vehicle is charged by special cord J1772 provided in EV. Maximum power it can store is 6000 to 12000 W. These chargers take 3 to 8 h and need to be fixed by technical expert only. The major advantages of these chargers are they work with Wi-Fi by notifying the consumers when the price of power from grid is less or more. The development in this field is going on to make smart grid in future (Fig. 8).

E-Mobility in India: Plug-in HEVs and Evs

433

Fig. 8. J1772 cord charger cable

Level 3: These chargers require 3 phase supply. Buses, commercial vehicle and business locations are major places for these charging points. 300 and more volts with 20000 W or more is supplied (Fig. 9).

Fig. 9. Level 3 charging stations

8 Conclusions This paper presents development and history of E-mobility in India and in other countries. Different types of mobility systems are presented. The latest technology in HEVand EV is given along with working principle. Various types of batteries and their advantages are given. Different types of charging stations and the schemes provided by the Indian Government are specified. To reduce the pollution and to save the resources for future generations shifting of smart vehicles is very much important. Many companies are coming forward to encourage with good schemes to make pollution free vehicle market by 2030.

434

G. Sree Lakshmi et al.

References 1. Khaligh A, Li Z (2010) Battery, ultracapacitor, fuel cell & hybrid energy storage systems for electric hybrid electric, fuel cell, and pull-in hybrid electric vehicles: state-of the art. IEEE Trans. Veh Technol 59(6):2806–2814 2. Morcos MM, Dillman NG, Mersman CR (2000) Battery chargers for electric vehicles. IEEE Power Eng Rev 20(11):8–11 3. Lopes JAP, Soares FJ, Almeida PMR (2011) Integration of electric vehicles in the electric power systems. Proc IEEE 99(1):168–183 4. Dum B, Kumath H, Tarascon JM (2011) Electrical energy storage for the grid: a battery of choices. Science 334:928–935 5. Hu X, Zou C, Zhang C, Liu Y (2017) Technologies development in batteries: a survey of principal roles, types and management needs. IEEE Power Energy Mag 15(5):20–31 6. Shahidinejad S, Filizadeh S, Bibeau E (2011) Profile of charging load on the grid due to plug-in vehicles. IEEE Trans Smart Grid 3(1):135–141 7. Rei RJ, Soares FJ, Pecas Lopes JA (2010) Grid interactives charging control for plug-in electric vehicles. In: 13th international conference on intelligent transportation systems, pp 386–391 8. Song Y, Yang X, Lu Z (2009) Integration of plug-in hybrid and electric vehicles experience from China in isolated systems. In: IEEE power engineering, energy and electrical drives, pp 49–54 9. http://auto.ndtv.com/news/top-5-hybrid-electric-cars-in-india-757164 10. http://www.pluginindia.com/powerandenergy.html 11. http://mahindrareva.com/faqs/charging

Unified Simulation Test Facility for Aerospace Vehicles K. Rama Rao1(&), B. Mangu2(&), and Manchem Rama Prasanna3 1

2

RCI, DRDO, Hyderabad, India [email protected] EEE Department, UCE, OU, Hyderabad, India [email protected] 3 Bangalore, India

Abstract. Aerospace vehicle carries On-Board Control & Guidance (OBCG) software (s/w) resident on On-Board Mission Computer (OBMC). This OBCG s/w is validated on various simulation test-beds for various mission critical errors. Non Real Time Simulation (NRTS) test-bed is used to validate the s/w for implementation, functionality and performance. Robustness of Controller design is validated through Monte-Carlo Simulation (MCS). Hardware-In-LoopSimulation (HILS) test-bed is used to validate OBCG s/w for functionality & performance in the presence of Real Time (RT) issues like communication delays, system lag, scheduling, interfacing, execution times & Hardware subsystems problems. Here, in this publication, a new concept of ‘Unified Simulation Test Facility for Aerospace Vehicles (USTFASV)’ is brought out, which facilitates Model Driven On-board Controller S/w Development (MDOBCSD), Non Real Time Simulation, Monte-Carlo Simulation with Test Automation (MCSTA), Real Time Full Simulation and Real Time Hardware-In-LoopSimulation. This Test Facility unifies all the simulations on single platform. Keywords: Unified simulation
Model driven
Model source code
Interface source code
Common utility files
NRTIS
NRTDS
MCS RTIFS
RTDFS




1 Introduction Aerospace vehicle is a fully automated intelligent system with no human- intervention, carrying OBCG s/w resident on OBMC integrated with other hardware subsystems to guide it towards the target point and deliver the payload. Once OBCG algorithms’ design [1, 2] for a given vehicle is completed, they are developed in to OBCG s/w & this s/w along with mission sequence is validated to make it free from mission critical errors before flight testing to ensure the flight success. This OBCG s/w is validated on various simulation test-beds for various issues. NRTS test-bed is used for development of OBCG s/w as well as to validate the s/w for implementation, functionality and performance [4]. These validated OBCG modules are then directly ported on to OBMC. Robustness of Controller design for deviations in the input data is tested and

© Springer Nature Switzerland AG 2020 S. C. Satapathy et al. (Eds.): ICETE 2019, LAIS 4, pp. 435–442, 2020. https://doi.org/10.1007/978-3-030-24318-0_52

436

K. Rama Rao et al.

validated through MCS with Test Automation. HILS test-bed [3] is used to validate OBCG s/w for functionality & performance in the presence of RT issues like communication delays, system lag, scheduling, interfacing, execution times & Hardware subsystems problems. Each test-bed has its own advantages and all are essential in various stages, to make the process of development & validation of OBCG s/w time and cost efficient, leading to timely delivery of OBCG s/w free of mission critical issues. Each test-bed has its own plant model s/w. Differences in the plant model cause variation in simulation results across different test-beds and lateral changes have to be updated on each testbed separately. Unified 6DoF simulation test-bed facilitates NRTS, MCS & HILS on a common platform, without making any changes in plant model s/w and to avoid repetitive model implementation and to avoid differences in model s/w across test-beds. Unified Simulation Test Facility (USTF) adopts the modularization concept. Each sub-system is modeled in to a single sub-routine based module so that sub-system can be accessed by calling this sub-routine. And total C/C++ Source Code Files (SCFs) are distributed among sub-folders meant for each subsystem as shown in Fig. 1. For example all files related to plant model are maintained in a folder named “SIXDOF” and all files related to control & guidance are maintained in a folder named ‘OBC’. It is continued for all other subsystems like TUR (Target Update Receiver), IIRS (Image Infrared Seeker), INS (Inertial Navigation System) etc. Each sub-system folder is selfsufficient with its SCFs to represent its mathematical model and interface with other sub-systems. Total SCFs of each sub-system are classified in to Model Source Code Files (MSCFs) and Interface Source Code Files (ISCFs). MSCFs represent the mathematical model of that particular subsystem whereas ISCFs are meant for interfacing i.e. to call sub-system module and for input/output interface. Same MSCFs are used for all modes of simulation, whereas ISCFs are different for different modes of simulation. ISCF file for integrated mode simulation is kept in ‘INTEGRATED’ folder. Two separate ISCFs are there in each folder one for Non Real Time (NRT) mode and other Real Time (RT) mode. Common Utility Files (CUFs) shared by multiple sub-systems are kept in ‘COMMON’ folder. One more ISCF is there in MCS folder, which invokes NRTIS workspace repetitively for MCS. MSCFs and ISCFs are physically kept at one fixed place i.e. in respective folders. Using same MSCFs and by selecting different ISCFs in Project Workspace (PWS), various modes of simulation are realized. So without making any changes in the MSCFs and just by selecting different ISCFs, all modes of simulation can be carried out. Making no changes in MSCFs ensures no model differences across all modes of simulation. Code in the MSCFs changes very frequently due to design/configuration changes during development phase whereas ISFCs remain more or less same, once sub-systems and interfaces are finalized. In case of three separate NRTS, MCS and HILS test-beds, all these changes are to be updated on three test-beds. But in USTF, once the changes are updated in NRTS mode, they are auto updated for MCS & HILS modes also, as same MSCFs are used for all modes. On USTF, all modes can be simulated in single computer (PC). In case of RTDFS, it can be simulated within a computer or different sub-system modules can be

Unified Simulation Test Facility for Aerospace Vehicles

437

distributed in to different computers keeping the Sixdof plant model in the same PC. In case of porting to other computers, sub-system folder can be directly copied as it is and run there. In HILS mode, Sixdof plant model is used as it is whereas other mathematical models are replaced with hardware subsystems. It consists of OBCG s/w integrated with mathematical models of hardware subsystems like INS, actuation system, aerospace vehicle (plant), seeker and radar in closed loop. It is a huge C/C++ code with greater complexity consisting of 20000 lines, 50 files and 300 Sub-routines. It is developed on Linux platform and compatible with Windows also. It has a provision to carry out Non Real Time Integrated Simulation (NRTIS), Non Real Time Distributed Simulation (NRTDS), Monte-Carlo Simulation (MCS), Real Time Integrated Full Simulation (RTIFS), Real Time Distributed Full Simulation (RTDFS) and Real Time Hardware-In-Loop-Simulation (RTHILS).

2 Development of USTF A systematic step-by-step method is followed for development of USTF. They are in sequence: (i) Sixdof plant model development, (ii) model driven controller s/w development, (iii) NRT integrated closed loop simulation [4], (iv) ensure complete decoupling, (v) NRT distributed closed loop simulation, (vi) RT integrated full simulation, (vii) RT distributed full simulation and (viii) RT distributed with hardware in loop. 2.1

Non Real-Time Integrated Simulation

A single project workspace is created in INTEGRATED sub-folder as shown in Fig. 2. All the MSCFs of all the sub-systems from their respective sub-folders along with ISCF meant for NRTS from INTEGRATED sub-folder are included in the workspace. In this mode individual ISCF are not needed and hence not included in the project workspace. In this mode, all the sub-system modules are called as per the mission sequence and @ specified time intervals from ISCF. And data exchange between various subsystems is through global variables. In this mode, controller and the plant interact with each other in closed loop through global variables. As it is Non Real Time Simulation or Free Running Simulation (FRS), execution times of all modules are considered as zero i.e. ideal execution times. Execution is sequential execution. And time is updated at each minor cycle after execution of all the modules. It is the fastest simulation mode. Its run time depends on the processor speed and algorithm complexity. Run time is completely independent of flight time. In high end work station, it hardly takes 1 min per run. Hence NRTS is carried out in this mode only. NRTIS being the fastest mode, majority of the issues are cleared and majority of failure-mode test-cases are simulated in this mode only. This mode facilitates subroutine level validation and has advanced debugging tools.

438

2.2

K. Rama Rao et al.

Non Real-Time Distributed Simulation

A separate project workspace is created for each sub-system in corresponding subfolders as show in Fig. 3. All the MSCFs along with ISCF meant for NRTS of that particular sub-system are integrated in this workspace. In this mode any of the ISCFs residing in INTEGRATED folder is not used. Data exchange between various subsystems is through file sharing. Hence in this mode each individual ISCF has three tasks i.e. input data reading, invoking that particular module and writing of output data. These tasks are executed in sequence @ specified time interval. All the workspaces/sub-systems will be running independently but again in sequence. One module will be waiting for the fresh input data till the other module completes execution and writes output data for the first one. Hence to avoid dead-lock at simulation start, it is to be ensured that the module which is run prior to all, generally Sixdof model should not wait for input data prior to module execution at least for the first cycle. Again it is pure, Non Real Time Simulation or Free Running Simulation (FRS), as the execution times of all modules are considered as zero i.e. ideal execution times. Execution is sequential execution. And time is updated at each minor cycle after execution of all the modules in each individual workspace. This mode of simulation ensures complete decoupling of each sub-system module from all other sub-system modules and decoupling of plant s/w & controller s/w from each other. This simulation is also called as ‘Virtual Distributed Simulation’, because of the fact that different sub-systems are distributed in different workspaces but in same computer. Simulation results of NRTIS and NRTDs should exactly match even at sixth/seventh decimal level also. In this mode, input/output files are to be accessed from hard disk. Hence it is a little bit slow. Some time it may take more than flight time also for one run. Generally it takes 5–10 min per run. Hence this mode is not used regularly for NRTS. It is used only for ensuring self-sufficiency and complete decoupling of each sub-system module from other modules. These self-sufficient modules can be then simulated in the same computer or in some other computer directly without any change in the source code. Once complete decoupling is ensured and simulation results are proper, now all sub-system models are ready for ‘Real Time Distributed Simulation (RTDS)’, where each subsystem model is simulated in a separate simulator/Computer in closed loop with other subsystem simulators through hardware interfaces. 2.3

Monte-Carlo Simulation

Monte-Carlo Simulation is a test automation technique, where number of runs (in thousands) can be given with a single click. This simulation is to validate the robustness of the Controller design. Some input parameters like thrust, drag, mounting misalignments and launcher parking angle are varied stepwise/randomly and performance is evaluated during simulation. In this mode, a separate project workspace is created in MCS sub-folder, and ISCF from MCS folder is included in the workspace. This ISCF calls integrated workspace created in the INTEGRATED folder for NRTS, for pre-decided number of times. Every

Unified Simulation Test Facility for Aerospace Vehicles

439

time input parameters are varied stepwise/randomly by introducing some noise and output is saved for off-line analysis. 2.4

Real-Time Integrated Full Simulation

This mode is same as NRTIS in the sense that all the MSCFs from all sub-system folders and ISCF meant for RTS from INTEGRATED folder are integrated together in single project workspace created in the later folder as shown in Fig. 2. Two main differences are there between NRTIS & RTIFS. In RTIFS, time is updated based on RT Clock (RTC) whereas it is Free Running Clock (FRC) in NRTIS. And it is sequential execution in NRTIS whereas it is parallel execution using threads in RTIFS. Library files related to RTC & Threads are included in ISCF meant for RTIFS. Again data exchange between various sub-system modules for closed loop simulation is through global variables only as in the case of NRTIS. In this mode, run time is exactly same as flight time. And simulation results should exactly match with NRTIS & NRTDS results provided nowhere execution time crosses the scheduled time interval throughout the run. This mode is to ensure nowhere execution time crosses the scheduled time interval throughout the run. Once results match with that of NRTS, we can go ahead with RTDFS. 2.5

Real-Time Distributed Full Simulation

As far as workspace creation is concerned, it is same as NRTDS as show in Fig. 3. Separate project workspace is created for each subsystem by integrating MSCFs & ISCF meant for RTS form that folder. Three main differences are there between NRTDS & RTDFS. In RTDFS, time is updated based on RTC in place of FRC. And it uses parallel execution using threads in place of sequential execution. This mode uses hardware interfaces like MIL-STD-1553/RS422 in place of file sharing for data exchange between sub-system modules for closed loop simulation. Two different ISCFs are maintained in each sub-system folder, one for NRTDS and other for RTDFS. Library files related to RTC, Threads & MIL-STD-1553/RS422 are included in each ISCF. In this mode multiple sub-systems can be simulated in a single computer or they can be distributed in multiple computers. But MIL-STD-1553/RS422 hardware interface remains the same. In this mode also, run time is exactly same as flight time. And simulation results may not exactly match with RTIFS results. But they should fairly match. This difference is due to message scheduling between various subsystems. Once the simulation is satisfactory and meeting the performance requirements, we can go ahead for RTDHILS. 2.6

Real-Time Distributed Hardware-In-Loop Simulation

Till this point, all sub-systems are represented with their mathematical models and all the above simulation modes are called as Full Simulation (FS) modes. Now these

440

K. Rama Rao et al.

models are replaced with their actual hardware sub-system one by one in HILS. Now validated OBCG s/w is directly ported on to OBC and integrated with scheduling and interfaces. Now OBC simulator i.e. Controller is replaced with actual OBC hardware. This set up, OBC connected in closed loop with mathematical models of all other subsystems is called OBC-In-Loop-Simulation (OILS) test set up. If INS model is replaced with INS hardware, then it is Full- Stimulation- sensor InLoop (FSIL) [3]. If INS is mounted on Fight Motion Simulator (FMS) and FMS is driven by SiXDof PC, then it Sensor-In-Loop (SIL) simulation [3]. Similarly ActuatorIn-Loop (AIL) Simulation when actuator model is replaced with Actuator and SeekerIn-Loop (SkIL) Simulation when Seeker hardware replaces model. Once HILS results are satisfactory and fairly match with NRTS results, OBC s/w is cleared for flight testing [3].

3 Figures See Figs. 1, 2, 3 and 4.

Fig. 1. Distribution of source code files

Fig. 3. Distributed NRT/RT simulation

Fig. 2. Integrated NRT/RT simulation

Fig. 4. Graphical user interface

Unified Simulation Test Facility for Aerospace Vehicles

441

4 Results and Validation NRTIS mode is validated for functionality and performance. Also NRTIS is validated by comparing closed loop simulation results and FLIP run results with reference to design level results. Advanced debugging tools aid in sub-routine level validation. Once NRTIS is validated, it becomes reference for NRTDS. NRTDS is validated by comparing closed loop simulation results with reference to NRTIS results. In the same way NRTDS is reference for RTIFS, RTIFS is reference for RTDFS and RTDFS is reference for RTHILS. Finally HILS results should fairly match with NRTS results.

Fig. 5. Comparision of all modes

Fig. 6. Monte-Carlo simulation

5 Advantages USTF unifies all the simulations on single test-bed. Same time it maintains all the unique advantages of each individual simulation modes. OBCG s/w is validated for majority of the mission critical issues in NRTIS mode itself. OBCG s/w is validated for implementation, functionality & performance and for any design deficiencies in this mode. As NRTIS mode is very fast and has advanced debugging tools, flip provision for validation, sub-routine level validation feature, it saves lot of time during HILS. Then NRTDS mode ensures self-sufficiency of each sub-system module and complete decoupling of each module from all other modules. Otherwise it would have taken lot of time in HILS. Because of this systematic procedure and clearing errors one by one in various modes of simulation, by the time HILS starts, bug-free OBCG s/w ready, which is directly ported on to OBC. At the same time test-bed is completely established for HILS. It saves lot of time and cost. Minimal effort for plant model integration and incremental changes thereafter. Repetition of plant model integration and incremental changes thereafter on each testbed is avoided. Once it is done for one mode of simulation, it is auto updated for other modes also. And hence chances of committing implementation mistakes are reduced. Hence USTF is proved as time effective. MSCFs are not disturbed. Only separate ISCFs are used for various modes of simulations. Distributed simulation mode can be used to validate any third party sub-system/model.

442

K. Rama Rao et al.

Challenges. As it is a common test facility for various simulations and handled by different people with expertise in different fields, implementation and maintenance becomes a little bit challenging. This can be overcome by following a systematic procedure for implementation and maintenance. The MSCFs along with ISCFs related to NRTIS, NRTDS & MCS are implemented and maintained by NRTS expert whereas ISCFs related to RTIFS, RTDFS & RTDHILS are implemented and maintained by HILS expert. Any changes thereafter are to be properly communicated among the experts. USTF does not demand any extra memory/processor cost if they are done in separate simulators as per the conventional method. In fact, resources requirement is reduced if all are done in a single system, which is the high end simulator among all.

6 Conclusions Scheme of USTF is presented, which is very time effective and cost effective. Development procedure, its superior features over other simulation test-beds are brought out. All three simulation test-beds (NRTS, MCS and HILS) are brought on to single platform. Instead of having three separate test-beds, all three modes of simulation are done on same test-bed. Step-by step systematic procedure for develop of USTF is presented, which ensures the minimal effort. Simulation results comparison of NRTIS, NRTDS, RTIFS, RTDFS and RTHILS is presented in Fig. 5. Results of NRTIS, NRTDS & RTIFS modes are exactly matching as expected. Results of RTDFS are slightly deviating from that of above three because of MIL-STD-1553/RS422 hardware interface. Again results differ between RTDFS & RTHILS as model is replaced with actual hardware. MCS results for varying thrust, drag and Centre of Pressure (CP) location are shown in Fig. 6 Acknowledgment. Authors are grateful to Dr. G. Sateesh Reddy, Chairman DRDO, Sri B.H.V. S.N. Murthy, Director RCI, G. Venkat Reddy, OS, Director, DNEC and Sri. Vijaya Sankar, Sc-‘G’ for their constant encouragement in carrying out this research work. The authors are also thankful to Sri U. Raja Babu, Programme Director AD, Dr. Y. Srinivasa Rao, PD (FV-Exo) and Sri R. Venkatrami Reddy, DPD (AD-Missions) for their useful suggestions in related areas.

References 1. Greensite L (1970) Control theory: volume-II: analysis and design of space vehicle flight control system 2. Garnell P (1980) Guided weapon control system, 2nd edn. Pergamon press, Oxford 3. Internal reports (2017) DHILS, RCI, Hyderabad 4. Internal reports (2018) DNEC, RCI, Hyderabad

Machine Learning Based Prediction Model for Monitoring Power System Security N. Srilatha(&), B. Priyanka Rathod, and G. Yesuratnam Department of Electrical Engineering, Osmania University, Hyderabad, India [email protected]

Abstract. Real-time operation of electric power systems requires monitoring and control on a continuous basis. This paper proposes Machine Learning (ML) based Prediction model for assessment and enhancement of static security. The objective of the model is to assess the security state of the current operating point by predicting Static Security Index (SSI). If the system is insecure or critically secure, generators are re-dispatched based on Relative Electrical Distance (RED) to bring back the system to secure state. The re-dispatch of generators is defined by a control vector, denoted as Corrective Generation Schedule (CGS). Most suitable Machine Learning based classifier is identified to make the necessary predictions of the security state and the required corrective action. Correlation-based Feature Subset Selection (CFS) is utilized to improve the classification process. The prediction statistics obtained for IEEE 39-bus New England system clearly indicate effectiveness of the proposed Machine Learning based model. Keywords: Security assessment
Security enhancement
Generation rescheduling
Machine Learning
Decision Tree
Random Forest

1 Introduction Online monitoring and control of power systems is the need of the day, as the power systems have become complex in structure and operation. Power system security is one of the key aspects that has to be monitored and managed in real time. Contingency analysis forms the basis for security state monitoring. All possible contingencies are analyzed in terms of their severity in an offline mode, and this analysis aids in online predictions. However, it involves considerable computation effort and large execution time [1]. Recently, artificial intelligent techniques like expert systems, fuzzy logic and neural networks are being used for security assessment. [2, 3] use neural networks to estimate the power system static security. Multiclass SVM, a pattern recognition technique is used in [4] for classification and assessment of security. Formation of appropriate input vectors and dealing with nonlinearly separable classes needs to be improved further. Hence there is a pronounced need for automatic learning and prediction methods. Machine Learning (ML) Techniques are more suitable for such capabilities, like learning fast and reproducing in reliable manner. Compared with other methods,

© Springer Nature Switzerland AG 2020 S. C. Satapathy et al. (Eds.): ICETE 2019, LAIS 4, pp. 443–450, 2020. https://doi.org/10.1007/978-3-030-24318-0_53

444

N. Srilatha et al.

automatic learning performs better in terms of computational efficiency and interpretability [5]. ML techniques utilize the knowledge from intensive offline calculations to discover the built-in characteristics of the system and develop prediction models for various desired objectives [5, 8]. ML techniques are of use in case of complex problems where large amount of data has to be handled without explicit rules or equations, and when the nature of data keeps on changing. These techniques have been widely used for different power system control applications in the literature, like stability analysis, voltage stability improvement, security assessment, etc. [2, 7–10, 13]. Security assessment is implemented using Multiway Decision Trees (MDT) in [14]. MFNN and RBFN models are used in [15] for the same objective. However, enhancement of security in critical situations is not considered. It is also equally necessary to quantify the security level, along with ranking of contingencies, as this ranking may change with the load in the system. Hence in this paper, security enhancement is also implemented along with security assessment; the usage of different ML techniques is examined on the test system, DT based Random Forest ML technique proves to be more efficient. The key contributions of the paper include (i) security assessment model for predicting the security level of the system and (ii) security enhancement using generation rescheduling using ML based prediction model. The paper is organized as follows: Sect. 2 presents the scheme adopted for security monitoring. The proposed prediction model using ML techniques is explained in Sect. 3. Section 4 deals with case study and results, proving the effectiveness of the selected technique. Relevant conclusions are presented in the next section.

2 Scheme of Security Monitoring Real-time operation of power systems requires monitoring and control of system security on a continuous basis. Security assessment is the process of determining if the power system is in a secure or alert (insecure) state. It evaluates the security level of the system to a set of probable contingencies. 2.1

Security Assessment

Static security of a power system addresses whether, after a disturbance, the system reaches a steady state operating point without violating system operating constraints called ‘Security Constraints’. Violation of any of these constraints may disturb the system and finally lead to ‘black-out’. Static Security Index (SSI) [4] is a measure of static security of the system. Line outages, generator outages and faults resulting in loss of lines or loads are examples of contingencies that could cause a threat to static security. SSI is used to measure the static security level for a given system operating condition and a specified contingency. SSI is expressed as a function of Line Overload Index (LOI) and Voltage Deviation Index (VDI) [4].

Machine Learning Based Prediction Model

445

Table 1. SSI values Static Security Level Class A: Secure Class B: Critically Secure Class C: Insecure Class D: Highly Insecure

2.2

Static Security Index (SSI) SSI = 0 SSI > 0 & SSI  5 SSI > 5 & SSI  15 SSI > 15

Security Enhancement

When the system is diagnosed to be insecure or critically secure, security enhancement is required. Security enhancement is attempted for the cases where SSI values are greater than 5, as indicated in Table 1. This work examines the security violations because of line outages, that result in transmission line congestion. The congestion in the line is reduced by re-dispatch of generators using RED concept. Economical rescheduling of the generators is also incorporated to reduce the cost of re-dispatch [12]. This method has an additional advantage of choosing minimum number of generators, reducing the number of control variables that have to be modified. RED (Relative Electrical Distance) of generators from the overloaded line is used as the basis for generation rescheduling. It represents the relative electrical distance of the generator to the overloaded line. The lesser this distance, the more is the desired contribution of a particular generator. The actual contribution of each generator to the congested line is first calculated. Among all the generators, those which are contributing to the congested line are identified as Generation Decrease group (GD group). This is the group where generation decrease is recommended to relieve the overloaded line; And the other generators are categorized under Generation Increase group (GI group). This is the group where generation increase is recommended. At any instant of time, total generation change in GD group must be same as the total generation change in the GI group. The amount of generation change required to relieve the congestion of the mostly congested line is calculated using RED [12]. In doing so, the new generation schedule not only reduces the overload in the congested line, but also improves the system parameters like voltage stability index and losses.

3 Proposed ML Based Prediction Model The first aspect of monitoring the power system is to identify the state of security of its operation, i.e., secure or critically secure or highly insecure. Hence tedious offline calculations are done for multiple operating conditions, encompassing all credible contingencies for different loading levels. These results are used to train the ML classifiers so as to learn and predict the security state for any operating condition in real-time. The next step in the process of monitoring is, if the system is not perfectly secure, then a corrective action has to be initiated according to the level of insecurity. Overloaded line or lines have to be identified, and accordingly the generations must be adjusted so as to decrease the overload in the line.

446

N. Srilatha et al.

The whole process of assessing and enhancing of system security is depicted in the prediction model presented in Fig. 1. Part-1 of the model assesses security in two different classification modules parallelly. One of them, Static Security Index Prediction Module (SSIP module) classifies and predicts the system state in terms of SSI value (numerical value), while the other module, Security Class Prediction Module (SCLP module) predicts in terms of the class of the security state (nominal value). These two modules work parallelly, and this redundancy is required in order to ensure that the predicted security state has a minimum chance of misinterpretation. Except for the extra training that is done offline, both these modules require the same inputs and the same processing time. Voltage magnitudes and angles of all buses, loads and generation values, and MVA flows of all lines are considered as inputs to these two modules. Part-2 of the model predicts the amount of rescheduling power required to reduce the overload, i.e., the corrective control action required in case of any type of insecurity. If the system is completely secure, then the value of rescheduling power is zero. It considers the SSI value predicted from Part 1 of the model as one of the inputs along with all line MVA flows. This module, Generator Rescheduling Power Prediction module (RPP module) is trained to predict the amount of power that has to be reduced on GD group generators and increased on GI group generators. RPP module is followed by a power calculation block. This block takes the input as the predicted rescheduling power from the module and allots this power economically to GI group generators, according to the concept of RED. Hence the output of this block is the CGS vector, that represents the new generation schedules of all generators in the system. The new generation values, when applied to the practical system, aim in reducing the overloads, thus enhancing the system security. Each of the three prediction modules (Security Class, SSI value and Rescheduling Power) have basically the same architecture, except that the data to be processed is different. The following steps are involved in all these modules. Step 1: Data Collection and Pre-processing Step 2: Feature Selection Step 3: Classification and Prediction The data from the state estimator, i.e., the bus voltage magnitudes and angles, load and generation values and line MVA flows are collected and normalized. All the inputs that are considered for training of the classifier from the input features of the system. Not all these features have the same effect on the security state of the system. Of all the various bus and branch features obtained, only few of them actually decide the system security state. Rest of the features are almost constant for different operating conditions. Hence, feature selection is performed. Correlation-based Feature Subset Selection method is used here. A subset of attributes is evaluated by CFS method for its suitability in terms of individual predictive ability. It also checks for the amount of redundancy between these attributes. Those subsets with a high value of correlation with the class attribute, and a low value of intercorrelation are chosen [11].

Machine Learning Based Prediction Model

447

4 Case Study A classifier purpose is to enable the model with automatic learning. It can predict the class if a new instance is provided to it. There are many classifier algorithms available in machine learning like Naive Bayes Classifier (NB), Support Vector Machines (SVM), Decision Trees (DT), Random Forest (RF), etc. Decision Trees are widely used ML algorithms owing to their simplicity and performance. It is a flowchart like structure, that consists of nodes, branches and leaves. It starts with the root node. Each node represents test on each attribute. The outcome of this test (true or false) represents a branch of the node. These branches are further divided into other branches with the help of other nodes. These finally terminate at leaf nodes, that represent class value. The path through root node to leaf node is the representation of classification rule. Random Forests are an extension of DT and are more robust. This work considers the use of RF for classification and prediction purposes. Random Forest (RF) is a very easy-to-use algorithm. It is an ensemble of multiple decision trees. Bagging (Bootstrap Aggregating) is generally used for the training of these trees. Figure 2 shows the RF that is developed with multiple trees. RF adds additional randomness to the model, while growing trees. It searches for the best feature from a random subset of features instead of searching for the most important feature while splitting a node. The result of this is wide diversity that enables the model to be better in terms of classification and ruggedness.

Fig. 1. ML based prediction model layout

Fig. 2. Model of Random Forest with ‘B’ trees

The IEEE 39-node New England system is chosen as the case study. It consists of 10 generators and 29 load nodes, represents 345 kV transmission system. The required data for training and testing the prediction model is generated offline for all credible contingencies using extensive calculations. The load is varied from 80% to 115% of the nominal values at a constant power factor, in small steps, considering the load curve of this particular system.

448

N. Srilatha et al.

System security is first monitored using part-1 of the model. Feature selection is initially performed on the data set to extract important attributes that contribute to the output. Table 2 gives the picture the quantity of reduction in dimensions of the input dataset. It is observed that the feature selection process results in considerably smaller number of attributes for learning and prediction purposes. Table 2. Classification metrics SSIP module SCLP module RPP module Number of instances 313 313 313 Number of attributes 174 174 48 Number of attributes after feature selection 21 6 7 % Reduction in dimensionality 87.9 96.5 85.5

For the security enhancement part, RPP module helps in predicting the rescheduling power required in case of emergency. Since the expected output of this block is the quantity of rescheduling power that is to be shared among the GI group of generators, this output is categorized into twelve classes, starting from C1 through C12. Different range of rescheduling powers are categorized under each class. The classifier is followed by a power calculation block (Fig. 1). In this block, rescheduling power that is the predicted in the RPP module is shared between generators on the basis of RED. This power is divided between GD group of generators in the ratio of their contributions. So, generation is decreased on this set of generators. To maintain the power balance, the same amount of power is shared among the GI group of generators in the ratio of their DLG values as mentioned in [12]. This group of generators witness an increase in their output generations. Moreover, the GI group generators are chosen to make the rescheduling economical too. All these calculations are done offline for all credible contingencies, and accordingly incorporated in this block. Hence the output of this block is the CGS vector, that represents the new schedules of the generators. In order to assess the performance of the designed classifier and prediction models, results of some random samples are drawn from the testing part of the whole set, and corresponding values are presented in Table 3. The module output is presented across the actual output for comparison. The predicted output obtained from SSIP module is directly given to RPP module as one of the inputs. The predicted output of the SCLP module justifies the correctness of SSIP module. Finally, the RPP module predicts the rescheduling power of the generators in case if enhancement of security is necessary. The power calculation block present after the RPP module calculates change in the generation (CGS vector) required for all the generators in the system. Table 4 gives a picture of change in generations of considered random samples. It represents the change in system security level before and after the application of this ML based prediction model. After the application of this model, flows in the overloaded lines are considerably reduced to normal values (below 100% loading) and the

Machine Learning Based Prediction Model

449

cases which are highly insecure are clearly indicated. Column 4 in this table indicates the control vector, known as CGS vector that represents the rescheduled generation values for the corresponding operating condition. Table 3. System performance after implementing ML based Prediction model Random samples 1 2 3 4 5 Classification accuracy

SSIP module output 2.71 17.59 3.44 0.234 7.50 0.86

Actual output 5.66 20 3.43 0.00 7.06

SCLP module output C B B A C 0.89

Actual output C D B A C

RPP module output C3 C11 C5 C1 C6 0.88

Actual output C3 C12 C5 C1 C6

Table 4. System performance after implementing ML based Prediction model Random samples

Generations before control

Flow CGS vector before control L 13–14 {2.81, 6.85, 5.08, 6.52, 141.4% 5.08, 6.87, 5.80, 6.55, 8.65, 12.00}

1

{2.81, 6.85, 7.18, 6.52, 5.08, 6.87, 5.80, 5.64, 8.65, 10.81}

2 3

Highly Insecure (L16–19 outage) {2.25, 3.49, 5.15, 6.18, {2.25, 5.79, 6.14, 6.18, L 5–6 134% 4.99, 6.31, 5.44, 5.71, 4.99, 6.31, 5.44, 5.29, 8.36, 12.00} 8.36, 9.12}

4 5

Secure; No rescheduling (L 28–29 outage) {2.81, 6.85, 7.18, 6.52, L 5–6 {2.81, 5.34, 5.14, 6.52, 5.08, 6.87, 5.80, 5.64, 153.2% 6.09, 7.00, 5.80, 7.00, 8.65, 10.81} 8.65, 12.00}

Flow after control

L 13–14 99.8% (DG32 = −2.1; DG37 = +0.91; DG39 = +1.20) L 5–6 86.4% (DG31 = −2.3; DG32 = −0.99; DG37 = +0.42; DG39 = +2.89) L 5–6 86.4% (DG31 = −1.5; DG32 = −2.1; DG34 = +1.02; DG35 = +0.13; DG37 = +1.36; DG39 = +1.20)

450

N. Srilatha et al.

5 Conclusion In this paper, machine learning based prediction model is used for both security assessment and enhancement of the IEEE 39-bus New England system. The assessment module predicts the security class of the system in terms of SSI value. This is very much required to know the present state of the system. In the case of any insecurity, this enhancement module presents with the CGS vector for improving security economically. Decision Tree based Random Forest classifier using ensemble methods like Bagging proves to be performing better than other methods. The results clearly indicate the designed model is accurate enough for the purpose and aids the operator in the control centre in case of emergencies in real time.

References 1. Dy-Liacco TE (1997) Enhancing power system security control. IEEE Comput Appl Power 10(3):38–41 2. Schilling MT, Souza JCS, Alves da Silva AP, Do Coutto Filho MB (2001) Power systems reliability evaluation using neural networks. Int J Eng Intell Syst Electr Eng Commun 9 (4):219–226 3. Niebur D, Germond A (1992) Power system static security assessment using the Kohonen neural network classifier. IEEE Trans Power Syst 7(2):865–872 4. Kalyani S, Shanti Swarup K (2011) Classification and assessment of power system security using multiclass SVM. IEEE Trans Syst Man Cybern Part C Appl Rev 5. Wehenkel L (1997) Machine-learning approaches to power-system security assessment. IEEE Expert 12(5):60–72 6. Shahidehpour M, Wang Y (2003) Communication and control in electric power systems. Wiley, New York 7. Wehenkel L (1999) Emergency control and its strategies. In: Proceedings of 13th power system computer conference, Trondheim, Norway, June 1999, pp 35–48 8. He M, Zhang J, Vittal V (2013) Robust online dynamic security assessment using adaptive ensemble decision-tree learning. IEEE Trans Power Syst 28(4):4089–4098 9. Xu Y, Dong ZY, Zhao JH, Zhang P, Wong KP (2012) A reliable intelligent system for realtime dynamic security assessment of power systems. IEEE Trans Power Syst 27(3):1253– 1263 10. Liu C et al (2014) A systematic approach for dynamic security assessment and the corresponding preventive control scheme based on decision trees. IEEE Trans Power Syst 29 (2):717–730 11. Hall MA (1998) Correlation-based feature subset selection for machine learning, Hamilton, New Zealand 12. Yesuratnam G, Srilatha N (2010) An expert system approach of congestion management for security and economy oriented power system operation. In: 9th IEEE international power and energy conference IPEC 2010 13. Sun H et al (2016) Automatic learning of fine operating rules for online power system security control. IEEE Trans Neural Netw Learn Syst 27(8) 14. Oliveira WD et al (2017) Power system security assessment for multiple contingencies using multiway decision tree. Electr Power Syst Res 148 15. Sekhar P et al (2016) An online power system static security assessment module using multilayer perceptron and radial basis function network. Electr Power Energy Syst 76

In-depth Automation and Structural Analysis to Enhance the Power Distribution Reliability of an Urban Neighbourhood Ranjith Kumar Mittapelli1, E. Tharun Sai2(&), and Navya Pragathi3 1

National Institute of Technology, Rourkela, India [email protected] 2 National Institute of Technology, Agartala, India [email protected] 3 National Institute of Technology, Warangal, India [email protected]

Abstract. Unreliable power supply is one of the many root causes for an inefficient power grid and a modest economic growth. The paper reflects upon the reliability of many network types available for a simple urban household. The assessment of reliability is based on the calculation of average annual power outage time and corresponding economic losses for networks with varying levels of automation and structures. The results of the studies are discussed and salient advantages of partial automation over complete automation is demonstrated. Keywords: Automation
Remote terminal units
Power grid
Reliability assessment
Customer Damage Function (CDF)
Extrapolation and interpolation techniques
Economic cost (ECOST) Pole mount switch




1 Introduction Power Distribution stands at the heart of upliftment in myriad sectors and henceforth the country as a whole, but unfortunately about 25% of energy produced in India is lost during power distribution [1]. This energy if harnessed will not only improve the efficiency of the system, it would inevitably reduce the power tariff for the customers, more importantly, enhance reliability and the quality of power at user’s end. The undertaking to improvise the reliability and efficiency of Power Distribution in the unplanned urban layout of India leaves us with many obstacles, two important of them are Most Cities of our nation didn’t stem in a planned layout, making it cumbersome for the distribution network to incorporate different overhead structures like SF6 structure [2], T-shape, and many other efficient structures. The network is highly complex and interconnected leaving scope to power theft that may include unauthorized connections, bypassing the energy meter etc. these malpractices result in decrements of power quality. Indicating that an efficient power grid should certainly possess a check over the amount and location of power connections where electricity is drawn from the grid. © Springer Nature Switzerland AG 2020 S. C. Satapathy et al. (Eds.): ICETE 2019, LAIS 4, pp. 451–461, 2020. https://doi.org/10.1007/978-3-030-24318-0_54

452

R. K. Mittapelli et al.

The Paper henceforth abides these notorious constrains while addressing them. One of the many ways to tackle these problems is by adaptation of Automation. The technique of Automation involves the use of RTUs, the Remote Terminal Unit are catering automation in an efficient manner across the globe. Automation in its basic form consists of RTU’S (remote terminal units) placed at “T” junctions (440 V and 11 kV line intersection), that essentially translates into a couple of million RTUs working together in a humongous network, communicating with each other and with their respective control centres at substations while handling various control services like fault location and isolation, feeder switching and service restoration. Henceforth, considering partial automation will reduce the density of RTU’s used and hence the overall economic investment, maintenance and work force but with the expense of efficiency. The paper reflects on the study of substituting complete automation and studying the degree of efficiency the latter.

2 Proposed Method and Techniques 2.1

Reliability Indices

Here we calculate the j (average annual failure rate), r(j) (average outage time), Uj (the Average Annual Outage Time) of different parts of transmission and distribution system. The data is taken from the Roy billington paper [2]. The paper evaluates discrete number of available techniques, but engenders a solution is one of the 38 stimulated reliability cases and five extrapolation and interpolation methods. So, the proposed solution is accurate with fewer assumptions. A standard feeder from the Roy Billington book [2] is considered with the above configuration as shown in the Table 1. The sample calculation of reliability indices is shown in the Table 2.

Table 1. Basic information about feeder

Automation and Structural Analysis of Power Distribution

453

Table 2. Sample calculation of reliability indices

2.2

Finding CDF Values

The CDF (Customer Damage function), the economic loss due to the power outage for a particular segment of the customer. The CDF Values readily available in [2] corresponds to power outage intervals of one hour, five hours and thirty minutes. Nevertheless, the use of automation brings the power outage time to five seconds and ten minutes etc. Hence the employment of mathematical formulations like interpolation and extrapolation are inevitable. The components tabulated above contain the specified length, respective failure rates and restoration time of components in grid. The four load points with the number of customers draining power, and the average load connected are also depicted. Each bus is 40% commercial and 60% residential load mix for both energy and peak load. On the 11 kV feeder, fuses, disconnectors, and alternative supply are connected to act and eject the faulty component for servicing, so that the unhealthy part of the feeder can be detached without affecting the remaining load points. Every component’s effect (CDF) on a particular load point varies, affecting the outage time and the ECOST too. The Five mathematical formulations namely, Newton [3], Lagrange [4], Curve Fitting [5], Trend analysis [6], Hybrid used for interpolation and extrapolation along with their result shown in Table 3. Tables 2 and 3 depict cumulative CDF value of all Loads Points for different values of power outage, after employing different types of interpolation and extrapolation methods for and commercial loads respectively. The sole use of single interpolation or extrapolation method doesn’t give results that are practically feasible, as newtons interpolations CDF values (y component) are negative when power outage (x component) tends to zero, which is clearly not a real life situation. Lagrange interpolation on the other hand has almost the same CDF value for 5 h and 10 h of power outage, again not a real life situation. Newton’s extrapolation results almost accurately duplicate the curve fitting method’s outcome, but, it fails to work when the depended variable is roughly stretching to zero. Hence, we need to use LaGrange interpolation for such values of CDF. The hybrid method is synthesized to use LaGrange method of inter-polation wherein

454

R. K. Mittapelli et al. Table 3. CDF values at different duration’s using different methods Commercial CDF Outage time Langrangia 0.0166 0.381 0.0833 0.9192 0.3333 2.969 0.5 4.3562 2 16.7266 5 38.7924 10 101.9317 200 21463.33 Residential CDF Outage time Langrangia 0.0166 0.001 0.0833 0.01516 0.3333 0.093 0.5 0.16606 2 1.51336 5 7.1648 10 19.5001 200 437.6521

Newton 4.074 4.33202 5.3557 6.0914 14.616 41.9357 117.998 31242

Trend −3.8825 −4.7716 −2.08522 −0. 2883 15.8517 48.1317 101.9317 2146.332

Curve fitti 4.074 4.33164 5.3554 6.09124 14.61 41.9544 118.0019 31243.65

Hybrid 0.381 0.9192 2.969 4.3562 14.616 41.9357 117.998 31242

Newton −0.28995 −0.24 −0.0785 0.0047 1.6117 7.086 23.1367 7073.738

Trend −2.4713 0 −1.77503 −1.4075 1.8937 8.4961 19.5001 437.6521

Curve fitti 0.28995 0.24822 −0.0785 0.0475 1.6117 7.37224 23.1629 7073.36

Hybrid 0.001 0.01516 0.093 0.16606 1.6117 7.086 23.1367 7073.738

Newton’s divided difference method deploys negative values, to mitigate negative values of CDF, (a negative CDF Value depicts the case of a profit for the customers without a power connection which clearly is not a real life situation), this process fails when the dependent variable is varying sharply, but Newton’s method can be used here because it doesn’t round off as in floating point arithmetic or Lagrange. ECOST Values using newton, LaGrange and hybrid methods are plotted and shown in the Fig. 1. Tables 2 and 3 depict cumulative CDF value of all Loads Points for different values of power outage, after employing different types of interpolation and extrapolation methods for and commercial loads respectively.

Fig. 1. Convergence of ECOST using different interpolation and extrapolation methods. 1. Newton 2. Lagrange 3. Hybrid

Automation and Structural Analysis of Power Distribution

455

Algorithm to find ECOST: The algorithm was designed and implemented for all the feeders in the network [2]. 1. For each Load Point p connected to the network, obtain kj, rj and Uj for each outage event j contributing to its outage. 2. Evaluate the customer Damage Function(CDF) in $/KW Cjp, for different time periods using different interpolation and extrapolation methods. 3. Evaluate the Cumulative Customer Damage Function in $KW. Depends on the different users, the summation term will come into picture. CCDFjp ¼ XCjp  kj

ð1Þ

4. Evaluate the corresponding Customer Interruption Cost CIC ($/year) [2] due to event j using the following equations. ECOSTjp ¼ CCDFjp  Lp  kj

ð2Þ

Where Lp is the Peak Load Demand 5. Repeat steps 2 and 3 for each outage event contributing to load point p. 6. The Total ECOST is obtained by summation of all ECOSTS at each loadpoint p. ECOSTjp ¼

1 X

ECOSTjp

ð3Þ

n¼1

3 Case Study A total of 38 Network case studies on the existing network is done incorporating, 6 Manual Cases, 28 Automation Cases (12 cases in 33% and 66% Automation, 4 cases In 100% Automation and 4 Loop cases.) of reliability on the existing network to find the optimal technique which is both efficient and less complex. 3.1

Manual Mode

The manual technique, incorporates the use of fuses, disconnectors and alternative supply to the radial system, as shown in Fig. 2, but there is no computer intervention in case of a fault, and the failure thus has to be first identified then rectified manually. The manual technique is categorized into six subcases, and the respective ECOST($/year), is summed over all the load points, are tabulated in Table 4.

456

R. K. Mittapelli et al.

Fig. 2. Structure of radial feeder

Case (1)- No Fuses, No disconnectors, No Alternative Supply Case (2)- Fuses, No disconnectors, No Alternative supply Case (3)- fuses, disconnectors, No Alternative supply Case (4)- fuses, disconnectors, Alternative supply Case (5)- No fuses, disconnectors, Alternative supply Case (6)- No fuses, disconnectors, no Alternative supply.

Table 4. Net ECOST of all the six cases in manual technique, using five mathematical formulation Cost TB Case 1 Newton 306.865 Lagrange 253.208 CF 306.818 Trend 319.87 Hybrid 306.865

3.2

Case 2 207.389 211.316 207.377 210.707 207.389

Case 3 134.389 134.97 134.52 124.046 131.044

Case 4 96.58 95.4343 96.5729 99.9023 96.584

Case 5 135.34 122.329 135.93 79.6546 121.544

Case 6 194.498 194.102 194.847 165.916 185.778

Automation

The method of Automation efficiently mitigates the outages and reliability than the traditional cases. The main constituents of automation are a substation, switching station, ring main unit, pole-mounted switch, feeders, intelligent remote terminal unit (IRTU), Global Positioning System (GPS) and the communication network. – Ring main Unit. The RMU is a factory assembled, a metal enclosed set of switch gear used at the load connection points of a ring-type distribution network. – Pole - mounted switch - The pole mounted switch is mounted on the pole and deploys when stimulated by RTU. – Intelligent Remote Terminal Unit or Remote terminal unit. The RTU gives protocol to the pole mounted switch, fuses, disconnectors and connects to Substation. RTU’s are located along the feeder; they continuously monitor node voltage, current, active power, reactive power, power factor, frequency, etc. Each RTU shares the real-time information with other RTUs in the same node (Group of RTU’s on the same feeder or feeder’s branch), all while every RTU shares information with the

Automation and Structural Analysis of Power Distribution

457

control centre (Advanced Distribution Management). The Advanced Distribution Management (ADM) incorporates the live feed from every RTU, acts as the information warehouse to the ADO (Advanced Distribution operation). The ADO obtains network topology data from the ADM system and directs the RTUs, The ADO is composed of a network of computers, merged with SCADA and IEC 61850 [8]. Automation further segregates into three different levels. Each level of automation is subdivided according to the type and number of equipment in use Case-1 Case-2 Case-3 Case-4

-

Fuses, disconnectors Fuses, disconnectors, Alternative Supply No Fuses, disconnectors, Alternative supply No Fuses, disconnectors, no Alternative supply

33% Automation. The case of 33% Automation deals with single load point automation, wherein an RTU is fixed on a feeder at any load point and is given access to corresponding fuse and a disconnector. 33% automation can be further divided into twelve sub-cases, i.e., the standard four cases as stated above, taken one at a time for one of the three load point being automated. The ECOST of all cases are tabulated in the Table 5. Table 5. Net ECOST of all the four cases, using five mathematical formulation 2 automation Case 1 Newton 134.0355 LaGrange 135.9331 Curve Fitting 134.154 Trend 124.0456 Hybrid 130.3114 3 automation case l Newton 132.481 Lagrange 140.1695 Curve Fitting 132.5429 Trend 124.0456 Hybrid 127.0877 4 automation Case 1 Newton 126.3783 Lagrange 139.5918 Curve Fitting 132.7626 Trend 124.0456 Hybrid 127.5273

Case 2 Case 3 Case 4 95.8783 127.2538 192.0253 97.35995 135.8087 200.8415 96.05827 127.6532 192.2837 79.16754 79.16754 165.9163 90.04279 108.1456 180.6496 case 2 case 3 case 4 93.89482 126.7592 190.0469 100.6335 137.1567 206.2333 94.00016 127.1405 190.2331 80.27281 79.167 165.9163 86.6725 107.1199 176.5466 Case 2 Case 3 Case 4 94.41129 126.3783 185.2536 100.056 140.0723 203.3449 94.53469 126.3656 185.2274 80.64123 121.3783 183.969 87.99131 119.8694 178.7446

458

R. K. Mittapelli et al.

66% Automation. The case of 66% Automation deals with two load point automation, wherein two RTU’s on the feeder are given access to two fuses and two disconnectors. 66% automation can be further divided into twelve sub-cases, i.e., the standard four cases as stated above, taken one at a time for one of the two load point being automated. The ECOST of all cases are tabulated in the Table 6. Table 6. Net ECOST of all the four cases of 66% automation, using five mathematical formulation. 2,4 automation Case 1 Newton 135.1886 LaGrange 140.5546 Curve Fitting 135.2456 Trend 135.9068 Hybrid 131.542 2,3 automation Case 1 Newton 132.1277 LaGrange 141.1323 Curve Fitting 132.1767 Trend 129.9001 Hybrid 126.355 3,4 automation Case 1 Newton 134.7114 LaGrange 144.791 Curve Fitting 134.7234 Trend 125.8761 Hybrid 127.497

Case 2 Case 3 Case 4 94.18253 132.0198 188.6338 101.9816 152.1565 210.0846 94.30065 132.3058 188.7685 91.60838 115.5832 181.2844 86.52597 109.5731 173.6159 Case 2 Case 3 Case 4 93.12267 121.8132 187.5739 104.87 150.6362 212.9731 87.94931 122.0141 187.6699 86.60658 105.025 177.6253 84.32796 96.86252 171.4179 Case 2 Case 3 Case 4 96.48359 123.3677 186.6554 105.2551 146.3997 215.4764 96.55445 123.6253 186.7179 86.60658 105.025 174.4541 87.08187 100.0863 169.513

100% Automation. The case of 100% Automation deals with every load point of automation where RTUs are fixed on every section and is given access to disconnectors. 100% automation can be further divided into twelve sub-cases i.e., the standard four cases as stated above, taken one at a time for one of the three load point being automated. The ECOST of all cases are tabulated in the Table 7.

Automation and Structural Analysis of Power Distribution

459

Table 7. Net ECOST of all the four cases in 100% Automation, using five mathematical formulation 100% automation Case 1 Newton 130.4319 Lagrange 135.5809 Curve Fitting 130.1491 Trend 124.0456 Hybrid 122.8382

Case 2 Case 3 Case 4 90.57894 115.8778 187.2306 111.8023 166.8116 222.2162 90.56572 115.8625 187.2017 79.1674 79.16754 170.4024 79.05274 84.55367 168.2463

Loop. Looping the radial system allows power to be supplied from both ends reducing the ECOST by a fold, as shown in Fig. 3. A lesser number of manipulations needed on the disconnectors and fuses for power to reach load points when compared to radial system. The proximity of every load point to the supply implies less voltage drop. Looping the Radial structure makes the systems ECOST as close as 100% Automation but the cost of making a loop is high because of the use of fuses and circuit breakers at every section of the loop and Unlike Automation, Looping the structure doesn’t provide any check over Power theft and a fast reaction to Faults. The ECOST values are tabulated in Table 8. The looped system is further subdivided into the following cases – Case-1: Disconnectors at the start of the loop, disconnectors at every load point, Alternative supply – Case-2: Circuit breaker at the beginning of the loop, disconnectors, fuses at every load point, Alternative supply. – Case-3: Disconnectors, Fuses and Circuit Breaker at each section of the loop but not at the start. – Case-4: Fuses, Circuit breakers at each section of the loop and at the beginning of the loop.

Fig. 3. Loop distribution system

460

R. K. Mittapelli et al.

Table 8. ECOST of all the four cases of Loop structure with different five mathematical formulation LOOP Newton Lagrange CF Trend Hybrid

3.3

Case 1 96.5847 95.43431 96.79061 99.90227 91.50813

Case 2 72.4098 84.38161 72.4652 74.79833 80.45543

Case 3 84.87585 86.59215 84.89321 88.19325 82.66597

Case 4 75.85014 79.77632 75.83814 79.16754 75.85014

Efficient Solution for All Cases

Criteria for finding the efficient and viable solution, out of the 38 cases scrutinized in the Table 9. The sum of ECOST and initial/infrastructural Cost should be minimum, only then the system can be implemented and maintained [10]. ECOST of the proposed solution should have a significant decrement from the ECOST for traditional technique. The Tables 6, 7, 8 have the values of Cumulative ECOST values of 33%, 66% and 100% automation respectively and by using different mathematical formulation, Clearly case 2 of every technique has the least ECOST because of the use of both fuses and disconnectors. Table 9 lists the cost of all the equipment that has been assumed, in order to find the sum of the ECOST and Cost of equipment that has to be borne for implementing techniques. The sum of ECOST and Cost of equipment should be minimum and only then the system can be implemented and maintained [9, 10]. A significant decrement of ECOST is expected that should outweigh the complexity of the system incorporated. Table 9. Costs of different devices used in the system Equipment Cost per equipment (S) Fuses 26.667 Disconnectors 100 Alternate supply (tie line switch) 26.667 RTU 1666.67 IEC 600 GPS 120 Ethernet 3000

4 Results Performance of the system has increased to the maximum, by 57.44% when 100% automation is implemented. Changing the structure from radial to loop has decreased the ECOST by 50.74% (case 1) and to a maximum decrement by 59.58%(case 4), though case 1 is relatively cheap than case 4. There is no scope of automation in a loop, as the cost of looping itself exceeds the price of 100% automation. The proposed

Automation and Structural Analysis of Power Distribution

461

method of 33% automation (case 2) at load point 3 drew down the economic loss due to power outage per year in the network by 59.58% While the total additional cost needed to integrate the new method into the system, as calculated, was about 1.55 times of the cost for traditional equipment. (all the percentages computed concerning the conventional method’s corresponding value).

5 Conclusion Looping the radial network results in an increase of reliability by 56.69%.100% automation doesn’t cater a vast difference than 66% automation. Out of all the techniques, 33% automation in load point 3 (case 2), is the most appropriate case, as the rest cases bear the high economic burden for equipment or do not decrease the existing ECOST by a significant margin. Concerning the case of a bare feeder (case 1), only use of fuses decreased the ECOST by 32% but by the solitary use of disconnectors the ECOST decreases by 39.5%, because the disconnectors decrease the restoration time of the Sects. 1, 2, 3, 4 and fault dependence between load points. But, the fuses only have an impact on the load point. Hence, disconnectors increase the reliability when compared to fuses.

References 1. 2. 3. 4. 5. 6. 7.

Central Electricity Authority (2016) Transmission and Distribution Losses (T&D Losses) Billington R, Allan RN. Reliability evaluation of engineering systems, vol 1 Liu EH-L. Fundamental Methods of Numerical Extrapolation with Applications Baker GB, Mills TM. Lagrange interpolation divergence Brown AM. A step-by-step guide to non-linear regression analysis of experimental data Anil Kumar P, Shankar J, Ashok Kumar G. Remote Terminal Unit for Smart Distribution Reichl J, Schmidthaler M, Schneider F. The value of supply security: the costs of power outages to Austrian households, firms and the public sector 8. Mohagheghi S, Mousavi M, Stoupis J, Wang Z (2009) Modelling distribution automation system components using IEC 61850 9. Kym Wootton. E source Market Research Reveals that Power Outage cost Businesses Over $27 Billion Annually: Winter Storm Jonas Makes it worse 10. Reichl J, Schmidthaler M, Schneider F (2013) The value of supply security: the costs of power outages to Austrian households, firms and the public sector

Implementation of Ant-Lion Optimization Algorithm in Energy Management Problem and Comparison P. S. Preetha1,2(&) and Ashok Kusagur3(&)

3

1 VTU, Belagavi, India [email protected] 2 Department of Electrical and Electronics Engineering, Jain Institute of Technology, Davanagere 577003, India Department of Studies in Electrical and Electronics Engineering, UBDT College of Engineering, Davanagere 577004, India [email protected]

Abstract. The Antlion algorithm (ALO) is used here for solution of energy management problem. This algorithm is compared with particle swarm optimization (PSO), genetic algorithm (GA) and BAT algorithms. The need of energy management increases as the power system loses more power due to wrong management of power. The energy management in industrial loads can save more for industry. Due to the recent tariff plans by Indian government makes the energy management mandatory in the industry to save the peak time load. Many management techniques are provided by the researchers. The segregation of loads to fixed, shiftable and uninterruptable load makes the management of load easy but when it is planed hourly it is not easy to schedule. The proper constraints also have to be considered to make the scheduling more practical. So, the problem is formulated and solved using meta-heuristic techniques in MATLAB. Keywords: Ant-lion optimization
Particle swarm optimization
Genetic algorithm
BAT algorithm
Energy management system

1 Introduction There might be a huge increase in energy demand in 2020 as per [1]. This shows the requirement of power. So, there is a huge need in optimal power scheduling, renewable resources base power generation [2] etc. Based on renewable resources the smart grids play important role. It uses the recent technologies to communicate between the customer and the generator [3]. The advanced metering concepts improves the optimal power delivery to the homes with smart meters. Another important factor is demand side management for improving the power deliver [4]. This concept uses a program which gives incentives to the customer if they are avoiding peak load. It communicates in two-way and it keeps informing the smart grid regarding pricing, load on utility, load shedding and any failures. Each customer © Springer Nature Switzerland AG 2020 S. C. Satapathy et al. (Eds.): ICETE 2019, LAIS 4, pp. 462–469, 2020. https://doi.org/10.1007/978-3-030-24318-0_55

Implementation of Ant-Lion Optimization Algorithm

463

can choose their load timing. The user comfort also should not be spoiled due to the scheduling. So, there is a tradeoff between cost of energy and user comfort [9, 10]. To give solution to the above problem the energy management problem is formulated and solved for industrial load. In this paper the new ant-lion optimization algorithm [8] is implemented to solve this energy management issue and compared with PSO [6], GA [6] and BAT [7] algorithm.

2 Problem Formulation In the energy management recent technology play important role. As the power of each devices are known already, the switching of the load can be done easily. In [5] the embedded scheduler is done for EMS system. Recent smart meters can transfer data two way. So, it transfers between customer and utility. The data can be transferred between, pricing signal from the energy market and the load demand. The architecture of EMS system for an industry is shown in Fig. 1.

Fig. 1. Proposed system

Table 1. Load type Load type Fixed load

Devices Furnace Refrigerator Air conditioner Shiftable Driller Water heater Water pump Fans Uninterruptible Tank cleaner Lights Dryer Mixer

Power in Kwh Hours operated 5 16 4.5 19 7.2 14 3.5 13 2.5 17 0.8 14 0.7 16 3.5 10 1.5 5 1.4 4 1 20

464

P. S. Preetha and A. Kusagur

Table 1 shows the load type and separated it as fixed, shiftable and uninterruptable and 11 loads are considered with its power rating and operating hours. 2.1

Fixed Devices X

24 X

fa 2Fap

t¼1

UðtÞ ¼

! qfa  cfa ðtÞ

ð1Þ

where a = {1, 2,…,n} and t = {1, 2,…, 24}. cfa(t) is ON/OFF state of the appliance in that timeslot. 2.2

Shiftable Devices

VðtÞ ¼

X

24 X

sa 2Sap

t¼1

! qsa  csa ðtÞ

ð2Þ

where csa(t) is the ON/OFF state of the appliance in that hour. Our focus is to minimize the per hour cost of each appliance, as a result the overall cost will be reduced. 2.3

Uninterruptible Devices X

24 X

uia 2UIap

t¼1

WðtÞ ¼

! quia  cuia ðtÞ

ð3Þ

UIap is the set of un-interruptible appliances such that uia
UIap and uia is the power rating of each appliance. The total power consumption W of this category appliances can be calculated using Eq. 3. 2.4

Energy Consumption Model

The total energy consumption of all the devices in each hour can be calculated using the following equation. PT ðtÞ ¼

P

 24 P



qfa  cfa ðtÞ  24  P P þ qsa  csa ðtÞ sa 2Sap t¼1  24  P P þ quia  cuia ðtÞ

fa 2Fap

t¼1

uia 2UIap

t¼1

ð4Þ

Implementation of Ant-Lion Optimization Algorithm

2.5

465

Energy Cost Model

CT ¼

T X

ðPðtÞ  kðtÞÞ

ð5Þ

t¼1

Where k is pricing signal used in our work.

3 Antlion Algorithm The hunting behavior of antlion is formed as algorithm for optimization solutions. Here ants are the population as per the power system term it is control signal. Here, Prey is the objective function. The hunting of objective function with random walks of the control signal, building the control traps, entrapment of population, catching the objective function and rebuilding the trap as known as control and rebuild new population are the different steps involved in it. The algorithm is given below, Step1: Step2: Step3: Step4:

randomly populate the ants and antlions calculate control error or fitness function find the antlion which finds prey faster and name it as elite. while the stop criteria are not satisfied then,

Do (for loop) every ant as below, Roulette wheel selection of antlion Update C and D using below equations, ct ¼

ct I

dt ¼

dt I

Create a random search and normalize it using X ðtÞ ¼ ½0; cumsumð2r ðt1 Þ  1Þ; cumsumð2r ðt2 Þ  1Þ; . . .:cumsumð2r ðtn Þ  1Þ Xit ¼

ðXi  ai ÞX ðdi  ci Þ þ ci ðdit  ai Þ

Update the position of each ant using Anti ¼

RtA þ RtE 2

end for Calculate the fitness of all the control variables Replace the antlion with the corresponding ant if the

466

P. S. Preetha and A. Kusagur

    Antliontj ¼ Antit if f Antit [ f Antliontj Update elite if newer is fitter End the while

4 Results and Discussion The data is collected in local sugar factory in Karnataka. The Karnataka tariff is provided in Table 2. The average cost of electricity taken is Rs. 3.50. The scheduling is done as per the load data available. There are 11 loads considered and it is separated as fixed, shiftable and uninterruptable. Fixed load cannot be scheduled. Shiftable can be scheduled at any time. Uninterruptable can be scheduled without switching it off intermittently.

Table 2. Tariff in Karnataka, India Time of day Tariff 22.00 h to 06.00 h (−) 125 paise per unit 06.00 h to 18.00 h 0 18.00 h to 22.00 h (+) 100 paise per unit

There are four algorithms used for solving the problem. GA, PSO, BAT and ALO. These four algorithms used here and solutions are shown in the figures. Figure 2 shows the convergence of all the four algorithms. Tables 3, 4, 5 and 6 shows the scheduling in that ‘1’ means on and ‘0’ means off. The electricity tariff in India is shown in Fig. 3.

Fig. 2. Compared cost function

Implementation of Ant-Lion Optimization Algorithm

Fig. 3. BESCOM electricity tariff

Table 3. GA Algorithm results Time Hours

Load

0:00 1:00 2:00 3:00 4:00 5:00 6:00 7:00 8:00 9:00 10:00 11:00 12:00 13:00 14:00 15:00 16:00 17:00 18:00 19:00 20:00 21:00 22:00 23:00

Fixed Load

1 1 1 1 1

2 1 1 1 1

Shiftable

1

1

load

type Un-interruptable

3 1 1 1 1

4 1 1 1 1

5 1 1 1 1

6 1 1 1 1

7 1 1 1 1

8 1 1 1 0

9 1 1 1 1

10 1 1 1 1

1

1

1

0

1

0

1

0

11 1 1 1 1

12 1 1 1 1

13 1 1 1 0

14 1 1 1 0

15 1 1 0 0

16 1 1 0 0

17 0 1 0 0

18 0 1 0 0

19 0 1 0 0

20 0 0 0 0

1

1

1

0

1

0

1

0

0

21 0 0 0 0

22 0 0 0 1

23 0 0 0 1

1

1

24 0 0 0 0

1

1

0

1

0

0

1

1

1

1

1

0

1

1

1

0

1

0

1

0

1

1

0

0

1

0

0 1 1 1 1

1 1 1 1 1

0 1 1 1 1

1 1 0 0 1

0 1 0 0 1

0 1 0 0 1

1 1 0 0 1

1 0 0 0 1

1 0 0 0 1

1 0 0 0 1

0 0 0 0 1

0 0 0 0 1

1 0 0 0 1

1 0 0 0 1

1 0 0 0 1

1 0 0 0 1

1 0 0 0 1

1 0 0 0 1

0 0 0 0 1

1 0 0 0 1

1 0 0 0 0

1 1 0 0 0

1 1 1 0 0

0 1 1 1 0

1

Table 4. PSO Algorithm results Time Hours Fixed Load

Load type

Shiftable load

Uninterruptable

0:00 1:00 2:00 3:00 4:00 5:00 6:00 7:00 8:00 9:00 10:00 11:00 12:00 13:00 14:00 15:00 16:00 17:00 18:00 19:00 20:00 21:00 22:00 23:00 1 1 1 1 1 1 0 0 1 1 1 1

2 1 1 1 1 1 1 1 1 1 1 1

3 1 1 1 1 1 0 0 1 1 1 1

4 1 1 1 1 1 0 1 1 0 0 1

5 1 1 1 1 1 1 0 1 0 0 1

6 1 1 1 1 0 1 0 1 0 0 1

7 1 1 1 1 1 1 1 1 0 0 1

8 1 1 1 0 0 1 1 0 0 0 1

9 1 1 1 1 1 1 1 0 0 0 1

10 1 1 1 1 0 0 1 0 0 0 1

11 1 1 1 1 1 1 0 0 0 0 1

12 1 1 1 1 1 1 0 0 0 0 1

13 1 1 1 0 1 1 1 0 0 0 1

14 1 1 1 0 0 0 1 0 0 0 1

15 1 1 0 0 1 1 1 0 0 0 1

16 1 1 0 0 0 0 1 0 0 0 1

17 0 1 0 0 1 1 1 0 0 0 1

18 0 1 0 0 0 0 1 0 0 0 1

19 0 1 0 0 0 1 0 0 0 0 1

20 0 0 0 0 1 1 1 0 0 0 1

21 0 0 0 0 1 0 1 0 0 0 0

22 0 0 0 1 1 0 1 1 0 0 0

23 0 0 0 1 1 1 1 1 1 0 0

24 0 0 0 0 1 0 0 1 1 1 0

Table 5. BAT Algorithm results Time Hours Fixed Load

Load type

Shiftable load

Uninterruptable

0:00 1:00 2:00 3:00 4:00 5:00 6:00 7:00 8:00 9:00 10:00 11:00 12:00 13:00 14:00 15:00 16:00 17:00 18:00 19:00 20:00 21:00 22:00 23:00 1 1 1 1 1 0 0 1 1 1 1 1

2 1 1 1 1 0 1 1 1 1 1 1

3 1 1 1 1 1 1 0 1 1 1 1

4 1 1 1 1 1 1 1 1 0 0 1

5 1 1 1 1 1 1 1 1 0 0 1

6 1 1 1 1 1 0 1 1 0 0 1

7 1 1 1 1 1 0 1 1 0 0 1

8 1 1 1 0 0 1 0 0 0 0 1

9 1 1 1 1 1 1 1 0 0 0 1

10 1 1 1 1 0 0 1 0 0 0 1

11 1 1 1 0 0 0 0 0 0 0 1

12 1 1 1 0 1 0 0 0 0 0 1

13 1 1 1 0 1 0 0 0 0 0 1

14 1 1 1 1 1 1 1 0 0 0 1

15 1 1 0 0 1 0 1 0 0 0 1

16 1 1 0 0 0 1 1 0 0 0 1

17 0 1 0 0 1 1 1 0 0 0 1

18 0 1 0 1 0 1 0 0 0 0 1

19 0 1 0 0 1 1 1 0 0 0 1

20 0 0 0 0 1 1 1 0 0 0 1

21 0 0 0 0 1 0 0 0 0 0 0

22 0 0 0 0 1 1 1 1 0 0 0

23 0 0 0 1 1 1 1 1 1 0 0

24 0 0 0 1 1 0 0 1 1 1 0

467

468

P. S. Preetha and A. Kusagur Table 6. ALO Algorithm results Time Hours Fixed Load

Load

Shiftable load

type

Uninterruptable

0:00 1:00 2:00 3:00 4:00 5:00 6:00 7:00 8:00 9:00 10:00 11:00 12:00 13:00 14:00 15:00 16:00 17:00 18:00 19:00 20:00 21:00 22:00 23:00 1 1 1 1 1 0 0 1 1 1 1 1

2 1 1 1 1 0 1 1 1 1 1 1

3 1 1 1 1 1 1 0 1 1 1 1

4 1 1 1 1 1 1 1 1 0 0 1

5 1 1 1 1 1 1 1 1 0 0 1

6 1 1 1 1 1 0 1 1 0 0 1

7 1 1 1 1 1 0 1 1 0 0 1

8 1 1 1 0 0 1 0 0 0 0 1

9 1 1 1 1 1 1 1 0 0 0 1

10 1 1 1 1 0 0 1 0 0 0 1

11 1 1 1 0 0 0 0 0 0 0 1

12 1 1 1 0 1 0 0 0 0 0 1

13 1 1 1 0 1 0 0 0 0 0 1

14 1 1 1 1 1 1 1 0 0 0 1

15 1 1 0 0 1 0 1 0 0 0 1

16 1 1 0 0 0 1 1 0 0 0 1

17 0 1 0 0 1 1 1 0 0 0 1

18 0 1 0 1 0 1 0 0 0 0 1

19 0 1 0 0 1 1 1 0 0 0 1

20 0 0 0 0 1 1 1 0 0 0 1

21 0 0 0 0 1 0 0 0 0 0 0

22 0 0 0 0 1 1 1 1 0 0 0

23 0 0 0 1 1 1 1 1 1 0 0

24 0 0 0 1 1 0 0 1 1 1 0

Table 7. Comparison Algorithms GA PSO BAT ALO

Cost in Rs. 1344.5 1344.45 1343 1342.97

Time in Sec 15.293922 16.32405 15.910565 12.394

Table 7 shows the final cost value with execution time. It shows that ALO algorithm shows the best result with lesser cost and time.

5 Conclusion The industrial loads are considered here for energy management problem. The loads are separated as different types by considering the requirement of the industries like fixed, shiftable and uninterruptable. Then the cost formula is formulated as objective function and the type of load are considered as constraints. The problem is solved by four different algorithms and the results are graphed and tabulated. Finally, the Ant-lion algorithm is identified as solution for this energy management problem.

References 1. Deploying a smarter grid through cable solutions and services (2010). http://www.nexans. com/Corporate/2010/WHITE-PAPERSMART-GRIDS-(2010).pdf. Accessed 31 Jan 2016 2. Guo Y, Pan M, Fang Y (2012) Optimal power management of residential customers in the smart grid. IEEE Trans Parallel Distrib Syst 23(9):1593–1606 3. Agnetis A, de Pascale G, Detti P, Vicino A (2013) Load scheduling for household energy consumption optimization. IEEE Trans Smart Grid 4(4):2364–2373 4. Nilsson H (1994) The many faces of demand-side management. Power Eng J 8(5):207–210 5. Logenthiran T, Srinivasan D, Shun TZ (2012) Demand side management in smart grid using heuristic optimization. IEEE Trans Smart Grid 3(3):1244–1252 6. Rao SS (2009) Engineering optimization: theory and practice, 4th edn. Wiley, Hoboken

Implementation of Ant-Lion Optimization Algorithm

469

7. Yang XS (2010) A new metaheuristic bat-inspired algorithm. In: Nature inspired cooperative strategies for optimization (NISCO 2010). Studies in computational intelligence, vol 284, pp 65–74 8. Fathy A, Abdelaziz AY (2018) Single and multi-objective operation management of microgrid using krill herd optimization and ant lion optimizer algorithms. Int J Energy Environ Eng 9:257 9. Manjunath TG, Kusagur A (2018) Robust fault detection of multilevel inverter using optimized radial basis function based artificial neural network in renewable energy power generation application. Int J Comput Appl 180(48):8–15 10. Manjunath TG, Kusagur A (2018) Analysis of different metaheuristics method in intelligent fault detection of multilevel inverter with photovoltaic power generation source. Int J Power Electron Drive Syst 9(3):1214–1222

Modulated Frequency Triangular Carrier Based Space Vector PWM Technique for Spreading Induction Motor Acoustic Noise Spectrum Sadanandam Perumandla1,2(&), Poonam Upadhyay3, A. Jayalaxmi4, and Jaya Prakash Nasam2 1

3

EEE Department, JNTUH, Hyderabad, Telangana, India [email protected] 2 EEE Department, Vaagdevi College of Engineering, Warangal, Telangana, India [email protected] EEE Department, VNRVJIET, Hyderabad, Telangana, India [email protected] 4 Center for Energy Studies, EEE Department, JNTUH, Hyderabad, Telangana, India [email protected]

Abstract. To spread the motor acoustic noise, various PWM techniques have been proposed for Voltage Source Inverter (VSI) fed Induction Motor Drives. Random PWM techniques are the most useful techniques to reduce motor acoustic noise. It is difficult for real-time implementation, unpredictable switching losses and the motor suffers from shaft torque dynamics due to adventitious excitation of natural frequencies. The Constant Frequency Triangular Carrier (CFTC) based Space Vector PWM technique will inject switching frequency harmonics into the Variable Frequency Drive, where the harmonics are concentrated at integral multiples of switching frequency which generate audible acoustic noise. To spread the audible harmonic spectrum, this paper presents a Modulated Frequency Triangular Carrier (MFTC) based SVPWM technique, in which the carrier frequency is modulated accordingly. Both the CFTC and the proposed MFTC based PWM techniques are simulated using MATLAB/Simulink. The experiments are conducted on a low-cost TMS320F28379D controller. The proposed MFTC based PWM technique shows better performance in spreading the motor acoustic noise spectrum as compared to the CFTC based SVPWM technique. Keywords: VSI


SVPWM
CFTC
MFTC
VFD
R - PWM
THD

1 Introduction For variable speed applications, DC Motor are more preferable due to the independent control of field and armature voltages for above and below rated speeds, but these motors are not recommended for applications like underwater, chemical and explosive © Springer Nature Switzerland AG 2020 S. C. Satapathy et al. (Eds.): ICETE 2019, LAIS 4, pp. 470–480, 2020. https://doi.org/10.1007/978-3-030-24318-0_56

Modulated Frequency Triangular Carrier Based Space Vector PWM Technique

471

environments. The advancements in the power electronics technology makes the control of Induction Motor simple. The decoupling model of induction motor with the voltage source inverter will give the performance which is nearly similar to the performance of DC Motor. When the induction motor is fed by a pure sinusoidal supply, it will not generate any harmonics but for variable speed control applications the supply frequency is to be controlled in proportional to the supply voltage which is possible with voltage sourced inverter only. When the induction motor is fed by VSI generates harmonics of integral multiples of switching frequency which will create acoustic noise. The reason behind this is the non sinusoidal nature of stator voltages and currents. Noise of Electrical Machines [1–3] is characterized by rapid changes in air pressure. The cause behind this is the vibration of machine parts and aerodynamics of moving parts of the machine. The frequencies of the noise generated due to the vibration and moving parts of the machine lies in lower band of the audible frequency. This noise is continued in VSI fed drives because they are independent to switching frequency. The noise due to non sinusoidal currents are discussed in this paper, it depends on the switching frequency of the VSI. The human ear can perceive sound waves of sufficient intensity and frequencies ranging from 20 Hz to 20 kHz. The switching frequency noise annoyance is dominant over mechanical and aerodynamic noise. This noise can be reduced by reducing the switching frequency but this method will generate low frequency harmonics which causes over heating of stator winding’s thus increasing torque and current ripples. Various PWM schemes [4, 5] are proposed to eliminate low frequency PWM noise and the high frequency PWM noise can be neglected due to lower magnetic forces. The Random PWM (RPWM) techniques are proposed, in which the magnitude of modulating wave or the position of the pulse is changed randomly [6]. Number of papers are published on RPWM to acquire the voltage continuously and the harmonic part is reduced significantly, which in turn reduces the motor acoustic noise. The problem with the RPWM techniques is its implementation for closed loop applications, unpredictable switching losses and shaft torque dynamics. The Constant Frequency Triangular Carrier based SVPWM technique is proposed [7, 8] and [11]. In this method, the harmonics are concentrated at integer multiples of switching frequency.

Diode Recfier

+

VSI

3- phase AC Supply

ib

DC Gen

IM

C

ic

Vd

INVERTER PWM Signals

Noise Measuring Unit

Mechanical Coupling

Resisve Load

ia

-

Field

+ Excitaon MATLAB/Simulink Soware interface

TMS320F28379D

Controller

Fig. 1. Block diagram representation of the controlled VSI fed Induction Motor drive

472

S. Perumandla et al.

Figure 1 represent the block diagram representation of the controlled VSI fed Induction Motor Drive. The input 3-phase rectifier will convert the available AC into DC. This DC voltage is used as a DC Bus for inversion by using VSI, which will feed the power to Induction Motor. The drive is controlled under open loop constant V/f mode, and the mathematical model of Induction Motor is considered in the reference [12]. To reduce the harmonics which are present at integer multiples of switching frequency and to spread the acoustic noise spectrum, this paper presents a Modulated Frequency Triangular Carrier based SVPWM technique. In this method, the carrier frequency is modified in a different manner as stated in the reference [10, 11, 14]. The carrier signal is modified according to the noise spectrum in CFTC based SVPWM, thus maintaining the average switching frequency same as that of CFTC based SVPWM technique. This paper is organized as, Sect. 1 Introduction as discussed, Sect. 2 describes the Motor Acoustic Noise and its Analysis, Sect. 3 discusses the CFTC and MFTC based PWM techniques, Sect. 4 explains about simulation and experimental results and Sect. 5 Conclusions.

2 Motor Acoustic Noise and Its Analysis In the motor drive VSI injects current harmonics of significant magnitudes into the motor which leads to the generation of acoustic noise where its frequency is in the range of carrier frequency and its integral multiples. The impact of harmonic currents on magnetic force wave and the acoustic noise are studied in [9]. The influence of phase current harmonics on magnetic force and acoustic noise is researched [11] and is presented here for the proposed PWM technique. The stator current contains fundamental ‘i1’ and harmonic currents ‘in’ which are defined as, i1 ¼ I1 cosðw1 t  /1 Þ

ð1Þ

in ¼ In cosðnwn t  /n Þ

ð2Þ

Where ‘I1’ and ‘In’ are the maximum values of fundamental and nth harmonic currents respectively and ‘u1’ and ‘un’ are the phase angles of fundamental and nth harmonic currents with respect to an arbitrary reference. When these currents pass through the stator winding a magnetic field is produced due to i1 and in which is expressed in Eq. (3). f ðhÞ ¼ F1 ðh; tÞ þ Fn ðh; tÞ ¼ F1 cosðph  w1 t  /1 Þ þ

X

Fn cosðph  wn t  /n Þ

ð3Þ

n

Where ‘h’ is the spatial angle in the air gap, ‘t’ is the time and ‘p’ is the number of pair of poles. ‘F1’ and ‘Fn’ are the maximum values of fundamental and nth harmonic mmf. The air gap flux density can be expressed in terms of air gap mmf and air gap

Modulated Frequency Triangular Carrier Based Space Vector PWM Technique

473

permeance. In general, the air gap flux density consists of numerous sinusoidal flux density waves is expressed as, X bðh; tÞ ¼ B1 cosðnh  w1 t  /1 Þ þ Bn cosðnh  wn t  /n Þ ð4Þ n

Where, ‘B1’ and ‘Bn’ are the maximum values of fundamental and nth harmonic flux density waves. These waves produce magnetic force waves are calculated by using Maxwell’s equations which is given in (5). pðh; tÞ ¼

b2 ðh; tÞ 2l0

ð5Þ

Where ‘µ0’ is the permeability of air. By substituting for b(h, t) from (4) and neglecting squared terms of harmonics, the force wave in the switching frequency range can be shown as the interaction of fundamental and harmonic flux density waves. pðh; tÞ ¼ pðh; tÞ ¼

X

X 1 B1 cosðph  w1 t  /1 Þ þ Bn cosðph  wn t  /n Þ l0 n

ð6Þ

Pn f½cosð2ph  ðwn þ w1 Þt  ð/n þ /1 ÞÞ þ ½cosððwn  w1 Þt þ ð/n þ /1 ÞÞg ð7Þ

n

Here, (wn − w1) and (/n − /1), respectively, indicate the relative angular frequency and relative phase angle of nth harmonic current with fundamental current. Furthermore, Eq. (7) can be compared to the standard force wave expression shown as follows: pðh; tÞ ¼

X

Pm cosðmh  wm t  /m Þ

ð8Þ

m

where ‘m’ is the mode order of the force wave; ‘wn’ is the angular frequency; ‘Pm’ and ‘/m’ are the amplitude and phase angle respectively of the mth order force wave. In fact wm = (wn − w1), if wn and w1 are rotating in the same direction; wm = (wn + w1) if both are rotating in the opposite direction.

3 CFTC and MFTC Based PWM Techniques The Space Vector PWM (SVPWM) technique [15] utilizes DC bus voltage which is 15% more than that of Sinusoidal PWM (SPWM) technique. The modulation index is varied in such a way to maintain motor V/f ratio constant. The same modulated

474

S. Perumandla et al.

waveform is used for both the PWM techniques. In the space vector approach, the commanded voltage is specified either as three phase voltage references (Va,Vb and Vc) or as a voltage space vector (VREF). When VREF is in sector 1, the active states applied are + − − (1) and + + − (2). The corresponding dwell times ‘T1’ and ‘T2’ are calculated which are proportional to (Va − Vb) and (Vb − Vc) respectively, as explained in the reference. 3.1

CFTC Based SVPWM

In the carrier based PWM techniques, generally its switching frequency is maintained at multiple of fundamental frequency to maintain symmetry. In this technique, the modulated signals are compared with the triangular carrier whose frequency is maintained constant. Based on the comparison, switching pulses are generated for controlling the VSI. Figure 2(a) represents the scheme of PWM generation. Due to constant carrier frequency, the harmonic energy is concentrated at integral multiples of switching frequency with larger magnitudes. These peaks will create acoustic noise which is uncomfortable to human ear. 3.2

MFTC Based SVPWM

In this technique, the carrier wave frequency is modified to spread the acoustic noise spectrum, in which the average switching frequency is maintained as that of CFTC based SVPWM. In a sector, the carrier frequency is varied in a stepwise manner for every 20° with two equal incremental steps, which is repeated for every sector (60°). Starting of a sector the carrier frequency is kept at lower value and is increased with a pre-calculated value after 20°, the same is repeated for next 20°, after completion of a sector, again frequency is brought to the lowest frequency and this variation is done in a recursive manner. The proposed technique achieves reduction in magnitudes which appears in the CFTC based SVPWM technique and also spreads the noise spectrum over a wide range. As the average frequency is maintained as same as that of conventional SVPWM technique, the switching losses, voltage stress across the switches and the thermal stress have not changed much. With this strategy, it is possible to disperse the motor acoustic noise spectrum towards higher frequencies with lower magnitudes and slight change in THD as compared to the existing. Figure 2(b) describes, the generation of PWM signals with the proposed PWM technique.

Modulated Frequency Triangular Carrier Based Space Vector PWM Technique

475

Fig. 2. (a) CFTC based SVPWM technique for PWM generation and the line voltage. (b) MFTC based SVPWM technique for PWM generation and the line voltage.

476

S. Perumandla et al.

4 Results and Discussion The existing and the proposed PWM techniques are simulated by using MATLAB/ Simulink software. These simulations are carried out on 3-phase, 415 V, 50 Hz, 4 kW Squirrel Cage Induction Motor with a DC link voltage of 566 V. The carrier frequency is considered as 5.4 kHz for CFTC based SVPWM and for MFTC SVPWM technique, it is chosen as discussed in Sect. 3 which are 4.5 kHz, 5.4 kHz and 6.3 kHz respectively. The motor is tested under no load and loaded conditions. 4.1

Simulation Results

Figure 2(a) and (b) describes the generation of PWM signals in the existing and the proposed PWM techniques. The carrier frequency modulation is clearly represented in Fig. 2(b). The PWM signals in a leg for upper and lower switches are also shown for both the PWM techniques. Figure 3(a) and (b) represent the full load line current waveform, the corresponding FFT spectrum and line voltages for both the methods. From the simulation results, it is observed that, there is no change in the THD levels. They are almost equal but the spread of the current FFT spectrum is more in the case of MFTC based SVPWM technique. It is clearly observed that, the motor acoustic noise is dispersed for a wide range with lower magnitudes at the integral multiples of the switching frequency as compared to the CFTC based SVPWM technique. The motor can be controlled by varying the modulation index. The dead time effect is neglected for simulation study and is considered for the experimental study.

Fig. 3. (a) Motor line current and its THD for CFTC and MFTC based SVPWM techniques (b) Motor Line voltage with CFTC and MFTC based SVPWM techniques

Modulated Frequency Triangular Carrier Based Space Vector PWM Technique

477

Figure 4(a) and (b) represent the speed, torque and phase current waveforms when the motor is under no load and light load conditions respectively. Figure 4(c) represent the motor speed, torque and phase current, when the load is varied in steps from 0 N-m to 10 N-m. From the results, it is observed that, the motor comes to steady state at less than 0.1 s. During the transition it comes to steady state very fast without any oscillations and overshoots with the proposed SVPWM technique.

Fig. 4. Speed, Torque and phase current of the Motor (a) under no load condition (b) light load of 10 N-m and (c) step change in load from 0 to 10 N-m with MFTC based SVPWM

4.2

Experimental Results

The experiments were conducted on a 3-phase 3-HP, 415 V, 4.4 A, 50 Hz Squirrel Cage Induction Motor. The motor is loaded through a directly coupled DC Generator with resistive load to analyse the dynamic performance of the motor. The control algorithms are implemented on a low cost TMS320F28379D TI micro controller. The MATLAB/ Simulink is used to generate code using add-on embedded coder and the code composer studio. The controller is interfaced with a computer via serial port. For measuring the line current two LA55-P current transducers are used. The drive is controlled in open loop constant V/f mode. The available DAC channels are used to measure and verify the modulated signal frequency and is mapped with the real time clock. The motor acoustic noise is measured by using a low cost noise measuring unit [13]. Figure 5 represent the experimental setup with low cost noise measuring unit and Induction Motor - DC Generator set. The SVPWM based modulated waveform and the carrier frequency modifier are shown in Fig. 6. Figures 7 and 8 represent the line current and the corresponding FFT spectrum when the motor is controlled with CFTC and MFTC based SVPWM techniques. These are captured after reaching the motor current to steady state. The magnitudes of motor current FFT spectrum at integral multiples of switching frequency are sharp and are clearly shown with an arrow marks but the magnitudes of motor current FFT spectrum is dispersed over a wide range and the peaks are reduced with the proposed PWM technique.

478

S. Perumandla et al.

Fig. 5. Experimental setup, low cost noise measuring unit and IM - DC Gen set

5.4 kHz

6.3 kHz

4.5 kHz Carrier Frequency Modifier Modulated Waveform

scale: X-axis 5ms/unit, Y-axis 1V/unit

Fig. 6. DAC Channel output modulated waveform with the carrier frequency modifier

Scale: X-axis 500ms/unit, Y-axis 5A/unit

5.4 kHz

Scale:X-axis2.5 kHz/unit, Y-axis 10 dB/unit 10.8 kHz

16.2 kHz

Fig. 7. Load current and its FFT spectrum for CFTC based SVPWM

Scale: X-axis 500ms/unit, Y-axis 5A/unit

Scale:X-axis2.5 kHz/unit, Y-axis 10 dB/unit

Fig. 8. Load current and its FFT spectrum for MFTC based SVPWM

Modulated Frequency Triangular Carrier Based Space Vector PWM Technique

479

Fig. 9. Motor Acoustic noise Spectrum with CFTC and MFTC based SVPWM technique at full load condition (scale: X axis 5 kHz/unit, Y-axis 10 dB/unit)

The wear and tear of the motor, friction between the bearings and the shaft coupling are not considered while measuring motor acoustic noise because they remain constant irrespective of type of PWM technique and independent of switching frequency. Figure 9 represent the measured motor acoustic noise by using a low cost noise measuring unit. With the proposed PWM technique, the spread in motor acoustic noise spectrum is observed more as compared to the CFTC based SVPWM technique. The peak magnitudes almost disappeared in the noise spectrum. The proposed PWM shows reduced acoustic noise annoyance by dispersing the peak noise energy levels. It is also found that, the stress levels on human ear are comparably reduced and comfortable. From the experimental results, it is observed that, the MFTC based SVPWM technique gives better performance in spreading the motor acoustic noise.

5 Conclusion In this paper, a complete simulation model incorporating the existing CFTC and MFTC based SVPWM techniques are simulated on a 3-phase SCIM. The simulation results show that, the proposed PWM method gives best performance in spreading the motor acoustic noise. The effectiveness of the proposed model is verified by conducting an experiment using a low cost TMS320F28379D TI processor and a low cost noise measuring unit. The experiments were carried out for both the PWM techniques. Experimental results shows that, the increased potentiality of the proposed PWM technique in spreading the motor acoustic noise as compared to the existing CFTC based SVPWM technique. Dynamic performance of the motor is also analyzed and it is observed that it comes to steady state in less than 0.1 s without any damping and overshoots. Moreover, when compared with conventional SVPWM technique the proposed SVPWM technique doesn’t require the look-up table and identification of the sector. The cost of the real time implementation is also reduced with the low cost processor instead of using high end dSPACE, Opal-RT, real time simulators and controllers. Acknowledgments. We would like to thank the University Grants Commission, Govt of India for sanctioning the research grant under Minor Research Project P.NO:4-4/2015-16(MRP/UGC SERO) to carry out this work. We thank the Principal and Management of Vaagdevi College of Engineering (Autonomous), Warangal, Telangana State, India, for their support and encouragement.

480

S. Perumandla et al.

References 1. Janda M, Vitek O, Hajek V (2012) Noise of induction machines, induction motors modelling and control. In: Araújo RE (ed). In Tech. https://doi.org/10.5772/38152 2. Vijayaraghavan P, Krishnan R (1999) Noise in electric machines: A review. IEEE Trans Ind Appl 35(5):1007–1013 3. Belmans RJM, D’Hondt L, Vandenput AJ, Geysen W (1987) Analysis of the audible noise of three phase squirrel cage induction motors supplied by inverters. IEEE Trans Ind Appl IA23(5):842–848 4. Lo WC, Chan CC, Zhu ZQ, Xu L, Howe D, Chau KT (2000) Acoustic noise radiated by PWM controlled induction machine drives. IEEE Trans Ind Electron 47(4):880–889 5. Besnerais JL, Lanfranchi V, Hecquet M, Brochet P (2010) Characterization and reduction of audible magnetic noise due to PWM supply in induction machines. IEEE Trans Ind Electron 57(4):1288–1295 6. Jadeja R, Ved A, Chauhan S (2015) An investigation on the performance of random PWM controlled converters. Eng Technol Appl Sci Res 5(6):876 7. Stemmler H, Eilinger T (1994) Spectral analysis of the sinusoidal PWM with variable switching frequency for noise reduction in inverter-fed induction motors. In: Proceedings of IEEE PESC 1994, Taipei, Taiwan, vol 1, pp 269–277 8. Ruiz-Gonzalez A, Meco-Gutierrez MJ, Perez-Hidalgo F, VargasMerino F, Heredia-Larrubia JR (2010) Reducing acoustic noise radiated by inverter fed induction motors controlled by a new PWM strategy. IEEE Trans Ind Electron 57(1):228–236 9. Holmes DG, Lipo TA (2003) Pulse width modulation for power converters: principles and practice, vol 18. Wiley, Hoboken 10. Binojkumar AC, Saritha B, Narayanan G (2015) Acoustic noise characterization of space vector modulated induction motor drives - an experimental approach. IEEE Trans Ind Electron 62(6):3362–3371 11. Binojkumar AC, Saritha B, Narayanan G (2016) Experimental comparison of conventional and bus-clamping PWM methods based on electrical and acoustic noise spectra of induction motor drives. IEEE Trans Ind Appl 52(5):4061–4073 12. Ratnani PL, Thosar AG (2014) Mathematical modelling of an 3 phase induction motor using MATLAB/Simulink. Int J Mod Eng Res 4(6):62–67 13. Binojkumar AC, Narayanan G (2011) A low-cost system for measurement and spectral analysis of motor acoustic noise. Presented at the national power electron conference, Howra, India, December 2011 14. Bhavsar T, Narayanan G (2009) Harmonic analysis of advanced bus-clamping PWM techniques. IEEE Trans Power Electron 24(10):2347–2352 15. Kim JS, Sul SK (1995) A novel voltage modulation technique of the space vector PWM. In: Conference proceedings of IPEC, pp 742–747

Hardware In Loop Simulation of Advanced Aerospace Vehicle A. Shiva Krishna Prasad1(&) and N. Susheela2 1

2

Research Centre Imarat, DRDO, Hyderabad 500 069, Telangana, India [email protected] Department of Electrical Engineering, University College of Engineering (A), Osmania University, Hyderabad 500 007, Telangana, India [email protected]

Abstract. This paper describes the design methodology of Hardware In Loop Simulation (HILS) for advanced aerospace vehicle. It exploits detailed six degree of freedom (6DOF) model for an asymmetrically configured aerospace vehicle. Criticalities exist in modelling and simulation of the vehicle has been discussed in detail. It explains the nonlinearities present in the actuators of the vehicle and its effect on the vehicle dynamics. Initially simulation is performed by integrating all subsystem models in a single Non Real Time (NRT) simulation test setup. Further HILS is carried out by replacing model with actual hardware in a stepwise manner. Simulation results of the vehicle are presented and also comparative results on actuator nonlinearities are presented. Keywords: Asymmetric vehicle
6DOF model Actuator nonlinearities dead band and backlash


HILS


Nomenclature

ax , ay , az p, q, r U, V, W m S d CN, CS Crm, Cm, Csm Crmd Cmd Csmd Ixx, Iyy, Izz, Ixy, Ixz, Iyz Vxiner, Vyiner, Vziner c Q

= Linear accelerations along the vehicle body axes (Xb, Yb, Zb) respectively, m/s2 = Angular velocities of the vehicle about the body axes, deg/s = Linear velocities of the vehicle in body frame, m/s = Mass of the vehicle, kg = Vehicle cross-sectional area, m2 = Mean diameter of the vehicle, m = Aerodynamic force coefficients = Aerodynamic moment coefficients = Stability derivative of the rolling moment = Stability derivative of the pitching moment = Stability derivative of the yawing moment = Moments of inertia of the vehicle, kg.m2 = Linear velocities of the vehicle in inertial frame, m/s = Distance between centre of gravity location to sensor location, m = Dynamic pressure, N/m2

© Springer Nature Switzerland AG 2020 S. C. Satapathy et al. (Eds.): ICETE 2019, LAIS 4, pp. 481–489, 2020. https://doi.org/10.1007/978-3-030-24318-0_57

482

A. Shiva Krishna Prasad and N. Susheela

1 Introduction Commonly, Aerospace vehicles have two axes of symmetry which is arranged in a cruciform configuration as shown in Fig. 1. These vehicles have control surfaces and/or lifting surfaces at 90 degrees from each other. These vehicles have identical lateral aerodynamic characteristics in pitch and yaw plane. Therefore the pitch and yaw channel autopilots are identical. Due to the symmetry of the vehicle about x-y and x-z plane, the moment of inertia in y-plane, z-plane is same i.e., Iyy = Izz and the product terms of inertia i.e., Ixy, Ixz, Iyz are usually omitted in its 6DOF model. Therefore the cross coupling effects among the roll, pitch and yaw plane of the vehicle are negligible [1].

+ Fig. 1. Cruciform configuration

The rapid advances in propulsion technologies like ramjet, scramjet have changed the airframe from symmetrical configuration to asymmetrical configuration as shown in Fig. 2. These vehicles show the highly nonlinear and non-identical lateral aerodynamic characteristics. Since the moment of inertia is not same in y-plane and z-plane, the cross coupling effect is more predominant in this type of configuration [2].

+ Air Intakes Fig. 2. Asymmetrical configuration

Any Aerospace vehicle need to be flight tested before finalizing its complete design. But there will be many issues with respect to subsystem design, system hardware, software, protocol and communication related issues among the sub systems which cannot be solved in one flight test and it needs lot many flight tests for finalizing the vehicle. This will drastically increases the design and development cost, time, man power etc. This can be overcome by HILS test facility on ground which will bring out maximum issues related to vehicle design and solved subsequently. It leads to finalize the vehicle design with very minimal flight tests which drastically reduce design and development cost, time and man power. Therefore HILS test setup acts as a simulated flight test tool in the design process.

Hardware In Loop Simulation of Advanced Aerospace Vehicle

483

It is a technique that is used in the development and testing of complex real-time systems. It is a real-time simulation test set up which simulates the entire system behaviour with actual hardware and software of the system in closed loop. It is a critical development tool which plays a key role in hardware realization (Design Process).

2 Asymmetric 6DOF Model of Dynamic Vehicle The six degrees of freedom consist of three translations and three rotations, along and about the vehicle (Xb, Yb, Zb) axes [3]. 6DOF equations of motion are used to represent the vehicle dynamics in the body frame. The vehicle dynamics are considered is governed by rigid body mechanics, which means that flexible modes due to aeroelasticity are neglected. The complete matrix of moment of inertia is considered in modelling to cater its asymmetric configuration. The entire cross coupling terms are taken in to consideration while modelling the vehicle. The complete 6DOF equations which have three force equations (Eq. (1)–(3)) and three moment equations (Eq. (4)– (6)) of vehicle therefore are: ðThrust  DragÞ  qVziner þ rVyiner m

ð1Þ

ðCS QS þ CSd QSÞ þ c_r  rVxiner þ pVziner m

ð2Þ

ax ¼ ay ¼

ðCN QS þ CNd QSÞ  cq_  pVyiner þ qVxiner m
ðCRM QSd þ Crmdroll QSd Þ  I_xx p  Izz  Iyy qr p_ ¼ Ixx Iyz ðr 2  q2 Þ þ Ixz ðpq þ r_ Þ  Ixy ðrp  q_ Þ Ixx az ¼

q_ ¼

ðCm QSd þ Cmd QSd Þ  I_yy q  ðIxx  Izz Þpr Iyy  Ixz ðp2  r 2 Þ þ Ixy ðqr þ p_ Þ  Iyz ðpq  r_ Þ Iyy

r_ ¼


ðCsm QSd þ Csmd QSd Þ  I_zz r  Iyy  Ixx pq Izz 2 2 Ixy ðq  p Þ þ Iyz ðrp þ q_ Þ  Ixz ðqr  p_ Þ Izz

ð3Þ

ð4Þ

ð5Þ

ð6Þ

484

A. Shiva Krishna Prasad and N. Susheela

A second order linear actuator model has been considered during initial stage of simulation [4]. Further actuator nonlinearities like dead band, backlash and saturation are incorporated in the model as shown in Fig. 3 [5].

Fig. 3. Actuator model with nonlinearities

3 Hardware In Loop Simulation HILS is a rapid proto test setup which consists of real hardware and software for finalizing the entire aerospace system design and performance evaluation before going for the actual flight tests [6]. The necessities of HILS are• • • •

To weed out the design deficiencies Validating dynamic response of the sub systems Performance evaluation of un modeled dynamics and nonlinearities Validation of On board software, interface and protocol clearance between sub systems • Subsystem design study HILS setup is designed and developed in a stepwise augmentation manner and performance evaluation of various vehicle subsystems such as On board computer, actuator and navigation sensor along with interfaces in flight configuration. This methodology allows ease of analysis and brings out the design related issues and deficiencies if any in systematic manner. The basic block diagram of aerospace vehicle simulation is shown in Fig. 4.

Fig. 4. Block diagram of aerospace vehicle simulation

Hardware In Loop Simulation of Advanced Aerospace Vehicle

485

HILS test setup is established in a step wise augmentation manner as given below. 1. 2. 3. 4. 5. 3.1

Non real time simulation On board computer In Loop Actuator In Loop Sensor In Loop Sensor Actuator In Loop Non Real Time Simulation (NRT)

Initially the NRT simulation test bed is established which comprises of 6DOF aerospace vehicle model (atmospheric model, thrust model, aerodynamics, propulsion etc.), actuator model and sensor model with guidance, control and navigation as shown in Fig. 5. This is a prerequisite step for conducting HILS. It is established in a single PC. Here the complete plant software and application software are developed which are decoupled but functionally cohesive. Microsoft Visual Studio C language in Windows environment is used for the development. This step ascertains preliminary design, control and guidance algorithms including 6 DOF plant model and it becomes the reference for establishing the HILS test bed [7].

Fig. 5. NRT simulation test setup

3.2

On Board Computer In Loop

In this configuration the actual hardware On board computer (OBC) with embedded software is introduced in HILS as shown in Fig. 6. A three loop autopilot is implemented in the OBC to control the vehicle body rates tightly. Power PC or System On Chip based systems are chosen as an OBC based on computational load, space requirements. The same I/O interfaces (Analog to Digital Converters, Digital to Analog Converters, Serial bus etc.) as exists in the vehicle are configured between controller and plant at HILS. Simulation is performed under this environment through which we can validate the On board computer software, its timings, scaling, interfaces throughout the vehicle profile. Hard real time task in On board computer gets thoroughly validated during simulation.

486

A. Shiva Krishna Prasad and N. Susheela

Fig. 6. On board computer In Loop test setup

3.3

Actuator In Loop

This is to validate the vehicle actuation system performance under dynamic conditions. The HILS test bed is augmented with actual hardware Actuators (Electro mechanical actuator) with On board computer and 6DOF model as shown in Fig. 7. Vehicle performance is checked with actual hardware actuators along with its inherent nonlinearities for entire profile.

Fig. 7. Actuator In Loop test setup

3.4

Sensor In Loop

This is to validate Inertial Navigation System’s (INS) hardware and software with trajectory dynamics and the effect of sensor lag and noise on vehicle performance. The real INS rate gyros get excited in the HILS test bed by strapping INS on the 3-Axis Flight Motion Simulator (FMS) as shown in Fig. 8. The pure rates are fed from 6DOF model PC to FMS controller and the sensors mounted on FMS experiences vehicle body rates. The sensed rates from the actual sensors and the computed accelerations from 6DOF model PC are used by the INS navigation processor for navigation to generate positions, velocities and quaternions of the vehicle. During this run INS with sensor electronics, navigation algorithm and its mechanical mounting along with electrical interfaces gets validated in the actual flight configuration. 3.5

Sensor Actuator In Loop

To validate the integrated dynamic performance of INS and real time actuation system, both the hardware’s are introduced along with the On board computer in HILS as shown in Fig. 8. Except the accelerometers, all other hardware’s are excited along with the mission software in HILS test bed (using the Flight Motion Simulator). However these accelerometers are tested in centrifuge and gravitational effects are observed.

Hardware In Loop Simulation of Advanced Aerospace Vehicle

487

Fig. 8. Sensor Actuator In Loop test setup

4 Simulation Results Since the vehicle has high dynamic performance, it is required to solve the 6DOF equations at finer time steps ( 1 through the recursive expressions. 3.3

NMPC Controller

Once the parameters are obtained using Eq. (7), we can design the control scheme for two-link planar manipulator system. To calculate the control actuator torques from controller, assume angular acceleration €hi is constant over one interval of instant t ¼ DðkÞ. We can predict the next instant of angular displacement hi ðk þ 1Þ in terms of measured angular displacement, angular velocity and angular acceleration hm i ðkÞ; _hm ðkÞ; and €hm ðkÞ respectively as, i i 1 €m 2 _m hi ð k þ 1Þ ¼ hm i ðk Þ þ hi ðk ÞD þ hi ðk ÞD 2

ð9Þ

To satisfy Eq. (9), we need to calculate h€i ðkÞ at instant, using reference trajectory R hi ðk þ 1Þ at instant k þ 1 and measured angular displacement hm i ðkÞ, angular velocity m h_ i ðkÞ, as given below, m

n o €hi ðk Þ ¼ 2 hR ðk þ 1Þ  hm ðk Þ  h_ m ðkÞD i i i 2 D

ð10Þ

Now the actuator torques which is control input to two link planar manipulator can be obtained by substituting Eq. (10) in Eqs. (1) and (2) as, T1 ¼

  2 R m _ m ðkÞD  L12 ðhR ðk þ 1Þ  hm ðkÞ ½L h ð k þ 1 Þ  h ð k Þ  h 11 1 1 1 2 2 D2  h_ m ðk ÞDÞ þ a1 þ b 2

1

ð11Þ

Adaptive NMPC Controller for Two-Link Planar Manipulator

T2 ¼

  2 R m _m ½L hR1 ðk þ 1Þ  hm 1 ðk Þ  h1 ðk ÞD  L22 ðh2 ðk þ 1Þ  h2 ðk Þ 2 21 D  h_ m 2 ðk ÞDÞ þ a2 þ b2

525

ð12Þ

Solutions of above two equations give the actuating torques on each link of two link planar manipulator. Thus by completing these three stages we can design the required NMPC controller for two link planar manipulator.

4 Simulation Results In order to investigate the performance of proposed NMPC - OPA estimator, the values of two link planar manipulator are shown Eq. (13). This equation given changes in values of physical values system for 10 s, say mass of first link m1 has modified from 2.21 kg to 1.77 kg after 10 s.  8 9  m2 : 2:21 ! 5:52½kg m1 : 2:21 ! 1:77½kg > >  > > > >  > > l1 : 0:1 ! 0:1:6½m l : 0:2 ! 0:3 ½ m  < = 2   ð13Þ lc1 : 0:145 ! 0:145½m  lc2 : 0:145 ! 0:145½m > > 2  2 > > I I : 0:0582 ! 0:0298 ½ kg:m  : 0:0582 ! 0:0091 ½ kg:m  > > 1 2  > > : ;  l1 : 0:01 ! 0:002 l2 : 0:01 ! 0:030 The unknown parameter vector also varies such that parameters of Eqs. (1) and (2) also be varied. These parameters ; ¼ ½c1 c2 c3 l1 l2 T are initially set to ;ð0Þ ¼ ½ 0:2977 0:1047 0:0641 0:01 00:01 T and final target values after 10 s these parameters are modified to ;ðtf Þ ¼ ½ 1:2339 1:0255 0:1282 0:002 0:03 T and these values shown in Eq. (14). 8 c1 : 0:2977 ! 1:2339 > > > > < c2 : 0:1047 ! 1:0255 c3 : 0:0641 ! 0:1282 > > l : 0:01 ! 0:002 > > : 1 l2 : 0:01 ! 0:03

ð14Þ

The estimated parameters b ; ¼ ½c1 c2 c3 l1 l2 T are shown Figs. 3 and 4 which represents the robustness of estimator. To investigate controller performance consider the control parameter values shown Eq. (15). 8 max s1 ¼ 0:15½sec > > > > smax ¼ 0:15½sec > > < 2 min s1 ¼ 0½sec ¼ 0½sec > smin 2 > > > c1 ¼ 2:5 > > : c2 ¼ 2:5

ð15Þ

526

B. Raja Gopal Redy et al.

The initial value of time constant sold is calculated as average of maximum and i minimum limiting values of time-constant as sold i ¼ 0:75 and sampling time is D ¼ 10 ms. The desired values of two link angular displacements are described by, 

hd1 hd2



 ¼

0:8 cosð6ptÞ 0:8signðsinð3ptÞÞ

 ð16Þ

The variation of each link desired angular displacements and corresponding actual output angular displacements are shown in Fig. 4. The system outputs of link1 and link2 are compared as shown in Fig. 5 and in Fig. 6 control errors are compared with existing results [11] and are evident that proposed method shows better performance.

Estimated Parameters

C1

2 1 0

0

2

4

6

8

10

12

14

16

18

20

0

2

4

6

8

10

12

14

16

18

20

0

2

4

6

8

10

12

14

16

18

20

0

2

4

6

8

10

12

14

16

18

20

0

2

4

6

8

10 Time

12

14

16

18

20

C2

2 1 0

C3

0.2 0.1 0

u1

0.01 0.005 0

0.02

Fig. 4. Estimated parameters

Angular Displacements of link1(Theta1) Comparisions 1 Reference PID NMPC

0.5

Theta1 in Rad

0.01

0

-0.5

-1

0

2

4

6

8

10 Time in Sec

12

14

16

18

20

Error Angular Displacements of link1(Theta1) Comparisions 0.7 PID NMPC

0.6 0.5

Error

u2

0.03

0.4 0.3 0.2 0.1 0

0

2

4

6

8

10 Time in Sec

12

14

16

18

Fig. 5. Angular displacement and error comparisons of link1

20

Adaptive NMPC Controller for Two-Link Planar Manipulator

527

Angular Displacements of link2(Theta2) Comparisions 1.5 Reference PID NMPC

Theta2 in Rad

1 0.5 0 -0.5 -1

0

2

4

6

8

10 Time in Sec

12

14

16

18

20

Error Angular Displacements of link2(Theta2) Comparisions 0.4 PID NMPC

Error

0.3

0.2

0.1

0

0

2

4

6

8

10 Time in Sec

12

14

16

18

20

Fig. 6. Angular displacement and error comparisons of link

5 Conclusions In this paper a new NMPC controller structure with OPA estimator is presented. A three-stage structure has reference input generation stage, parameter estimator stage and controller design stages ensure minimum control error, better parameter estimation. The simulation results show the robust performance of both estimator and controller which are better than existing results.

References 1. Maciejowski JM (2002) Predictive control with constraints. Prentice Hall. ISBN 0201 398230 2. Kwon WH, Pearson AE (1948) On feedback stabilization of time-varying discrete linear systems. IEEE Trans Autom Control 23:479–481 3. Cheng VHL (1979) A direct way to stabilize continuous-time and discrete-time linear timevarying systems. IEEE Trans Autom Control 24:641–643 4. Chen CC, Shaw L (1982) On receding horizon feedback control. Automatica 18:349–352 5. Mayne DQ, Michalska H (1990) Receding horizon control of nonlinear system. IEEE Trans Autom Control 35:814–824 6. Jung S, Wen JT (2004) Nonlinear model predictive control for swing- up of a rotary inverted pendulum. Trans ASME 126:666–673 7. Feng G, Lozano R (1999) Adaptive control system. Newnes. ISBN 07506 39962 8. Henmi T, Ohta T, Deng M, Inoue A (2009) Tracking control of the two-link manipulator using nonlinear model predictive control. In: Proceedings of IEEE international conference on networking, sensing and control, pp 761–766 9. Siciliano B, Khatib O (eds) (2008) Springer handbook of robotics. Springer, New York 10. Henmi T, Deng M, Inoue A (2010) Adaptive control of a two-link planar manipulator using nonlinear model predictive control. In: Proceedings of the 2010 IEEE international conference on mechatronics and automation, 4–7 August 2010, Xi’an, China 11. Mahamood RH, Pedro JO (2011) Hybrid PD/PID controller design for two link flexible manipulators. In: Proceedings 8th Asian control conference (ASCC)

Solving the Complexity of Geometry Friends by Using Artificial Intelligence D. Marlene Grace Verghese1(&), Suresh Bandi2, and G. Jacob Jayaraj3 1

2

Kommuri Pratap Reddy Institute of Technology, Ghanpur, Ghatkesar, Telangana, India [email protected] Bhimavaram Institute of Engineering and Technology, Pennada, Bhimavaram, AP, India 3 SV College of Engineering and Technology, Moinabad, Hyderabad, Telangana, India

Abstract. Artificial Intelligence is the most widely used in many applications. Especially, applications such as pattern recognition, gender classification and other types of applications. Geometry friends are characters used in the puzzle games and these are in various shapes such as circle or rectangle. Designing the agent based games with geometry friends and AI adopted techniques gives the better result. Many existing systems have the agent mis-understanding problem which may cause confusion between the geometry friends. In this paper, an amalgamative artificial intelligence and adopted with convolution neural networks technique developed a puzzle game to increase the tedious to the player. Keywords: ML


Geometry friends
Artificial intelligence
Generalization

1 Introduction Recently, another stage AI competition has been presented utilizing “Geometry Friends”. The player can control rectangle and circle objects to collect diamonds on the level. The objective is to collect the majority of the diamonds for a short range. The rectangle and circle objects have diverse limit. For instance, rectangle objects can slide sideway and resize (no mass change). Regardless, circle objects can roll sideway and resize (mass change). To manage complex issues, the two indisputable objects need to cooperate. It has a physics engine to simulate friction, acceleration and gravity. Geometry Friends has besides physics based redirection condition, with highlights, for example, growing rate and squashing. This dynamic condition requires a specific extent of capacity and coordination from the two players so as to effectively signify an estimation. This makes an unconventional test to the players, making it an entrancing pertinent examination for Artificial Intelligence (AI).

© Springer Nature Switzerland AG 2020 S. C. Satapathy et al. (Eds.): ICETE 2019, LAIS 4, pp. 528–533, 2020. https://doi.org/10.1007/978-3-030-24318-0_62

Solving the Complexity of Geometry Friends

529

2 Challenges for the AI A few highlights of Geometry Friends make it an especially interesting AI test bed, the most essential being the satisfying fundamental thinking between the two individual players. For human players participation turns into all-great easily as a kind of cooperation and correspondence as technique for accomplishing a typical target. Consistently, to win in Geometry Friends players need to: • Determine which collectibles (important stones) each character can get with no other person’s data and which require joint development of the two players; • Determine the interest by which the collectibles must be acquired, in light of the manner in which those specific activities may incite unrecoverable conditions (for example in the event that the Rectangle falls it can’t ricochet back); • Divide the assignment of gathering the diamonds by the two players, as indicated by the obstacles picked in the two past center interests. • For every collectible, pick the strategy of controls integral for the relating characters to stay it, including shaped activities. • Execute the controls in the distraction condition, pondering the imitated physics. • Do all these always and as quick as would be sensible. • For the greater part of the above clearly, to perform well, players need to achieve some attention and basic commitment with respect to parts of the task. • An AI player should manage all the six, which present inconveniences in three phenomenal estimations: • Dealing with coordination at various estimations: from advancement control to shared planning; • Dealing with fine-grained material science based actuation; • Dealing with inquiry understanding. Since the two characters have specific actuation limits, they move in the game world in various courses and, in light of the likelihood of the estimations, they reliably need to join endeavours to abuse each other’s attributes. This cooperation is required at arranging level, for distributing assignment to manufacture execution, and at actuation level, to urge joint activities to achieve certain parts of the game world. Furthermore, despite the manner in which that Geometry Friends is seen and played like most other two dimensional platform games, for example, Super Mario Bros., the control of the headway is dynamic and testing, in perspective of the physics engine. Specifically, there are different conditions that require right foreseeing synchronous made activities. At last, Geometry Friends, as an inquiry preoccupation, highlights particular estimations where players must use some major thinking, in light of the path that there are requirements in the interest by which collectibles must be picked. This recommends completing a wrong, irreversible activity (e.g. tumbling from a stage too early) can predict completing an estimation with headway.

530

2.1

D. Marlene Grace Verghese et al.

Artificial Intelligence in Geometry Friends

Carlos Fraga was the primary individual making artificial intelligent agents for Geometry Friends [1]. His answer used a navigational graph and its progression was before the essential variant of the competition and was totally occupied with the supportive state of Geometry Friends. The graph has distinctive kinds of edges relying upon whether the edge is traversable by the circle alone, by the rectangle alone, or by the two characters. The nodes of the graph are arranged on both agents’ starting positions and on the diamonds’ positions. These fundamental nodes are expanded, making distinctive nodes all through the measurement. Exactly when the agent will play out a movement, it chooses its way through the A* algorithm. Consequent to picking which approach to take, that way is confined into a great deal of exercises, which are a movement of improvements that will empower the expert to accomplish express nodes of the graph that are en route. In the wake of completing every movement, the authority keeps an eye in the unlikely event that it accomplished the organized action result. In case it had done all things considered, the agent on and on proceeds to next action on the endeavour until the moment that that task is fulfilled. At whatever point a movement is unsuccessful or the task is done, the agent calculates the accompanying task performed. The basic downside of this system is the getting ready overhead caused by running the A* algorithm each time the agent needs to calculate a task. Another issue is that each agent is simply prepared to work together with another expert having a comparative algorithm, which powers an imperative limitation when playing with an optional partner. CIBot (Sejong University): These gathering organized agents that, in a fundamental stage, prioritized the individual collection of diamonds, tolerating indistinct exercises from if these were performing in single-player tracks. After this movement, the two agents start playing supportively, where the circle plays the guideline work in the gettogether endeavour and the rectangle carries on as a moving stage to the drift, in order to accomplish distinctive stages or diamonds that were hard to reach through individual play. The rectangle agent seeks after the circle agents advancement, while endeavouring to be underneath it, and stretches vertically when the two agents are under a diamond, lessening its partition to it. It is the circle agent that picks the target diamond and, after it is assembled, the agent picks the accompanying closest diamond to its position, reiterating the strategy until the moment that the measurement is done. 2.2

The Circle Track Received Two Submissions

CIBot (Sejong University): By creating a directed graph in the start of the every estimation, this agent utilizes the Dijkstra’s calculation to accomplish the most compelled course through the resulting graph. Each stage’s edge is a node and links between nodes are made at whatever point the circle agent can cross beginning with one edge then onto the following one. The estimation most lessened stage, the floor, is solidified into the chart and edges having a place with a similar stage are constantly related. The Dijkstra’s check makes an outline of diamonds requested by their separation to the key position to the circle agent. It picks the first and moves towards it and

Solving the Complexity of Geometry Friends

531

when the diamond is gathered, the overseer moves to the going with one, repeating this procedure until the point that the important stone once-over is vacant, finishing the estimation. The circle moving strategy is masterminded in an immediate standard based-structure that utilizes voracious rolls and bounces to affect the agent to draw nearer to the objective diamond (Fig. 1).

Fig. 1. CIBot circle agent’s graph. The nodes of the graph are the edge points.

KUAS-IS Lab (National Kaohsiung University of Applied Sciences): The circle agent utilizes A* ask for and Q-production feeling of how to finish every estimation. The fundamental check enables the agent to locate the most brief way to deal with amass the majority of the diamonds utilizing a graph that tends to the estimation, in like manner to the CIBot manager, with the capability that this graph takes data of the diamond’s positions and the demand heuristic uses the segment to the diamonds. The Q-learning estimation is utilized to set up the agent to address its pathing by utilizing the Q-table attributes search for heuristic, enabling the head to make stay away from gridlocks and unreachable diamonds, which would make the estimation complete immense.

3 Proposed Method Designing the game with various amalgamative algorithms merged and developed an intelligent game program which is used to provide the complex features for the players. In this paper, the amalgamative algorithm provides the integrated features of artificial intelligence and deep learning algorithms. AI in Game Design: AI in the game is very important task to help the agents and to improve the performance of the system players to handle the critical situation. Role of Geometry Friends: GF is one of the platforms to design the puzzle game. Various reinforcement learning algorithms need a lot of data until they reach accurate performance. The Deep Q-learning from Demonstrations (DQfD) solution [3]

532

D. Marlene Grace Verghese et al.

proposes using a low quantity of demonstration data in order to utilize the learning process. The proposed method combines both supervised learning and reinforcement learning for the agent to be able to learn how to imitate the demonstrator and have selfawareness. The proposed system initialize the threshold value for the level based increase of complexity of the game and also improve the agents performance (Table 1). Table 1. Level based performance in PS. Levels 1 2 3

Existing (Collected cubes) Proposed (Collected cubes) Complexity (PS) 4 6 10% 3 7 70% 2 8 90%

In this game, there are three levels. For every level if the performance is high the complexity becomes high.

4 Conclusion Geometry Friends platform game requires multi target improvement (Time and Diamond Collection) to finish each level. In this paper, the proposed system works not only to improve the performance of the players but also to create the complexity in playing game.

References 1. de Sousa Fraga C (2011) Motion control for an artificial character teaming with a human player in a casual game. PhD thesis, Instituto Superior Técnico 2. Browne CB, Powley E, Whitehouse D, Lucas SM, Cowling PI, Rohlfshagen P, Tavener S, Perez D, Samothrakis S, Colton S (2014) A survey of Monte Carlo tree search methods. IEEE Trans Comput Intell AI Games 4(1):1–43 3. Seetharaman G, Lakhotia A, Blasch EP (2006) Unmanned vehicles come of age: The DARPA grand challenge. Computer 39(12):26–29 4. Billings D, Papp D, Schaeffer J, Szafron D (1998) Poker as a testbed for AI research. In: Advances in artificial intelligence, pp 228–238 5. Genesereth M, Love N, Pell B (2005) General game playing: overview of the AAAI competition. AI Mag 26(2):62–72 6. Togelius J, Karakovskiy S, Baumgarten R (2010) The 2009 mario AI competition. In: Proceedings of the IEEE congress on evolutionary computation 7. Hingston P (2009) A turing test for computer game bots. Comput Intell AI Games. IEEE Trans 1(3):169–186 8. Lucas SM, Runarsson TP (2006) Temporal difference learning versus co- evolution for acquiring othello position evaluation. In: 2006 IEEE Symposium on Computational Intelligence and Games, pp 52–59. IEEE

Solving the Complexity of Geometry Friends

533

9. Buro M (2003) ORTS: a hack-free RTS game environment. In: Computers and games, pp 280–291. Springer 10. Ontanón S, Synnaeve G, Uriarte A, Richoux F, Churchill D, Preuss M (2016) A survey of real-time strategy game AI research and competition in starcraft. IEEE Trans Comput Intell AI Games 5:293–311 11. Loiacono D, Lanzi PL, Togelius J, Onieva E, Pelta DA, Butz MV, Lonneker TD, Cardamone L, Perez D, Sáez Y et al (2010) The 2009 simulated car racing championship. IEEE Trans Comput Intell AI Games 2(2):131–147 12. Mohan S, Laird JE (2012) Relational reinforcement learning in infinite Mario. arXiv preprint arXiv:1202.6386 13. Prada R, Lopes P, Catarino J, Quiterio J, Melo FS (2015) The geometry friends game AI competition. In: 2015 IEEE conference on computational intelligence and games (CIG), pp 431–438. IEEE 14. Raju NS, Kumar S, Gupta R Artificial intelligence in games 15. Rocha JB, Mascarenhas S, Prada R (2008) Game mechanics for cooperative games. In: ZON digital games 2008, pp 72–80

An Efficient Classification Technique for Text Mining and Applications Vb. Narasimha(&) and Sujatha(&) Department of CSE, Osmania University, Hyderabad, India [email protected], [email protected]

Abstract. Content Mining has transformed into an indispensable research zone. Content Mining is the disclosure by PC of new; effectively darken data, by means of therefore isolating data from different created resources. Content Mining is the path toward isolating charming data or taking in or cases from the unstructured substance that are from different sources. The case disclosure from the substance and record relationship of report is an exceptional issue in information mining. These days, the measure of set away data has been hugely extending well ordered which is all things considered in the unstructured shape and can’t be used for any planning to remove supportive data, so a couple of frameworks, for instance, portrayal, clustering and data extraction are available under the arrangement of substance mining. Remembering the true objective to find a capable and effective system for content request, distinctive techniques of substance game plan is starting late made. Some of them are overseen and some of them unsupervised method for record arrange. In this paper, focus is on thought of substance mining, content mining process, systems used as a piece of substance mining in like manner demonstrating some honest to goodness employments of substance mining. Additionally, brief talk of substance mining preferences and imperatives has been shown. keywords: Text mining
Information extraction
Topic tracking Summarization
Clustering
Question answering etc.




1 Introduction Content Mining [1] is the disclosure of new cloud data, by PC and by means of this isolating data different resources are made. A key part is the association of the removed data together to shape new substances or new theories which are to be explored by the help of more standard techniques for experimentation. Content mining is exceptional in connection to what we think about look for in web. As recorded, the customer is usually scanning for something that is starting now, known and has been created by someone else. The issue is pushing aside all the material that at display is not relevant to your necessities with an ultimate objective to find the imperative data. The goal in content mining is to discover cloud data, something that no one yet knows and consequently couldn’t have yet recorded as well. Content mining is a minor takeoff from information mining, a field that tries to find intriguing cases from colossal databases. Content mining which is generally called Intelligent Text Analysis, Text Data Mining © Springer Nature Switzerland AG 2020 S. C. Satapathy et al. (Eds.): ICETE 2019, LAIS 4, pp. 534–544, 2020. https://doi.org/10.1007/978-3-030-24318-0_63

An Efficient Classification Technique

535

or Knowledge-Discovery in Text (KDT), insinuates all things considered for the path towards vacating entrancing and non-insignificant data and gaining from unstructured substance. Content mining, an energetic interdisciplinary field draws on data convalescence, information mining, machine learning, estimations and computational historical underpinnings. As most data (over 80%) is secured as substance, content mining is acknowledged to have a high business potential regard. Taking in may be found from numerous wellsprings of data, yet, unstructured work remains the greatest available wellspring of data. The issue of Knowledge Discovery from Text (KDT) [6] is to isolate, express and irrefutable thoughts and semantic relations between thoughts using Natural Language Processing (NLP) systems. Its point is to get bits of learning into extensive measures of substance information. KDT, while significantly settled in NLP, draws on methodologies from estimations, machine getting the hang of considering data extraction, learning organization, and others for its divulgence technique. KDT accepts a confirmable imperative portion in creating applications, for instance, Text Understanding. Content mining bear a resemblance to information mining, beside that information mining gadgets [2] are expected to be able to sort out information from databases, yet message mining can work with unstructured or semi-composed informational collections, for instance, messages, full-content reports and HTML records et cetera. In this way, content mining is an extremely enhanced retort for associations. Till date, nevertheless, most creative work tries have concentrated on information mining attempts using sorted out information. The issue displayed by content mining is plainly obvious: regular tongue was created for individuals to talk with each other and to record data, and PCs are a long way from acknowledging trademark vernacular. Individuals can perceive and apply semantic cases to substance and individuals can without a doubt crush blocks that PCs can’t without a lot of an extend handle, for instance, slang, spelling assortments and legitimate essentialness. Regardless, in spite of the way that our lingo limits empower us to get a handle on unstructured information, we don’t have the PC’s ability to plan message in considerable volumes or at high speeds. Figure 1 on next page, depicts a non particular process show [3] for a substance mining application. Starting with an aggregation of reports, a substance mining instrument would recuperate a particular chronicle and preprocess it by checking setup and character sets.

Fig. 1. An example of text mining

By then it would come across a substance examination organize, as a rule reiterating strategies until the point that data is evaluated. Three substance examination frameworks are showed up in the delineation, yet various diverse blends of techniques could be used depending upon the targets of the affiliation. The consequent data can be set in an organization data system, yielding a plenty measure of learning for the customer of that structure.

536

Vb. Narasimha and Sujatha

2 Technology Foundations Regardless of the way that the qualifications in human and codes are extensive, there have been mechanical advances which have begun to close the crevice. The field of ordinary tongue get ready has made headways that show PCs trademark lingo with the objective that they may look at, appreciate, and even create content. A segment of the advances [4] that have been created and can be used as a piece of the substance mining process are data extraction, point following, plot, arrangement, clustering, thought linkage, data observation, and question answering. In the coming sections we will look at each of these advances and the part that they play in content mining. We will similarly outline the sort of conditions where each advancement may be significant with a particular true objective to empower per users to perceive mechanical assemblies vital to themselves or their affiliations. 2.1

Data Extraction

The initial stage for PCs to separate unstructured substance is to use data extraction. Data extraction programming identifies key expressions and associations inside substance. It does this via hunting down predefined groupings in content, a strategy called configuration planning. The item initiates the associations between all the recognized people, places, and time to outfit the customer with imperative data. This development can be greatly important while overseeing extensive volumes of substance. Customary information mining acknowledges that the data to be “mined” is currently as a social database. Disastrously, for a few applications, electronic data is quite recently open as free trademark tongue reports rather than sorted out databases. Since IE addresses the issue of changing a corpus of artistic documents into a more sorted out database, the database created by an IE module can be given to the KDD module to moreover mining of data as depicted in Fig. 2.

Fig. 2. Overview of IE-based text mining framework

2.2

Topic Tracking

A point following system works on keeping customer profiles and, in light of the reports the customer witnesses it predicts diverse records imperative to the customer.

An Efficient Classification Technique

537

Yippee offers a free subject after gadget (www.alerts.yahoo.com) that empowers customers to pick catchphrases and advises them when news relating to those focuses winds up clearly available. Subject after development has limitations, in any case. For example, if a customer sets up an alert for “content mining”, she/he will get a couple of news stories on digging for minerals, and not a lot of that is very message mining. A part of the better substance mining contraptions let customers select particular classes of interest or the item thusly can even infer the customer’s preferences in perspective of his/her examining history and explore data. There are various locales where subject after can be associated in industry. It [5] can be used to prepared associations at whatever point a contender is in the news. This empowers them to remain mindful of centered things or changes in the market. So additionally, associations may need to track news isolated association and things. It could in like manner be used as a piece of the helpful business by experts and different people scanning for new solutions for infections and who wish to keep up on the latest types of progress. Individuals in the field of guideline could similarly use subject after to ensure they have the latest references for explore in their general region of premium. Watchwords are a plan of basic words in an article that gives unusual state depiction of its substance to per users. Perceiving catchphrases from a good deal of on-line news info is particularly helpful therein it will convey a brief framework of reports articles. As on-line content reports chop-chop enlarge and measure with the improvement of computer network, slogan extraction [6] has reworked into associate introduce of some of substance mining applications, for example, web record, content request, diagram, and field of study. Manual watchword extraction may be a to an improbable degree hard and monotonous task; honestly, it’s for all intents and functions laborious to expel catchphrases physically if there need to emerge an occasion of reports articles disseminated during a singular day as a results of their volume. For a speedy use of watchwords, we want to develop a motorized methodology that concentrates catchphrases from news articles. The coming up with of watchword extraction structure is given in Fig. 3.

Fig. 3. The architecture of keyword extraction system

538

Vb. Narasimha and Sujatha

HTML news pages are congregated from an Internet passage site. In addition, confident watchwords are isolated hurl catchphrase extraction module. In conclusion catchphrases are removed by cross-space examination module. Catchphrase extraction module is portrayed in detail. We make tables for ‘chronicle’, ‘word reference’, ‘term happen conviction’ and ‘TFIDF weight’ in social database. At first the downloaded news records are secured in “File” table and things are expelled from the reports in ‘Record table. Then [7] the facts which words are appeared in documents are updated to ‘Term occur fact’ table. Next, TF-IDF weights for each word are calculated using ‘Term occur fact’ table and the result are updated to ‘TF-IDF weight’ table. Finally, using ‘TF-IDF weight’ table, ‘Candidate keyword list’ for each news domain with words is ranked high. Keyword extraction module is given in Fig. 4.

Fig. 4. Keyword extraction module

Lexical tying [5] is a strategy for gathering lexically related terms into purported lexical chains. Subject following includes a given news occasion in a surge of news stories i.e. discovering all the ensuing stories in the news stream. In multi vector [8] point following framework legitimate names, areas and typical terms are separated into unmistakable sub vectors of report portrayal. Measuring the likeness of two archives is led by contrasting two sub-vectors at once. Number of elements that impact the execution of subject following framework are examined. To start with decision is to pick one trademark, for example, the selection of words, words or expressions, for example, string as an element in this term to make highlights for instance. That examine the given occasion. 2.3

Outline

Content outline is colossally useful for attempting to make sense of regardless of whether a protracted archive addresses the client’s issues and merits perusing for

An Efficient Classification Technique

539

additional data. With huge writings, content synopsis programming forms and compresses the report in the time it would take the client to peruse the main section. The way to synopsis is to lessen the length and detail of a report while holding its primary focuses and general importance. The test is that, in spite of the fact that PCs can distinguish individuals, places, and time, it is as yet hard to instruct programming to investigate semantics and to translate meaning. By and large, when people condense content, we read the whole choice to build up a full understanding, and after that compose a synopsis highlighting its fundamental focuses. Since PCs don’t yet have the dialect abilities of people, elective techniques must be well thought out. One of the methodologies most generally utilized by content outline devices, sentence extraction, removes vital sentences from an article by factually weighting the sentences. Advance heuristics, for example, position data are additionally utilized for synopsis. For instance, outline instruments may extract the sentences which take after the key expression “in conclusion”, after which normally lie the primary purposes of the record. Outline devices may likewise look for headings and different markers of subtopics keeping in mind the end goal to recognize the key purposes of a record. Microsoft Word’s AutoSummarize work is a straightforward case of content outline. Numerous content synopsis devices enable the client to pick the rate of the aggregate content they need removed as an outline. Rundown can work with subject following instruments or classification devices keeping in mind the end goal to abridge the records that are recovered on a specific point. In the event that associations, restorative staff, or different specialists were given many reports that tended to their subject of intrigue, at that point rundown devices could be utilized to diminish the time spent dealing with the material. People would have the capacity to all the more rapidly evaluate the relevance of the data to the point they are occupied with. A programmed outline [9] process can be separated into three stages: (1) In the preprocessing step an organized portrayal of the first content is acquired; (2) In the preparing step a calculation must change the content structure into a rundown structure; and (3) In the era step the last synopsis is acquired from the outline structure. The strategies for outline can be ordered, as far as the level in the etymological space, in two general gatherings: (a) shallow methodologies, which are confined to the syntactic level of portrayal and attempt to separate notable parts of the content advantageously; and (b) more profound methodologies, which accept a semantics level of portrayal of the first content and include phonetic preparing at some level. In [10] the main approach the point of the preprocessing step is to lessen the dimensionality of the portrayal space, and it typically incorporates: (i) stop-word disposal – common words with no semantics and which don’t total pertinent data to the assignment (e.g., “the”, “an”) are killed; (ii) case collapsing: comprises of changing over every one of the characters to a similar sort of letter case - either capitalized or bring down case; (iii) stemming: linguistically comparable words, for example, plurals, verbal varieties, and so forth are viewed as comparable; the reason for this system is to get the stem or radix of each word, which stress its semantics. A much of the time utilized content model is the vector show. After the preprocessing step every content component –a sentence on account of content rundown – is considered as a N-dimensional vector. So it is conceivable to utilize some metric in this space to gauge closeness between content components. The most utilized metric is the cosine measure, characterized as cos q = ()/(|x| . |y|) for vectors x and y,

540

Vb. Narasimha and Sujatha

where () shows the scalar item, and |x| demonstrates the module of x. Thus, greatest comparability relates to cos q = 1, though cos q = 0 demonstrates adding up to disparity between the content components. To execute content synopsis in light of fluffy rationale, MATLAB is normally utilized since it is conceivable to mimic fluffy rationale in this product. Select normal for a content, for example, sentence length, closeness to pretty much nothing, comparability to watchword and so on as the contribution of fluffy framework. At that point, every one of the tenets required for rundown are entered in the information base of this framework. A short time later, an incentive from zero to one is acquired for each sentence in the yield in view of sentence qualities and the accessible guidelines in the learning base. The development of incentive in the yield decides the level of the significance of the sentence in the last synopsis. The Kernel of producing content outline utilizing sentence determination based content rundown approach [11] is appeared in Fig. 5.

Fig. 5. Kernel of text summarization

An Efficient Classification Technique

541

3 Text Mining Applications The fundamental Text Mining applications [12] are frequently utilized as a part of the accompanying segments: • • • • •

Publishing and media. Telecommunications, vitality and different administrations ventures. Information innovation part and Internet. Banks, protection and monetary markets. Political establishments, political experts, open organization and authoritative records. • Pharmaceutical and think-tanks and human services. The areas broke down are portrayed by a reasonable assortment in the applications being tested. In any case, it is conceivable to distinguish some sectorial determinations in the utilization of TM, connected to the kind of formation and the destinations of the information administration driving them to utilize TM. The distributing segment, for instance, is set apart by ubiquity of Extraction Transformation Loading applications for the recording, creating and the encroachment of the information recovery. In the keeping money and protection parts, then again, CRM applications are common and gone for enhancing the administration of client correspondence, via programmed frameworks of message re-steering and with applications supporting the web search tools making inquiries in characteristic dialect. In the therapeutic and pharmaceutical parts, utilizations of Competitive Intelligence and Technology Watch are across the board for the investigation, characterization and extraction of information from articles, logical edited compositions and licenses. A division in which a few sorts of utilizations are broadly utilized is that of the broadcast communications and administration organizations: the most imperative targets of these enterprises are that all applications discover an answer, from advertise investigation to HR administration, from spelling amendment to client feeling study. A. Content Mining Applications in Knowledge and Human Resource administration Text Mining is generally utilized as a part of field of learning and Human Resource Management. Following are its couple of utilizations in these territories: (1) Competitive Intelligence: The need to sort out and adjust their methodologies as per requests and to the open doors that the market show requires that organizations gather information about themselves, the market and their rivals, and to oversee colossal measure of information, and investigating them to make arrangements. The point of Competitive Intelligence [13] is to choose just pertinent information via programmed perusing of this information. Once the material has been gathered, it is characterized into classes to build up a database, and investigating the database to find solutions to particular and essential information for organization techniques. The run of the mill questions concern the items, the areas of speculation of the competitors, the organizations existing in business sectors, the apposite budgetary markers, and the names of the representatives of an organization with a specific profile of abilities. Prior to the presentation of TM, there was a division that was totally devoted to the consistent checking of information (budgetary, geopolitical, specialized and monetary) and noting the inquiries originating from different segments of the organization. In these cases the

542

Vb. Narasimha and Sujatha

arrival on speculation by the consumption of TM improvements was plainly obvious when contrasted with comes about already accomplished by manual administrators. Sometimes, if [13] a plan of classifications is not characterized from the earlier, burning techniques are utilized to arrange the arrangement of reports (considered) significant concerning a specific subject, in groups of archives with comparable substance. The examination of the key ideas gives a general vision of the subjects managed in the single writings that are shown in single groups. More organization and news information are progressively accessible on the web. All things considered, it has turned into a gold mine of online information that is critical for focused insight (CI). Content mining strategies have been created to assemble and sort out this information and different web search tools are used for the same. Nonetheless, the client has no control on how the information is composed through these devices and they may not coordinate their needs. Physically aggregating records are indicated by a client and significantly inclined to work, and is extraordinarily opened up when it should be refreshed much of the time. Updates to what has been gathered often need a rehashed look, separating of beforehand recovered archives and re-sorting out. FOCI [14] (Flexible Organizer for Competitive Intelligence), will facilitate the training specialist within the occasion, finding out, following, and spreading of aggressive insight or info bases on the online. FOCI modify a consumer to characterize and customize the association of the data teams as indicated by their needs and inclinations into portfolios. Figure 6 demonstrates the planning of FOCI. It includes associate degree military operation module for convalescent pertinent info from the online sources; a Content Management module for transcription info into portfolios and customizing the portfolios; a Content Mining module for locating new info and a Content publication module for distributing and sharing of data and a UI side for graphical illustration and shoppers associations. The portfolios created square measure place away into CI info bases which may be shared by the shoppers within associate degree association.

Fig. 6. FOCI system architecture

An Efficient Classification Technique

543

Content mining can speak to adaptable ways to deal with information administration, research and investigation. Consequently message mining can grow the clench hands of information mining to the capacity to manage literary materials. The accompanying Fig. 7 addresses the way toward utilizing content mining and related strategies and procedures to separate business insight [15] from multi wellsprings of crude content information. Despite the fact that there appears something to that effect of information mining, this procedure of content mining picks up the additional energy to remove growing business knowledge.

Fig. 7. Text mining in business intelligence

4 Conclusion Finally we have a tendency to infer that, Text mining is otherwise known as Text data processing or Knowledge-Discovery in Text (KDT), alludes by and enormous to the method toward extricating fascinating and non-insignificant data and learning from unstructured content. Content mining may be a vernal knowledge domain field which pulls on data recovery, data mining, machine learning, measurements and process linguistics. As most data (more than 80%) is place away as content, content mining is accepted to own a high business potential esteem. Learning could be found from several wellsprings of knowledge, yet, unstructured writings stay the most important promptly accessible wellspring of learning.

544

Vb. Narasimha and Sujatha

References 1. Berry Michael W (2004) Modified discovery of similar words. In: Outline of text mining: clustering, classification and retrieval. Springer, LLC, New York, pp 24–43 2. Navathe SB, Elmasri R (2000) Data warehousing and data mining. In: Nuts and bolts of database systems. Pearson Education Pvt. Inc., Singapore, pp 841–872 3. Fan W, Wallace L, Rich S, Zhang Z (2005) Exploiting the power of text mining. J ACM 4. Bolasco S, Canzonetti A, Della Ratta-Rinald F, Singh BK (2002) Understanding text mining: a pragmatic approach, Roam, Italy 5. Liu L, Chen J (2002) Research of web mining. In: Proceedings of the fourth world congress on intelligent control and automation. IEEE, China, pp 2333–2337 6. Karanikas H, Theodoulidis B (2001) Learning discovery in text and text mining software. Center for Research in Information Management, Manchester 7. Liritano S, Ruffolo M (2001) Managing the knowledge contained in electronic documents: a clustering method for text mining. IEEE, Italy, pp 454–458 8. Brin S, Page L (1998) The life frameworks of a largescale hyper printed web file. Comput Netw ISDN Syst 30(1–7):107–117 9. Kleinberg JM (1999) Authentic sources in hyperlinked condition. J ACM 46(5):604–632 10. Dean J, Henzinger MR (1999) Finding related pages in the web. Comput Netw 31(11– 16):1467–1479 11. Kanya N, Geetha S (2007) Data extraction: a text mining approach. In: IET-UK international conference on information and communication technology in electrical sciences. IEEE, Dr. M.G.R. School, Chennai, Tamil Nadu, India, pp 1111–1118 12. Godbole S, Roy S (2008) Substance to intelligence: building and deploying a text mining solution in the services industry for customer satisfaction analysis. IEEE, India, pp 441–448 13. Lee S, Kim H (2008) News keyword extraction for topic tracking. In: Fourth international conference on networked computing and advanced information management. IEEE, Koria, pp 554–559 14. Carthy J, Sherwood-Smith M (2002) Lexical chains for point following. In: International conference. IEEE SMC WP1M1, Ireland 15. Wang X, Jiang L, Ma J, Jiangyan (2008) Utilization of NER information for improved topic tracking. In: Eighth international conference on intelligent systems design and applications. IEEE PC Society, Shenyang, pp 165–170

Clinical Big Data Predictive Analytics Transforming Healthcare: - An Integrated Framework for Promise Towards Value Based Healthcare Tawseef Ahmad Naqishbandi1(&) and N. Ayyanathan2 1

2

Department of Computer Science and Engineering, B.S. Abdur Rahman Crescent Institute of Science and Technology, Chennai 600048, India [email protected] Department of Computer Applications, B.S. Abdur Rahman Crescent Institute of Science and Technology, Chennai 600048, India [email protected]

Abstract. Modern data driven clinical healthcare application system development requires interdisciplinary and technical expertise to find hidden values from large volume of clinical data. Predictive big data analytics in combination with other technologies like machine learning is growing and is attracting much attention. Therefore, there is a need of integrated healthcare framework which can utilize the power of Predictive analytics, big data; Machine learning. In this paper, we have presented an integrated frame work for handling clinical data, which can act as reference for adoption and integration of clinical data. The purpose of the proposed integrated framework for Healthcare Clinical big data predictive analytics is to explore and combine the power of various analytical techniques and technologies so as to provide a comprehensive solution for value based healthcare. This framework is further committed to transform our perspective towards value based healthcare. Keywords: Big data Predictive analytics


Clinical data
Healthcare
Healthcare analytics



Value
Volume

1 Introduction From common cold to life threatening diseases like CVD, sepsis, cancer, diabetes the potential of predictive health care analytics to improve care and clinical patient outcome can be tremendous if applied properly. However, the application and use of predictive health care is at its early stage. Predictive analytics, with focus on healthcare, is exponentially growing as a transformative tool that can empower progressively elucidating, proactive and preventative treatment choices. The complicated set of comorbidities, unfavorable natural and social conditions among patients makes the medicinal services and health related medical problems extraordinarily difficult and

© Springer Nature Switzerland AG 2020 S. C. Satapathy et al. (Eds.): ICETE 2019, LAIS 4, pp. 545–561, 2020. https://doi.org/10.1007/978-3-030-24318-0_64

546

T. A. Naqishbandi and N. Ayyanathan

requires an adaptable, imaginative, and multidisciplinary way to deal with so as to achieve value based healthcare. The global market place for clinical data analytics is growing at very fast pace and is taking vast approaches into consideration to address their analytical needs so that value based quality care can be achieved. According to Global Clinical Data Analytics [1] market value is expected to reach USD by 2022 and by application, quality care holds the maximum market share globally and is expected to reach USD 3,443.0 million by 2022. In the era of modern data analytical methodologies and technologies like data science, deep learning, machine learning, artificial intelligence no clinical data should not go waste. Clinical big data predictive data analytics is expected to grow at a CAGR of 31.42% [1] in Asia during the forecast period from 2016 to 2022. The present scenario of clinical as well as traditional Electronic healthcare systems has many pitfalls, drawbacks and inefficiencies. Approximately 1.2 billion clinical documents [2] are produced in the United States alone each year. These documents comprise around 60% of all clinical data but unfortunately the insights, events and patterns from this tremendous data and other data sources available around the world is underutilized [8], and poorly managed resulting in wastage of resources, opportunities and potential delay in value based patient care due to unavailability of proactive predictions. Applying Predictive health analytics on data will boost not only clinical healthcare research According to Mckinsey [66] “it will boost the new health economy the ensuing economic impact could reach $845 billion to some $2.5 trillion globally by 2025” and examining the key trends and insights in healthcare clinical analytics will bring these unutilized healthcare transformations into focus. Time has come to harness and explore interdisciplinary nature of Health care and computing. Bringing together these areas will boost excellence in clinical research and decision making so that extra care of patients can be taken by doctors and nurses by recognizing adverse events well before events might happen, by visualizing trends and insights in the healthcare/clinical data, for this to happen proactive predictive healthcare analytics is the key to success. However, in the name of academic research, there remains always a potential threat and greater concerns about the illegal use of clinical data as data science analytics has started revealing some unusual alarming patterns and insights. As rightly said by Johan van der Lei “data shall be used only for the purpose for which they were collected” and was called as “1st Law of Medical Informatics” [4]. Van der Lei [4] argued that “health care data can easily be misused outside the context where they were collected”, but it would be a blunder to deny the actual fact that the Big Data era has created promising, new opportunities for hidden information discovery from data. Predicting behavior of an individual well before adverse event “Detecting the deteriorating patient… is a major goal,” wrote Moorman et al. [3]. As mentioned in EY’s Progressions 2018 – “human body is the biggest data platform, but the question is who will capture value. “It can be a paradigm shift in how we practice medicine”. Focusing on value based care, turning big data into actionable clinical intelligence. Generating Patterns, insights and predictions from clinical data using various machine learning and big data analytical techniques will not only help value based patient care, but will also help healthcare organizations to deliver better service. However due to

Clinical Big Data Predictive Analytics Transforming Healthcare

547

unpredictable and diversified nature of clinical data sources, task of transforming raw data to valuable data can have greater impact on healthcare. Predictive analytics, big data and Machine-learning (ML) technologies play a key role in turning this faraway dream into reality. Machine-learning (ML) extends an alternative approach for standard prediction modeling which may address present and existing limitations. It has potential to transform predictive healthcare by better exploiting clinical big data [7], so that healthcare related risks can be diagnosed and prevented. Keeping the potential of big data predictive analytical capability in mind, we hope to give healthcare organizations and stake holders, more latest and comprehensive understanding of clinical big data predictive analytics which will ultimately help in transforming volume based healthcare data to value based healthcare. Accordingly, this Comprehensive integrated technological outlook frame work will explore various conceptual aspects of applying big data and related technologies in combination with predictive analytics.

2 Research Study Method The systematic ongoing literature study was done by accessing and capturing relevant research information, focusing on the following key aspects: Focus on concepts and characteristics of big data, clinical predictive health analytics and related technologies. Exploring various perspectives of big data. Identification, Integration and understanding of healthcare Predictive analytical techniques and technologies in healthcare. To develop an integrated frame work combining various perspectives of healthcare, Predictive analytics, Big Data, Machine learning. To understand various predictive analytical strategies for tackling advanced challenges of predictive analytics health care. 2.1

Information Data Sources

Latest and relevant research articles from year 2015 to 2018 were searched. Following databases: Science Direct, Springer, Taylor & Francis, Pub Med, IEEE Explore, Emerald were accessed, inclusion of highly cited references articles other than accessed from the above databases and period were also included based of the relevance of study. 2.2

Search Query Keywords

On the basis of research relevance, consideration and inclusion of research articles, following Search query was utilized on all the above databases defined in Sect. 2.1. SQC 1:- Those research papers that talk about “predictive health analytics”. SQC2:- Those research papers that deal with “predictive health analytics”, “machine learning”.

548

T. A. Naqishbandi and N. Ayyanathan

SQC 3:- Those research papers that talk about “predictive health analytics, big data, machine learning. SQC4:- Those research which focus on “Clinical health analytics, big data, machine learning”. SQC5:- Those research which focus only on “Clinical big data health analytics”. 2.3

Methodology for Study Selection

The methodology pursuit of research literature was carried out in a predefined manner as per the requirement and relevance of study (Fig. 1): The search and retrieval of research publications from scientific databases containing keywords “Predictive Health Analytics”, “big data”, “Clinical big data analytics”, and “healthcare” or “clinical health analytics” or “Machine learning”. Screening of the Articles was based on Title, abstract and keywords and inclusion of articles with high relevance on the basis of predefined Selection method. Screening and disposal of research articles that was not disposed in the previous phase of review. Search and inclusion of interdisciplinary articles for detailed study.

Fig. 1. Study selection process

Clinical Big Data Predictive Analytics Transforming Healthcare

2.4

549

Research Quality Assessment

To ensure the quality of research a detailed literature study was done so as to ensure minimal diversion and confusion from the research area under consideration. The searches were made in relevance to the title to avoid irrelevant source which can lead to chaos and confusion in latter stage. Further necessary screening of abstracts was carried out and it was properly discussed and decided which research articles should be incorporated or rejected from the study. 2.5

Study Result Summary

The popular scientific databases were accessed for literature study and a predefined selection process was followed. Science Direct, Springer, Pub med, Emerald and Taylor & Francis returned a total hit of (8537, 680,106, 1039, 2463) as on October 2018. On the basis of the title, 130 articles were selected as per the already decided search query criteria, out of these research articles based on the relevance of abstracts and keywords 95 relevant articles were selected for next phase of study. The main crux of these 95 articles read thoroughly. Articles discussing security and privacy issues of Big Data predictive analytics and traditional models for Big Data predictive analytics in healthcare were out of scope and excluded. Out of 95 thoroughly screened articles, a total of 65 articles were included for further study. On the basis predefined research objectives, the content from these research articles was extracted and the articles were categorized into different groups. The subsequent section summarizes the findings in each of these categories.

3 Technological Overview 3.1

Clinical Big Data Predictive Health Analytics

By 2020, the amount of data generated worldwide is estimated to hit 44 zettabytes across the entire world against 4 zettabytes in 2013. There is an enormous amount of data available within the healthcare sector [13]. Clinical big data here refers to the data generated by human body in different blends. This clinical healthcare data is stored in large databases commonly known as EHR’S (Electronic Health Records) [14–16]. With diverse structured, semi-structured, unstructured broad range availability of healthcare data sources from traditional lab records to genome data, healthcare data is enabled by the continuous acquisition & aggregation of large amounts of healthcare data [13], which is growing exponentially making it practically impossible for medical professionals to extract hidden values from data, even it has become difficult to extract Useful hidden, patterns and proactive events while using conventional data techniques and tools due to the nature of data [9–11] “With growing digitization that enables us to collect vast amounts of data, we have to start leveraging the value of data,” said Dr. Ian Chuang. By leveraging the most advanced technologies and methods the ultimate goal is to supply these predictive patters and insights to the right decision makers at right time to help in eliminating healthcare related risks to improve care. By examining the various perspectives in detail, we can make a strong argument that the proposed

550

T. A. Naqishbandi and N. Ayyanathan

integrated framework in future can make a practically remarkable research contribution towards understanding the complex clinical care workflow, future aspects and application across the broad domain of clinical healthcare predictive Big Data analytics. Clinical Big data analytics is not only about data, but it also talks about analytics that can be performed while analyzing the data. Analytics is a multidimensional discipline which extensively uses mathematics, statistics, descriptive techniques, predictive models etc. to uncover hidden patterns from data and appropriately communicate those patterns. Clinical Raw data from different sources is taken and transformed by appropriate data preprocessing methodologies. This transformed data is then fed to various big data platforms and tools for necessary analytics. Outcomes of the analytics are then passed on to end user in required formats using appropriate applications. Figure 1 illustrates the conceptual pre-processing architecture of Clinical big data predictive analytics in the healthcare domain. Clinical healthcare big data predictive analytics is made possible with an combination of various platforms and tools like Hadoop, Hive, HBase, MapReduce, CouchDB, MySQL, NoSQL, UnSQL, Jaql, Cassandra, MongoDB, Pig, SOLR, flume, sqoop, Ambari, Oozie, ZooKeeper, R, Mahout and such others. The applications of predictive healthcare analytics are numerous and hold tremendous potential patient outcomes. For example, at the lowest level of analytics one should be able predict which patients are having high risk of cardiovascular disease, which patient are likely to be readmitted after pacemaker implantation, or which patient will stay longer than the average after implantation [12]. For these tremendous applications, predictive healthcare analytics is receiving consistent amount of focus from last couple of years. The knowledge gained through by applying data science, machine learning and predictive analytics in healthcare will change the narrative of old way of practicing medicine while enhancing our ability to prevent and treat significant diseases and illnesses. 3.2

Characteristics and Challenges

Clinical Big data is characterized by volume, velocity, variety, veracity, variability and value [3–6]. The term “Clinical “attached with big data will shift complete focus on healthcare domain so that value based healthcare can be achieved as depicted in Fig. 1 and can be termed as “clinical big data” characteristics. There are various characteristics of big data described in literature; this study focuses on some characteristics mainly because of clinical healthcare importance. Clinical Volume-in healthcare refers to the large amount of clinical data generated by human body, which is predicted to rise dramatically to 35 zettabytes by 2020 [21]. The clinical variety on the other hand talks about the diversified, nature of clinical data with different blends, types and formats (e.g., EHR’S, diagnostic labs, pharmacy, clinical laboratories, clinical trials, genomics, medical data etc.) and involves advanced data sources like clinical research data repository, clinical content review, remote patient monitoring data, location and demographic data, treatment based data, illness based, injury based) aggregated and extracted from different sources [18, 20] but it’s more than clinical volume and variety. Clinical velocity refers to receiving of data from multiple sources with enormous speed and complexity and clinical veracity refers to

Clinical Big Data Predictive Analytics Transforming Healthcare

551

those sources which can impact accuracy and conformity of facts [47]. Having accurate information is of greater concern in clinical healthcare [44] therefore data quality and veracity plays an important role in clinical healthcare analytics. These above defined clinical characteristic also contribute to challenges in the successful implementation of predictive big data applications in healthcare analytics [21]. Therefore there is dire need to address this clinical Volume, velocity [11], veracity and complexity keeping various critical consequences and numerous factors in mind, so that clinical value –from big data using can be fully achieved as it exhibits a wide range of medical and healthcare functions. With increasing complexity their needs to be greater focus towards values [19] based healthcare. Due to volume, variety and veracity Clinical Healthcare data [18] does not remain consistent and complete. With the incorporation of fragmented data [5, 26, 44], which ultimately leads data inaccuracy and inconsistency (veracity) [5, 46–48], data reliability [45], interoperability [25, 59, 87], network bandwidth, scalability, and cost [8]. In other words, performing analytics, finding hidden patterns and then taking decisions based on patient engagement and clinical practice is complex and tedious in nature due to clinical veracity [22] and Security related issues such as Data theft and breaches can be significant threat in healthcare [28, 43]. Apart from technical issues, big data predictive healthcare analytics face certain key challenges for use in health care: confidentiality and data security [52] access to information [53] data reliability, interoperability [54, 55] and management and governance [56, 57]. As rightly said by Mittelstadt et al. [49] informed consent and privacy are the key areas of concern. 3.3

Proposed Integrated Clinical Big Data Predictive Framework

By definition, Clinical healthcare predictive analytics focuses on finding hidden patterns and insights by extracting and analyzing large volume and variety of clinical data sets sources using various powerful analytical tools. The multivariate, multidimensional, diversified nature of healthcare data and its dynamic nature make clinical data difficult to analyze [26]. To handle diversified data, the proposed framework model should have the capability of utilizing various analytical techniques, machine learning, big data techniques and tools to develop predictive models so as estimate health care hidden patterns, events & insights [23]. To use and sense the power of big data predictive analytics, healthcare stakeholders can take extra advantage of predictive analytics framework which can incorporate data, techniques, technology and algorithms [27]. Theses model are aimed at generating accurate predictions from newly observed data, where new data can be analyzed [24]. The overall goal of the proposed framework is to reduce various clinical risks which will ultimately help decision makers to make better decisions better and faster. There are various analytical challenges associated with clinical big data which makes traditional methods invalid in analyzing clinical big data. The new analytical methods developed to overcome these challenges are often referred to as “predictive analytics”. The narrative of big data towards clinical analytics is depicted in Fig. 2. Big data reflecting usage of healthcare clinical data will help to build predictive models [25]. With the inclusion of machine learning based analytics, healthcare organizations have seen improved quality of care. However, the long-term substantial benefits can be

552

T. A. Naqishbandi and N. Ayyanathan

Fig. 2. Characteristics of big data healthcare

accumulated by utilizing the power of predictive analytics [28]. The proposed clinical predictive healthcare framework offers a systematic analytical processing pipeline from data acquisition stage to decision making stage. This proposed framework as shown in Fig. 3 marks the foundation for development of clinical big data healthcare predictive system which can have the capability of extracting, transforming, aggregating, accumulating and analyzing the exponential growing data in terms of clinical variety(structured, unstructured and semi-structured) so that valuable insights can be gained to improve care. 3.3.1 Clinical Data Sources To improve the state of art value based healthcare certain well-defined measures should be taken to identify various clinical data sources to keep pace between development of latest healthcare technologies and analyzing the clinical data. The healthcare data which is acquired from various sources is an essential component for any predictive system. In the proposed clinical big data healthcare predictive framework, data is captured from various healthcare sources or is available as part of large open data stores or is generated at from different web sources or is captured through EMRS diagnostic labs, medical images, medical claims, clinical trials, internet, medical imaging, patient reported wearable’s and sensors, smart phones, prescription claims [5]. The data sets acquired from these sources can be used for personalized medicine and large cohort studies [58]. Healthcare data can also be collected from various social media applications like face book, twitter etc. gathering information about a specific flu affected population from social media application like Twitter is faster than using any traditional method. Websites like patients like me (www.patientslikeme.com) has more than

Clinical Big Data Predictive Analytics Transforming Healthcare

553

Fig. 3. Predictive frame work—towards value based healthcare

600,000 patients and is tracking more than 2,800 diseases [59]. Patients as a source are themselves one of the largest sources of clinical data. Combining, integrating and analyzing data aggregated from various sources is a real challenge. Aggregated healthcare patient data provides compendious view of an individual’s health. To find new innovations in healthcare and subsequent clinical diagnostic or interventional clinical trials with next-generation technologies clinical data will play an important role [41]. The next stage after data acquisition involves data storage so that pre-processing of data or events can be performed. Utilizing data captured by different healthcare smart applications or online programs will not only accelerate clinical trial but will improve trial efficiency, and will solve personalized and population based healthcare during clinical follow-up. This is where Big Data predictive analytical techniques come into play. 3.3.2 Pre-processing Healthcare data can be used for diagnosis and forecasting certain medical conditions which are hidden and can only be revealed using predictive analytics, big data analytics, machine learning & deep learning. To achieve this data need to be pre-processed so that error rate will be minimal. Clinical data exhibit unique features and numerous attributes including missing values, noise resulting from human as well as system errors. Nearly all clinical diagnoses, treatments, prescriptions in medicine are inexplicit, and are subject to rate of errors [33]. For purposeful analysis to be performed on data in order to achieve best and optimized results data cleaning, finding missing values,

554

T. A. Naqishbandi and N. Ayyanathan

dimensional reduction, and noise is to be detected and removed [60]. The processed data is then used for feature selection process so that performance of training model can be enhanced. For example, a clinical dataset can contain 30 attributes or features, few among them can be significant in decision making process. In order to achieve value based care most significant values will be only taken into consideration. 3.3.3 Clinical Data Analytics Fuel for performing analytics comes from clinical data that is human generated data which consists of Electronic Health Records (EHR) AND Electronic Patient Records (EPR) which is one of the important types of clinical data and can be applied to the larger population for better healthcare [32]. A predictive analytical model helps in understanding a biological system and has the capability to uncover the underlying cause of diseases [30]. Predictive analytics can take a specified clinical dataset from source as input as depicted in Fig. 2 and make predictions based on historical or past data. Predictive healthcare clinical applications are numerous and significant, and personalized care patient outcomes are also possible [31]. For example, one can predict, what are the signs that a particular or a group of patients needs continuous follow-up? Which drugs should be used for better recovery or whether a patient will stay admitted than the expected after a surgery [12]. There are five core key maturity stages in the clinical predictive data analytics, as shown in Fig. 2. The first step into healthcare predictive data analytics is Descriptive analytics: - i.e. what happened? For example for a particular surgical intervention, many questions can arise like, what is the success rate of cardiovascular surgeries performed in last 3 months? What are the signs that a particular or a group of patients needs continuous follow-up? The second step is:Diagnostic Analytics: - i.e. why it happened? Here the use of advanced analytics techniques is used to get insights from the present and past data such that the model can answer various questions like. Why has the success rate of cardiovascular surgical intervention increased in last 3 months? Why has the success rate of cardiovascular intervention decreased in last 3 months? This can potentially suggest or initiate corrective actions itself, based on what it learns from the past data. The third step towards destination is Predictive analytics: - i.e. what will happen? For example which cardio-patients are likely to respond to a given treatment after cardiovascular surgery? Which patients are likely to survive? Which patients need readmission? There can be numerous such questions which can be answered with the use and power of predictive analytics. The real use of clinical predictive analytics therefore lies in taking action to influence or change the predicted outcome. In other words, predict the future, and then act on it and change. Estimating Risk is another wonderful property of predictive analytics. The fourth step is:Prescriptive analytics: - i.e. what should be done? For example, what surgical procedures to be followed in order to increase success rate? Which drugs should be

Clinical Big Data Predictive Analytics Transforming Healthcare

555

used for better recovery? Which diet needs to be followed? Which policies need to be changed in order to increase cardiovascular outcome. Finally, the last but not the least analytical capability is Pre-emptive analytics: - i.e. what more can be done? What preventive drugs to be suggested during routine check up? What recommendations to be suggested like No intervention Exercise Physiotherapy Dietary modification Drug modification. Therefore applying various predictive and big data analytical techniques is the key to solution. 3.3.4 Clinical Big Data Processing Clinical Big data Analytical processing capability refers to the use of modern art technologies and techniques used to handle process and analyze data with unique and supportive data storage, management, analytical, and visualization capability. Analytical techniques can be used to identify hidden patterns, insights and discover hidden associations from large volume of healthcare clinical records, thus providing way for evidence-based personalized as well as population based care. Modern Healthcare analytical techniques fulfill the growing need of generating hidden patterns and insights to allow healthcare research professionals and organizations to capture, process, manipulate and analyze data or clinical records. In doing so, integrated frame work output model should be able to identify unnoticed associations, hidden patterns, insights, readmission rate etc. in order to support a better balance between care and value based care. Use of data mining together with machine learning in clinical healthcare can prove tremendous However, the success of healthcare data mining hinges on the availability healthcare data. On the other hand Machine-learning (ML) with clinical big data offers exponential vision of using big data technologies for predictive modeling that may address various limitations which can prove disadvantageous at certain times. It has a tremendous potential to transform unutilized clinical data of different blends. We found no evidence of machine learning based large-scale investigation using clinical big data based predictive evaluation for general population, and the same argument is supported by [40] except for diagnostic and prognostic prediction [36–39] in literature. As the sensing and computational capability of clinical health care systems is increasing tenfold due to the incorporation of various technologies like artificial intelligence, the opportunities to exploit machine-learning to enhance prediction of disease risk in clinical practice will become a realistic option for improved clinical care outcomes [7], for example for predicting individual health related risk factors to support clinical decisions using genomic and related factors. Modern analytical techniques for example deep learning based predictive analytics is nowadays considered as major technology in analyzing different varieties of healthcare data. Deep learning models have already demonstrated great performance than potential natural language processing tasks [23–26]. Big data predictive analytics isn’t new but it’s the combination of unprecedented volumes of data clinical data where 90% data is unstructured in nature, big data analytical techniques and tools play a great role in clinical analytics by examining the various critical factors which can have negative impact on environmental, and lifestyle factors associated with an individual and the same impact can extend to a large

556

T. A. Naqishbandi and N. Ayyanathan

population [34]. Structured or Unstructured health data of different blends once acquired and aggregated as a part of pre processing phase (as defined in Sect. 3.3.2) is stored in a distributed manner known as hadoop distributed file system and is called upon on users request. The databases in use i.e. NOSQL supports unstructured as well as semi-structured data. In order to perform analytics Map Reduce algorithms by Apache Hadoop [29] need to be implemented. Map Reduce is having the capability of capturing the data from the database and processes it by executing “Map” and “Reduce”, which break down large jobs into a set of small discrete task once the data is analyzed, the results gets stored and is made visually accessible for users to facilitate decision-making. Big data predictive analytical capability in combination with artificial intelligence and machine learning is giving healthcare recommendation systems unprecedented capability to think solve and understand language. Healthcare Organizations that are able to harness these analytical capabilities effectively will be able to produce vital values and differentiate themselves, while others will find themselves increasingly at a disadvantage. 3.3.5 Healthcare Predictive Analytics Application Layer The analytical Application layer is to realize potential of clinical big data predictive analytics. As we know Clinical Healthcare aims to utilize the data sources in order to help healthcare users, practitioners achieve improvements in value based clinical effectiveness. As healthcare data continues to grow massively, healthcare quality is needed to coordinate and support the work done to achieve value based care. Predictive Analytics together with clinical big data can transform healthcare industry [22] as it enables different uses of data collected and aggregated from de-identified sources [35]. It also refers to the collection of data from different healthcare sub-domains. The Examples of these applications are: Specified Disease Analytics (cardiovascular diseases, nephrology, diabetes etc.), Clinical Analytics (association between various diagnostic events and results, pattern generation from clinical values), Financial Analytics(insurance claims and diseases drug cost, Administrative Analytics (hospital admission, re-admission, discharge), Insurance Analytics (claim bill, total insured cost, lump sum claim amount)Genomics, Proteomics, Behavioral Analysis Human Phenotypic Analysis. These represent the applications as well as the sources of clinical big data analytics. Mere by collecting and storing data does not full the potential of big data, but the real promise and potential lies in generation of insights and predicting the unseen. For example big data predictive analytics in genomics will help in preventative treatment and will deliver personalized care to patient at individual level [61]. Research in genomics using big data and related technologies is in still in its earlier stage with certain crucial areas of study, for example leukemia, diabetes, and cancer [62, 63]. The various challenges present in today’s health sector organizations is lack in predictive and prescriptive analytics due to shortage of well trained data scientists that is why they are either still at the descriptive analytics phase or are transitioning between descriptive and predictive analytics. 3.3.6 Users Users use and interact with the system directly or indirectly. The holistic use of predictive system or framework depends on the type user that is provider, payer,

Clinical Big Data Predictive Analytics Transforming Healthcare

557

researcher and the type application and objective.. As depicted in (Fig. 2) users can be categorized into following categories: Healthcare Providers: - Health care providers are those who provide healthcare services For example Medical doctors who have specialization in medicine, general medical practice, or pediatrics. Obstetrician/Gynecologists, Licensed practical nurses (LPNs), Clinical nurse specialists and various other sub domains like Cardiology – heart disorders, Dermatology skin disorders, Endocrinology-hormonal and metabolic disorders, including diabetes Gastroenterology - digestive system disorders, General surgery-common surgeries involving any part of the body, Hematology – blood disorders, Immunology – disorders of the immune system [67]. Healthcare Payers: - Healthcare payer is that percentage of population or users who pay to avail health various health services and related products and services. Payers admit patients as beneficiaries. Payers procure healthcare services from the healthcare providers on behalf of their patient beneficiaries. Healthcare providers, use big data predictive analytics in understanding fraud detection and claim management. Researchers:’- Here “researchers” are those individuals who use the clinical data for various purposes, for example to reveal hidden patters from large volume of data using various algorithmic tools and techniques. Hospitals and Labs: - They provide health various services to patients from common cold to cancer treatment. They store and maintain health related information about them. Hospitals and diagnostic labs are one of the primary sources of data. The healthcare service providers coordinate patient care with other providers. Many providers are independent businesses. Insurance Companies: A business house that provides health coverage for certain diseases and for those health related events which can damage life. The coverage is in the form of injury, accident, surgery, treatment, hospital expenses, death due to accident, heart attack, death while on line of duty etc. A huge volume of data is also generated from insurance companies. Pharmaceutical Companies:-Pharmaceutical companies produce huge volume data. Analyzing this data can reveal hidden patterns. Using big data predictive analytics for various research opportunities like Precision medicine is emerging as an important form medical diagnosis and treatment of diseases using relevant data. During pharmaceutical development for example drug discovery big data is used during all phases [64]. Pfizer [65] initiated Precision Medicine Analytics Environment program which the crucial gaps among electronic medical record data, clinical trial, and genomic data to rapidly develop innovative medicines for particular patient population. 3.3.7 Visualization Visualization tools and techniques help to identify various patterns and insights which include clusters, association among clusters, outliers, weak and strong associated points. It provides an interactive visual analytics in the shape of graphs, histogram by extracting and analyzing data from one or more data sources.

558

T. A. Naqishbandi and N. Ayyanathan

4 Conclusion and Future Work The literature study taken into consideration reveals limited number of practical application using big data predictive analytics in healthcare. Most published work talks about traditional frame works for collection of data, policies, road maps, applications and varying characteristics of clinical big data predictive analytics, but no real world case study supports the technological clinical perspective framework. Literature study reveals that big data predictive analytics have tremendous potential and can improve value based care. However, to get these expectations fulfilled, various technical issues must be looked upon. The analytics and technical issues are largely fragmented. Lack of availability of open source clinical data sets, informed consent and privacy [49], data quality and interoperability are few of the major issues. Working on big data predictive analytics in clinical healthcare is very challenging until there is high-quality genuine data collection, aggregation and analytical processing systems in place with welldefined privacy and governing rules. Big data predictive analytics in combination with other techniques and technologies is emerging a new state of art technology with greater opportunities in healthcare with amazing applications and features. In this article we examined the use of big data predictive analytical framework keeping various characteristics and challenges in mind with broader research objectives. We proposed a framework that supports the integrated use of Clinical Big Data with greater Promise towards value based healthcare. The integrated framework holds the capability of performing meaningful analytics across the sub domains of healthcare. In future it is planned to make use of this framework with a possible practical application for performing clinical data analytics on an open available data set to study a particular disease and related risks.

References 1. Market research future. https://www.marketresearchfuture.com/reports/clinical-data-analyticsmarket-2520. Accessed 21 Nov 2018 2. Health Information and management system. www.himss.org/health-story-project. Accessed 05 Dec 2018 3. Moorman JR, Carlo WA, Kattwinkel J, Schelonka RL, Porcelli PJ, Navarrete CT et al (2016) Mortality reduction by heart rate characteristic monitoring in very low birth weight neonates: a randomized trial. J Pediatr 159:900–916. https://doi.org/10.1016/j.jpeds.2011.06.044 [PMC free article] [PubMed] 4. van der Lei J (1991) Use and abuse of computer-stored medical records. Methods Inf Med 30(2):79–80 5. Szlezák N, Evers M, Wang J, Pérez L (2014) The role of big data and advanced analytics in drug discovery, development, and commercialization. Clin Pharmacol Ther 95:492–495. https://doi.org/10.1038/clpt.2014.29 6. Baig MM, Gholamhosseini H (2013) Smart health monitoring systems: an overview of design and modeling. J Med Syst 37(2):9898 7. Bellazzi R, Ferrazzi F, Sacchi L (2011) Predictive data mining in clinical medicine: a focus on selected methods and applications. Rev Data Min Knowl Discov 1(5):416–430 8. Belle A, Thiagarajan R, Soroushmehr SMR, Navidi F, Beard DA, Najarian K (2015) Big data analytics in healthcare. 1–16. https://doi.org/10.1155/2015/370194

Clinical Big Data Predictive Analytics Transforming Healthcare

559

9. Martin-Sanchez FJ, Aguiar-Pulido V, Lopez-Campos GH, Peek N, Sacchi L (2017) Secondary use and analysis of big data collected for patient care. IMIA Yearb 26:1–10. https://doi.org/10.15265/IY-2017-008 10. Wang Y, Hajli N (2017) Exploring the path to big data analytics success in healthcare. J Bus Res 70:287–299. https://doi.org/10.1016/j.jbusres.2016.08.002 11. Cyganek B, Graña M, Krawczyk B, Kasprzak A, Porwik P, Walkowiak K, Woźniak A (2016) A survey of big data issues in electronic health record analysis. Appl Artif Intell 30:497–520. https://doi.org/10.1080/08839514.2016.1193714 12. Admes J, Garets D (2014) The healthcare analytics evolution: moving from descriptive to predictive to prescriptive. In: Gensinger R (ed) Analytics in healthcare: an introduction. Health Information and Management System Society (HIMSS), Chicago, pp 13–20 13. El-Sappagh SH, El-Masri S, Riad AM, Elmogy M (2013) Data mining and knowledge discovery: applications, techniques, challenges and process models in healthcare. Int J Eng Res Appl 3(3):900–906 14. Cortes PL, Cortes EG (2011) Hospital information systems: a study of electronic patient records. J Inf Syst Technol Manag 8(1):131–154 15. Qiu Y, Zhen S, Zhou M, Li L (2012) Continuously improve the medical care quality and hospital management level through medical information system construction. J Transl Med 10(Suppl. 2):A56 16. Gunter TD, Terry NP (2005) The emergence of national electronic health record architectures in the United States and Australia: models, costs, and questions. J Med Internet Res 7(1):e3. https://doi.org/10.2196/jmir.7.1.e3.pmc1550638.pmid15829475 17. Fang R, Pouyanfar S, Yang Y, Chen C (2016) Computational health informatics in the big data age: a survey. ACM Comput Surv 49(1):1–36. https://doi.org/10.1145/2932707 18. Dinov ID (2016) Volume and value of big healthcare data. J Med Stat Inf 4. http://dx.doi. org/10.7243/2053-7662-4-3 19. Manyika J, Chui M, Brown B et al (2011–2018) Big data: the next frontier for innovation, competition, and productivity. McKinsey Global Institute 20. Rumsfeld JS, Joynt KE, Maddox TM (2016) Big data analytics to improve cardiovascular care: promise and challenges. Nat Rev Cardiol 13:350–359. https://doi.org/10.1038/nrcardio. 2016.42 21. Gui H, Zheng R, Ma C (2016) An architecture for healthcare big data management and analysis. In: International conference on health information science, pp 154–160 22. Sukumar SR, Natarajan R, Ferrell R (2015) Quality of big data in health care. Int J Health Care Qual Assur 28:621–634 23. Wessler M (2013) Big data analytics for dummies. Wiley, Hoboken 24. Shmueli G, Koppius OR (2011) Predictive analytics in information systems research. MIS Q 35(3):553–572 25. Wu P-Y, Cheng C-W, Kaddi CD, Venugopalan J, Hoffman R, Wang MD (2017) Omic and electronic health record big data analytics for precisionnNmedicine. IEEE Trans Biomed Eng 64:263–273. https://doi.org/10.1109/TBME.2016.2573285 26. Dimitrov DV (2016) Medical internet of things and big data in healthcare. Healthc Inf Res 22:156–163. https://doi.org/10.4258/hir.2016.22.3.156 27. Taglang G, Jackson DB (2016) Use of big data in drug discovery and clinical trials. Gynecol Oncol 141:17–23. https://doi.org/10.1016/j.ygyno.2016.02 28. Asante-Korang A, Jacobs JP (2016) Big data and paediatric cardiovascular disease in the era of transparency in healthcare. Cardiol Young 26:1597–1602. https://doi.org/10.1017/ s1047951116001736 29. Wang Y et al (2016) Big data analytics: understanding its capabilities and potential benefits for healthcare organizations. Technol Forecast Soc Change. http://dx.doi.org/10.1016/j. techfore.2015.12.019

560

T. A. Naqishbandi and N. Ayyanathan

30. Gifford DK (2001) Blazing pathways through genetic mountains. Science 293(5537):2049– 2051 31. Alharthi H (2018) Healthcare predictive analytics: an overview with a focus on Saudi Arabia. J Infect Public Health. https://doi.org/10.1016/j.jiph.2018.02.005 32. Hripcsak G, Duke J, Shah N, Reich C, Huser V, Schuemie MS et al (2015) Observational health data sciences and informatics (OHDSI): opportunities for observational researchers. Stud Health Technol Inf 216:574–578 33. Tanwani AK, Farooq M (2009) Performance evaluation of evolutionary algorithms in classification of biomedical datasets. In: Rothlauf F (ed) The 11th annual conference companion on genetic and evolutionary computation: late breaking papers, GECCO 2009. ACM, Canada, pp 2617–2624 34. Auffray C, Balling R, Barroso I, Bencze L, Benson M, Bergeron J, BernalDelgado E, Blomberg N, Bock C, Conesa A, Del Signore S, Delogne C, Devilee P, Di Meglio A, Eijkemans M, Flicek P, Graf N, Grimm V, Guchelaar H-J, Guo Y-K, Gut IG, Hanbury A, Hanif S, Hilgers R-D, Honrado Á, Hose DR, Houwing-Duistermaat J, Hubbard T, Janacek SH, Karanikas H, Kievits T, Kohler M, Kremer A, Lanfear J, Lengauer T, Maes E, Meert T, Müller W, Nickel D, Oledzki P, Pedersen B, Petkovic M, Pliakos K, Rattray M, i Màs JR, Schneider R, Sengstag T, Serra-Picamal X, Spek W, Vaas LAI, van Batenburg O, Vandelaer M, Varnai P, Villoslada P, Vizcaíno JA, Wubbe JPM, Zanetti G (2016) Making sense of big data in health research: towards an EU action plan. Genome Med 8:71. http://dx. doi.org/10.1186/s13073-016-0323-y 35. Cano I, Tenyi A, Vela E, Miralles F, Roca J (2017) Perspectives on big data applications of health information. Curr Opin Syst Biol 3:36–42. https://doi.org/10.1016/j.coisb.2017.04.012 36. Betancur J, Otaki Y, Motwani M et al (2017) Prognostic value of combined clinical and myocardial perfusion imaging data using machine learning. JACC Cardiovasc Imaging 11:1000–1009 37. Haro Alonso D, Wernick MN, Yang Y, Germano G, Berman DS, Slomka P (2018) Prediction of cardiac death after adenosine myocardial perfusion SPECT based on machine learning. J Nucl Cardiol 11:1654–1663 38. Betancur J, Commandeur F, Motlagh M et al (2018) Deep learning for prediction of obstructive disease from fast myocardial perfusion SPECT: a multicenter study. JACC Cardiovasc Imaging. (2018) 39. Arsanjani R, Dey D, Khachatryan T et al (2015) Prediction of revascularization after myocardial perfusion SPECT by machine learning in a large population. J Nucl Cardiol 22 (5):877–884 40. Weng SF, Reps J, Kai J, Garibaldi JM, Qureshi N (2017) Can machine-learning improve cardiovascular risk prediction using routine clinical data? PLoS ONE 12(4):e0174944. https://doi.org/10.1371/journal.pone.0174944 41. Sherman RE, Anderson SA, Dal Pan GJ et al (2016) Realworld evidence—what is it and what can it tell us? N Engl J Med 375:2293–2297 42. McConnell MV, Shcherbina A, Pavlovic A et al (2017) Feasibility of obtaining measures of lifestyle from a smartphone app: the MyHeart counts cardiovascular health study. JAMA Cardiol 2:67–76 43. Weng C, Kahn MG (2016) Clinical research informatics for big data and precision medicine. Yearb Med Inf 211–218. http://dx.doi.org/10.15265/IY2016-019 44. Raghupathi W, Raghupathi V (2014) Big data analytics in healthcare: promise and potential. Health Inf Sci Syst 3. http://dx.doi.org/10.1186/2047-2501-2-3 45. Salas-Vega S, Haimann A, Mossialos E (2015) Big data and health care: challenges and opportunities for coordinated policy development in the EU. Health Syst Reform 1:285–300. https://doi.org/10.1080/23288604.2015.1091538

Clinical Big Data Predictive Analytics Transforming Healthcare

561

46. Cox M, Ellsworth D (1997) Application-controlled demand paging for out-of-core visualization. In: Proceedings of the Visualization 1997 (Cat. No. 97CB36155), pp 235– 244. https://doi.org/10.1109/visual.1997.663888 47. Bian J, Maslov D, Angel Mayer M, Seevanayanagam S, Toldo L (2016) Challenges and opportunities of big data in health care: a systematic review. J Med Internet Res 4. http://dx. doi.org/10.2196/medinform.5359 48. Ghani KR, Zheng K, Wei JT, Friedman CP (2014) Harnessing big data for health care and research: are urologists ready? Eur Urol 66:975–977. https://doi.org/10.1016/j.eururo.2014. 07.032 49. Wu J, Li H, Cheng S, Lin Z (2016) The promising future of healthcare services: when big data analytics meets wearable technology. Inf Manag 53:1020–1033. https://doi.org/10. 1016/j.im.2016.07.003 50. Mehta N, Pandit A (2018) Concurrence of big data analytics and healthcare: a systematic review. Int J Med Inf 57–65. https://doi.org/10.1016/j.ijmedinf.2018.03.013 51. Bellazzi R (2014) Big data and biomedical informatics: a challenging opportunity. Yearb Med Inf 9(1):8–13 52. Keen J, Calinescu R, Paige R, Rooksby J (2013) Big data + politics = open data: the case of health care data in England. Policy Internet 5(2):228–243 53. Curtis LH, Brown J, Platt R (2014) Four health data networks illustrate the potential for a shared national multipurpose big data network. Health Aff 33(7):1178–1186 54. Edwards A, Hollin I, Barry J, Kachnowski S (2010) Barriers to cross-institutional health information exchange: a literature review. J Healthc Inf Manag 24(3):22–34 55. Elliot T, Holmes JH, Davidson AJ, La Chance PA, Nelson AF, Steiner JF (2013) Data warehouse governance programs in healthcare settings: a literature review and a call to action. eGEMS 1(1):1–7 56. Sinclair S, Norris JM, McConnell SJ et al (2016) Compassion: a scoping review of the healthcare literature. BMC Palliat Care 15:6 57. Karlberg M, Skaliotis M (2013) Big data for official statistics: strategies and some initial European applications. In: Conference of European statisticians, 25–27 September, Geneva, Switzerland. http://www.unece.org/fileadmin/DAM/stats/documents/ece/ces/ge.44/2013/ mgt1/WP30.pdf 58. Stempniak M (2014) Beyond buzzwords: two state hospital associations collaborate around big data. Hosp Health Netw 88:18 59. PatientsLikeMe (2018). https://www.patientslikeme.com. Accessed 10 Dec 2018 60. Archenaa J, Anita EAM (2015) A survey of big data analytics in healthcare and government. Procedia Comput Sci 50:408–413. https://doi.org/10.1016/j.procs.2015.04.021 61. Belle A, Thiagarajan R, Soroushmehr SMR, Navidi F, Beard DA, Najarian K (n. d.) Big data analytics in healthcare. https://doi.org/10.1155/2015/370194 62. van Allen EM, Wagle N, Levy MA (2013) Clinical analysis and interpretation of cancer genome data. J Clin Oncol 31:1825–1833 63. Andre F, Mardis E, Salm M, Soria JC, Siu LL, Swanton C (2014) Prioritizing targets for precision cancer medicine. Ann Oncol 25:2295–2303 64. Schultz T (2013) Turning healthcare challenges into big data opportunities: a use-case review across the pharmaceutical development lifecycle. Bull Assoc Inf Sci Technol 39:34– 40. https://doi.org/10.1002/bult.2013.1720390508 65. https://www.pfizer.com/ 66. https://www.mckinsey.it/ 67. http://www.mountsinai.org/

A Study of Different Techniques in Educational Data Mining Nadia Anjum(&) and Srinivasu Badugu Department of Computer Science and Engineering, Stanley College of Engineering and Technology for Women, Hyderabad, India [email protected], [email protected]

Abstract. Educational Data Mining (EDM) is a dawning interdisciplinary research field that concerns with the development of tools/methods to analyze enormous amount of data generated by or related to an educational framework or system. Computational approaches may be employed to explore the educational data and study the educational queries. This paper surveys the important studies/debates carried out in EDM. It talks about the various components that form a part of the EDM system, and lists the goals of EDM. Firstly, it identifies the different tasks that can be applied in educational environment. It then provides the most common tasks/problems in the educational system that have been solved through data mining (DM) techniques. It also compares the different techniques employed in terms of the merits and demerits. Keywords: Educational data mining (EDM)
Data mining
Classification Clustering
Sequential pattern mining
Association rule mining




1 Introduction EDM is an emerging field. Huge amounts of information generated in educational systems can be availed to find the patterns in the educational data. EDM deals with developing techniques to analyze the various types of data in educational environment and, to better contemplate the learners and the settings in which they learn by making use of these techniques [1]. Conversion of raw data from educational systems into instrumental information may be used by instructors, software developers, teachers, researchers, etc. The EDM process pictorially can be depicted as- (Fig 1). In the figure below, the data must be either be made available through public repositories or can be generated by an educational environment. This data, then is picked up for performing the following three steps: Pre-processing : First step in the process in which the data from the educational system is pre-processed to convert it into required format such that mining techniques can be applied. The different pre-processing techniques are noise reduction, data cleaning, attribute selection, etc. Data mining: The second step in the EDM process, it is an intermediary step using which the data mining techniques are applied to the pre-processed data. The different DM techniques used are: Clustering, Analysis, Visualization Regression, Classification, etc. © Springer Nature Switzerland AG 2020 S. C. Satapathy et al. (Eds.): ICETE 2019, LAIS 4, pp. 562–571, 2020. https://doi.org/10.1007/978-3-030-24318-0_65

A Study of Different Techniques in Educational Data Mining

To Pick up

Data

To Apply

Educational Environment

Preprocessing

Data Mining

563

PostProcessing

Knowledge

Fig. 1. Educational data mining process

Post-processing: It is the concluding step in which the outcomes or model gathered are interpreted and used to form decision regarding the educational environment. After the steps are performed the patterns are found and can be used to apply in the educational environment so that the performance of the students and various activities can be improved. The study of educational data is an old practice but the contemporary advances in educational technology have led to concern towards developing techniques for analyzing enormous quantity of data that is being generated in educational environment. EDM workshops were held from 2000–2007 as part of several international research conferences [2]. In 2008, an annual international research conference on EDM was founded. The first conference of annual international research was started in Montreal, Canada [3]. The charm in EDM attracted many academicians/researchers towards EDM, as a consequence the EDM researchers formed an academic journal in the year 2009, named as “the Journal of Educational Data Mining”, for partaking and propagating research outcomes. The International Educational Data Mining Society was established in the year 2011 by EDM researchers to connect the researchers from all around the world and to widen the scope of the field. As an outcome of the awakening of public educational data storage in 2008, such like the Pittsburgh Science of Learning Centre’s (PSLC) Data Shop and the National Center for Education Statistics (NCES), the educational data mining has become more feasible and easily reachable, the public data sets have contributed to EDM’s growth [1]. This survey is structured as stated below: Sect. 2, lists the goals of the EDM. Section 3, provides a brief about the applications of EDM. Section 4, lists the common tasks in educational data mining and the different data mining techniques that can be employed. Section 5, discusses some of the most noticeable future research lines. Lastly, conclusions are laid out in the Sect. 6.

2 Goals of Educational Data Mining Baker and Yacef [4] provides the succeeding four goals of EDM: (i) Predicting student’s future learning behaviour (ii) Discovering/improving domain models

564

N. Anjum and S. Badugu

(iii) Studying the effects of educational support (iv) Advancing scientific knowledge about learning and learners (Fig 2).

Prediction student’s performance Discovering domain models

Goals Of EDM

Educational Support

Scientific knowledge

Fig. 2. Goals of EDM

Predicting student’s future learning behaviour – By considering the student’s behaviour, this goal can be achieved by creating student models that assimilating the learner’s characteristics, including detailed facts such as their knowledge, behaviours, intelligence, and motivation to learn. Discovering/improving domain models – By the use of various methods and techniques in the applications of EDM, invention of alternatives and improvements to current models may be achievable. Educational support – It can be attained by learning systems. Scientific knowledge – The advancement in knowledge can be done by forming and integrating student models, the EDM research and the technology and software being used.

3 Applications of EDM Authors in [1] suggest four important areas of application for EDM: improving student models, improving domain models, studying the pedagogical support provided by learning software, and scientific research into learning and learners. (1) Improving Student Models: Students models are those models that give elaborative facts about a student’s characteristics or states, such as knowledge, motivation, meta cognition, intelligence, and attitudes. Modelling the individual contrasts between students, by allowing the software to react to those individual contrasts, is prominent theme in educational software research. These models are important for finding new patterns.

A Study of Different Techniques in Educational Data Mining

565

These models have proven to boost the ability to anticipate whether students are well informed academically or not and to anticipate student’s future achievement – including models of guessing and slipping into predictions of student’s future achievement has increased the accuracy of these predictions by up to 48% (Fig 3).

Fig. 3. Applications of EDM

(2) Improving Domain Models: Improvement in domain models may be attained by discovering techniques of domain models directly from the data. These methods have integrated psychometric modelling frameworks with highly evolved space-searching algorithms. It is used to predict whether student’s actions will be accurate or inaccurate in different domains. (3) Studying the pedagogical support: Modern educational software provides different types of pedagogical support to students. Finding which pedagogical support is the most effective has been an important area of interest for educational data miners. (4) Scientific Research into learning and learners: The final important section of application of the EDM is for scientific discovery about learning and learners. This takes different forms. By the application of EDM to answer questions in any of the three areas previously discussed (e.g. student models, domain models, and pedagogical support) can have broader scientific benefits.

4 Tasks and Techniques in EDM The following paragraphs deal with the different tasks and their objectives and the data mining methods. The different tasks are: (1) Analysis And Visualization Of Data: The aim is to draw attention towards useful information and support decision making. In web based education learning, instructors find the highly complex statistics in log data

566

N. Anjum and S. Badugu

too difficult to inspect or tedious to interpret. It may prove to be fruitful to the instructors in the educational environment to analyze the student’s activities and to use the information generated by student to get a general view of his/her learning experience. Statistical analysis of educational data (logs/files/databases) is effective in obtaining assessing reports. Visual representation permits users to understand, analyze the huge information all at once which may turn out to be useful in decision making support systems. (2) Predicting a student’s performance: The goal is to predict the unknown value of a variable/characteristic describing the student. In education the values normally predicted are performance, knowledge, score or mark. This value can be numerical/continuous value (regression task) or categorical/ discrete value (classification task). Prediction of a student’s performance is one of the oldest and most popular applications of Data Mining in education, and various techniques and models can been applied (neural networks (NN), Bayesian networks (BN), rule-based systems, regression). (3) Providing feedback for supporting instructors: The idea is to get the feedback in order to help the authors/administrators and teachers in better decision making so that the performance of students may be increased and the resources maybe used efficiently etc. It may also prove to be fruitful in taking appropriate actions like remedial classes etc. Several Data Mining techniques have been used in this task, however, association rule mining is the most common. Association rule mining shows intriguing relationships among attributes in large databases and presents them in the form of strong rules according to the different degrees of interest they might present. (4) Grouping students: The students are made into a similar groups called clusters wherein the students are categorized into clusters based on their features like personal characteristics, studying pattern, etc. Then, the cluster of students obtained may be used by the instructor/developer to build a personalized learning system, to promote effective group learning, to provide adaptive contents, etc. The DM techniques used for this task are classification and clustering. (5) Detecting a typical student behaviours: The goal of detecting a typical student behavior is to discover/detect those students who might posses some kind of problem or strange behavior such like: erroneous actions, depression, playing game, misuse, cheating, dropping out, academic failure, etc. Several DM techniques (mainly, classification and clustering) have been used to reveal these types of students in order to provide them with appropriate help in stipulated time. (6) Recommendations for students: Recommendations are made directly to the students according to their activities such as studying pattern, etc. Various data Mining methods are used for this task but the most common are association rule mining, clustering and sequential pattern mining. Students may avail of this recommender system and use it for their benefit.

A Study of Different Techniques in Educational Data Mining

567

(7) Social network analysis Social Networks Analysis (SNA): The main aim is to analyze the connections/relationship among individuals. A social network is considered to be a group of people, an organization or social individuals who are connected by social relationships like friendship. Various DM techniques have been used to mine social networks in educational environments however, collaborative filtering is the most common (Table 1). Table 1. Different Tasks and techniques in EDM S. Tasks in EDM no 1 Analysis And Visualization Of Data: The main aim is to draw attention towards useful information and support decision making. In web based education learning, instructors find the highly complex statistics in log data too difficult to inspect or tedious to interpret 2 Predicting a student’s performance: The aim is to predict the unrecognized value of a variable/characteristic describing the student

3

Techniques (i) Statistics (ii) Visualization

(i) Regression analysis finds the relationship between a dependent variable and one or more independent variables [7] (ii) Classification is also used to predict the student’s performance [8] (iii) Bayesian network have been used to predict student academic performance [9] (iv) Correlation Analysis have been applied for anticipating school students’ probabilities of success in university [10] (v) Neural Networks (i) Association rule mining is used to Providing feedback for supporting confront the problem of continuous instructors: The idea is to get the feedback in order to feedback in the educational process [11] help the authors/administrators and teachers (ii) Association analysis, case-based in better decision making process so that the reasoning is employed to structure the course performance of students may be increased material and to give assignments at different and the resources maybe used efficiently etc. levels of difficulty [12] It may also prove to be fruitful in taking appropriate actions like remedial classes etc. Several Data Mining techniques have been used in this task, although association rule mining has been the most common. Association rule mining reveals interesting relationships among variables in large databases and presents them in the form of strong rules according to the different degrees of interest they might present (continued)

568

N. Anjum and S. Badugu Table 1. (continued)

S. Tasks in EDM no 4 Grouping students: The students are made into a similar groups called clusters wherein the students are categorized into clusters based on their features like personal characteristics, studying pattern, etc. Then, the cluster of students obtained can be used by the instructor/developer to build a personalized learning system, to promote effective group learning, to provide adaptive contents, etc. The DM techniques used are classification and clustering

5

Detecting a typical student behaviours: The goal of detecting a typical student behavior is to detect those students who have some type of problem or strange behaviour such as: faulty actions, little motivation, depression, playing game, misuse, cheating, dropping out, academic failure, etc.

6

Recommendations for students: Recommendations are made directly to the students according to their activities such as studying pattern, the consecutive task to be done etc., Different data Mining techniques are used for this task but the most common are association rule mining, clustering and sequential pattern mining. Students may make avail of this recommender system and use it for their benefit Social Networks Analysis (SNA): Relationships among individuals is analyzed and a network of group of people is created, the people in the group are associated by any of the social relations

7

Techniques (i) Hierarchical agglomerative clustering, Kmeans and model based clustering to detect groups of students with akin skill profiles [13] (ii) Various classification techniques have been employed in order to group students, such as: DT organizes university students into 3 groups(low-risk, medium risk and high risk) [14] (iii) NB classifier to classify learning or reading styles that describe learning behavior and course content [15] (iv) Grouping students based on their profiles in a peer review content can be achieved by making use of genetic algorithms [16] (i) Several clustering techniques like kmeans have been used [13] (ii) Genetic Algorithms have also been used [16] (iii) Random Forests are used to determine the actions of the students [14] (iv) Association rule mining [11] (v) Classification Techniques used to anticipate whether a student would drop out or not (i) Sequential pattern mining is applied for individual recommendations on the learning content based on reading style and web usage patterns [17] (ii) Association rule mining is used in online learning environment to prescribe online learning activities or shortcuts on a course web site [18]

(i) Collaborative filtering or social filtering is a process of making automatic predictions about the concerns of a user [19] It gives personal recommendations by calculating the similarity

A Study of Different Techniques in Educational Data Mining

569

In the below Table 2, we provide a comparison of different techniques that were used in achieving the tasks (prediction of student’s performance, grouping students, providing feedback) of EDM in terms of merits & demerits. Table 2. Comparison of EDM tasks and techniques. References Techniques used [6] Visualization Of Data: The aim is to draw attention towards prominent information and support the instructors in decision making In e-learning environment teachers find the huge statistics in log burdensome to inspect/analyze and tedious to interpret [8, 9] Classification: The different classification algorithms used are decision trees, Bayesian networks

Merits Demerits Helps in effective report assessing Visual representation permits users to understand, analyze the huge information all at once which may turn out to be effective in decision making support systems

It was seen that DT works well with discrete or categorical data C4.5 has a great amalgam of error rate and speed Bayesian network (BN) takes into consideration the prior information It is seen that linear regression has proven to be competent of anticipating student’s academic performance much more effectively

[7]

Regression: Regression techniques applied on datasets available publicly

[13]

Clustering: It is used to identify common traits among the students Association rule mining: Identifies weak students Used to furnish feedback so that remedial classes can be provided to the instructors. It represents relationship among variable in enormous datasets

[11, 12, 18]

Decision trees showed higher accuracy when compared to Naive bayes in certain cases BN is irrelevant for datasets with an ample number of features and the numerical features must be normalized Linear regression describe only linear relationship if there exists a non linear relationship then we may have a poor working model Overfitting is again a problem K must be specified by the instructor so as to group the students in specific categories May discover too many unwanted rules Rules discover may be arduous to extrapolate in the case of huge datasets (continued)

570

N. Anjum and S. Badugu Table 2. (continued)

References Techniques used [17] Sequential Pattern Matching: It finds the relationships between Sequential events to know whether there is a sequence in their occurrences

Merits Personalized recommender system based on individual’s interests, learner’s habits etc.

Demerits

5 Future Work An ample of future work that can be thought of to work on but, the most important or prominent areas can be EDM tools that must be designed to be easier for the educators and instructors with very less or no complexity and more flexibility. Tools must be designed specifically for applying the DM algorithms. The other area is integration with e-learning systems and standardization of methods and data to mine data for a specific purpose. Focus must be more on making data available i.e., public domain datasets. At present, there is a single public educational data repository, the PSLC Data Shop that has an ample of educational data sets from across the globe and also supports analysis of those data sets.

6 Conclusion This paper gives a concise description of the different methods involved in educational data mining. However, the focus is not just on the methods involved but also the type of data and consequentially, by the kind of educational task that they solve. The techniques may be utilized in more than one task. EDM is an ever growing and one of the most promising fields using which the standards of education can be enhanced both online and offline, the student’s performance can be enhanced by providing recommendations. EDM can be helpful to authors, instructors, researchers, academicians, faculty by getting a feedback etc.

References 1. Baker R (2010) Data mining for education. In: McGaw B, Peterson P, Baker E (eds.) International encyclopedia of education, 3rd edn. Oxford, UK Elsevier 2. Romero C, Ventura S (2010) Educational data mining: a review of the state-of-the-art. IEEE Trans Syst Man Cybern Part C Appl Rev 40(6):601–618 3. “http://educationaldatamining.org/EDM2008/. Accessed 04 Sept 2013 4. Baker RS, Yacef K (2009) The state of educational data mining in 2009: a review and future visions. JEDM-J. Educ Data Min 1(1):2017

A Study of Different Techniques in Educational Data Mining

571

5. Castro F, Vellido A, Nebot A, Mugica F (2007) Applying data mining techniques to elearning problems. In: Jain LC, Tedman R, Tedman D (eds.) Evolution of teaching and learning paradigms in intelligent environment. studies in computational intelligence, vol 62, Springer, pp 183–221 6. Cristóbal R, Sebastian V (2007) Educational data mining: a survey from 1995 to 2005. Expert Syst Appl 33:135–146. https://doi.org/10.1016/j.eswa.2006.04.005 7. Draper NR, Smith H (1998) Applied regression analysis. Wiley, Hoboken 8. Phyu TN (2009) Survey of classification techniques in data mining. In: International multi conference of engineers and computer scientists, Hong Kong, pp 1–5 9. Haddawy P, Thi N, Hien TN (2007) A decision support system for evaluating international student applications. In: Frontiers in education conference, Milwaukee, pp 1–4 10. Mcdonald B (2004) Predicting student success. J Math Teach Learn 1–14 11. Psaromiligkos Y, Orfanidou M, Kytagias C, Zafiri E (2009) Mining log data for the analysis of learners’ behaviour in web-based learning management systems. Oper Res J 1–14 12. Sheard J, Ceddia J, Hurst J, Tuovinen J (2003) Inferring student learning behaviour from website interactions: a usage analysis. J Educ Inf Technol 8(3):245–266 13. Ayers E, Nugent R, Dean N (2009) A comparison of student skill knowledge estimates. In: International conference on educational data mining, Cordoba, Spain, pp 1–10 14. Superby JF, Vandamme JP, Meskens N (2006) Determination of factors influencing the achievement of the first-year university students using data mining methods. In: International conference on intelligent tutoring systems, educational data mining workshop, Taiwan, pp 1–8 15. Kelly D, Tangney B (2005) First aid for you: getting to know your learning style using machine learning. In: IEEE international conference on advanced learning technologies, Washington, DC, pp 1–3 16. Crespo RM, Pardo A, Pérez JP, Kloos CD (2005) An algorithm for peer review matching using student profiles based on fuzzy classification and genetic algorithms. In: International conference on innovations in applied artificial intelligence, Bari, Italy, pp 685–694 17. Zhang L, Liu X, Liu X (2008b). Personalized instructing recommendation system based on web mining. In: International conference for young computer scientists, Hunan, China, pp 2517–2521 18. Zaïane O (2002) Building a recommender agent for e-learning systems. In: Proceedings of the international conference in education, Auckland, New Zealand, pp 55–59 19. Herlocker J, Konstan J, Tervin LG, Riedl J (2004) Evaluating collaborative filtering recommender systems. ACM Trans Inf Syst J 22(1):5–53

A Study on Overlapping Community Detection for Multimedia Social Network Sabah Fatima(&) and Srivinasu Badugu Department of Computer Science and Engineering, Stanley College of Engineering and Technology for Women, Hyderabad, India [email protected], [email protected]

Abstract. For studying the network system the important task is community detection as it provides the overall information of network structure. Community structure shows that there is a relationship between individuals of that community and is found in many social networks which is an interesting feature for research. There are communities that tend to overlap as there are nodes that may belong to multiple communities at the same time which makes the task even more challenging. This paper reviews overlapping community detection techniques. Keywords: Overlapping community detection Complex networks
Community structure


Online social networks


1 Introduction The concept of social network has become popular after websites like Facebook and Google+ emerged and become a part of our everyday life. Entities and the relationships within these entities participating in the network are the two main properties of the social networks. Entities might be “people” and relationships might be the “friendship” of these people like on Facebook and like most of the other social websites but they are not limited to “people” and “friendship”. Entities might be entirely different e.g., organizations, websites and relationships might be something else e.g., business, trade, collaboration. For decades, social networks and their analysis has been a very popular research area since the recent revolution on the internet. There is a need for changing how to handle analyzing and processing networks as the size of Real world networks are very large, reaching billion of nodes for which large number of methods have been proposed. In a randomly generated network, there is homogeneous edge distribution is mostly homogeneous and similar vertex degree. However, the degree distributions of real networks are not homogeneous, edges might be denser within some group of entities and might be rarer within other group of entities. This feature of real networks that edges within some specific group are denser, is called community structure or clustering. The entities of a social network naturally fall into communities which the relationships within a community is dense while the relationships between different communities are rare.

© Springer Nature Switzerland AG 2020 S. C. Satapathy et al. (Eds.): ICETE 2019, LAIS 4, pp. 572–578, 2020. https://doi.org/10.1007/978-3-030-24318-0_66

A Study on Overlapping Community Detection for Multimedia Social Network

573

Examples of network are: people that are friends, computers that are connected, and web pages that point to each other. In graph theory, vertices represent entities and edges represent the links. Large graph of real life are called complex network. Many real world problems dataset can be modeled with networks. To represent the relations in data a powerful mathematical tool is provided by the networks. The real world dataset is divided depending on the networks generated into social, information, technological, and biological networks. A social network is a network connecting the people who contact or interact with each other. Social networks are not limited to “online social networks” such as Facebook, Twitter, or LinkedIn. An information network is a network of entities that contains the information of the network such as World Wide Web, word co-occurrence networks, citation networks etc. A technical network is a network made by man such as the airline routes, Internet, railways, the electric power grid, and road networks. A biological network represents a biological system such as a network of metabolic pathways, protein-protein interactions, the food web, and the network of blood vessels. Examples of Complex network are: Internet, Web pages, Food webs, Http, Telecom networks, Biological networks, Social Networks etc [1]. Online social networks have attracted a lot of attention as it has become popular means of communication today. In today’s world every person has a account on Social Networking Sites (SNS) such as Facebook, Twitter, Youtube etc. People communicate to others using these social media sites and has attracted a lot of attention in recent years for research focus under Social Network Analysis domain. Graph is a convenient tool for modeling network. If we define a graph as G = (V, E) then V is the set of nodes and E is the set of edges that if an edge exists between the nodes Vi and Vj then we can say that nodes have link to each other and that there is an edge between them. While modeling social networks as graphs, an entity is modeled as a node and the relationship that connecting two entities are modeled as an edge. A graph can be used to represent social network, where people are represented as nodes and the edges between nodes represent link between the people in social network. Formation of communities is an interesting feature found in social networks. A community is a group of individuals communicating frequently with each other than with people outside the group in a social network. For eg: If Social network is represented as a graph G (V, E), then the individuals are represented by vertex V and edge E represent the links among them, then a if a community is given as CG edges inside the community C is more than edges outside the community. Communities in social networks, tends to overlap with each other which means that a node which is a person of one community can also be a member of another community. Existence of numerous small communities in a network makes it a complex problem for identifying communities. Community: Community is a group of nodes that have some common properties and have common role in organization. Group of nodes in the same community are densely connected than those outside the community [2]. Types of Communities: There are two types of communities. They are: (a) Disjoint community: In disjoint community is a community where a node belongs to single community. Disjoint community there is a one to one relationship being held between a

574

S. Fatima and S. Badugu

node and a community . A node belongs to at-most 1 community and at-least 0 community (none) [3]. (b) Overlapping Community: In this type community, a node belongs to muliple communities at the same time [4] or have more than one community in common. In social network extracting such community structure is very useful as it helps us to study the overall structure of network. Community detection has several application areas in the real world. It is beneficial in commercial, security and academic areas. Recommending same products or services to individuals who are in the same community and using community of the individual as a feature for the recommendation systems are two very common and known application types. Another application area of community detection in social networks is that, usage of community detection in social networks to reveal the fraud events and other suspicious leakages of money is proposed by generating a network of customers using the text messages and telephone communications between the individuals and identifying community structures. It is stated that unexpected communications between individuals and their type of social structure can enable us the necessary information to find suspicious groups or individuals. For another example in the academic area, we can show that dividing citation network into communities can help researchers who are looking for a cooperation for a specialized field. There are also studies to detect hidden criminals in the networks that we do not have any or have too little prior knowledge about individuals’ identity. In this kind of networks, since there are not much data the to characterize the individuals, the relationships between the entities become important. The relationships in criminal networks in this type of studies are built from several resources like police arrest data, crime location data, kinship or hometown data.

2 Literature Survey A great deal of work has been devoted to study the structure and dynamics of networks generated from real-world data. These networks are not random networks and the nodes in these networks are organized into specific structures. A wide variety of network mining methods and algorithms exists which can be used to uncover the structure of such networks. Community detection is an imperative field of research. There are different calculations for network recognition however the majority of the calculations fail in discovery of covering networks they can just distinguish disjoint networks. Some of the algorithm proposed are graph partitioning algorithms, hierarchical clustering, multilevel graph partitioning algorithms, divisive algorithms etc [5]. Palla et al. in 2005 [6] presented the algorithm for discovering communities that are overlapping. K-cliques are used to identify the communities where a node may belong to multiple communities. Subset of nodes in which each node is adjacent to each node. Size of the clique is represented by k-clique. Eg. subgraph with 6 nodes means it a 6-Clique. Raghavan et al. in 2007 [7] proposed Label propagation algorithm for detecting communities. Its extremely fast method for finding community. Label propagation algorithm can be used for both detecting both disjoint as well as overlapping

A Study on Overlapping Community Detection for Multimedia Social Network

575

communities. The concept behind this algorithm is that nodes takes the label that majority of its neighboring nodes have and labels are used in forming the communities. Shen et al. in 2008 [8] proposed algorithm which is agglomerative hierarchical clustering method for detecting overlapping community. Eagle algorithm uses set of maximal cliques instead of set of nodes. This algorithm is based on two stages. In the first stage, dendogram is created and in the second stage appropriate cut method is selected to partition this dendogram into communities. Lancichinetti et al. in 2009 [9] presented a method which reveals both covering network and hierarchical complex network properties. This algorithm is based on fitness function local optimization. This algorithm maximizes fitness value for detecting overlapping communities. Peaks in the fitness function of the histogram is used for finding structure of the community. Shen et al. in 2009 [10] proposed another approach for detecting overlapping community. Depending on number of maximal cliques they are used for finding overlapping communities. Modularity optimization method is used to find the overlapping communities by partitioning the network of maximal clique. With the measure Qc, to identify the overlapping community structure of network we find the optimal cover with maximum Qc. Maximal clique network is constructed from the original network to determine the optimal cover. Experimental tests were done with known community structure on both the real world and artificial networks Gregory in 2009 [3] presented algorithm for discovering communities that are overlapping community in two phases. In the first phase split betweeness is used to form new network from the existing network by splitting node. In second phase algorithm for detecting disjoint community is applied to new network. Therefore it converts the disjoint community to overlapping community detection. The author proposed two phase approach for discovering overlapping communities. Experiment is done on eight real world networks like protein-protein, PGP, email etc as well as synthetic networks. Ahn et al. in 2010 [11] presented link partition for finding the overlapping communities. The existing algorithms focus mainly on grouping nodes so link communities are used as it reveals the hierarchical organisation of the network while preserving the overlap. A dendrogram is built using Hierarchical clustering where a link from the original network is a leaf a and then partition links in dendogram. Then the link communities are extracted when the dendogram is cut at certain threshold point to detect overlapping communities. Partition density is used as a measure for modularity in this algorithm. The algorithm was tested in group of 11 networks. These networks vary from small to large, from sparse to dense, and from those with modular structure to those with highly overlapping structure. Chen et al. in 2010 [12] proposed algorithm for weighted network for discovering the communities that are overlapping. It expands partial community using the algorithm starting from a single node to identify overlapping communities. Here the main stratergy is to discover the partial community of a node and expand the partial community by adding tight nodes from node having maximum node strength. The data sets used to evaluate the algorithm are seven complex networks from real world and one synthetic network. The proposed algorithm is found to be efficient for discovering communities that are overlapping in weighted networks.

576

S. Fatima and S. Badugu

Lazar et al. in 2010 [13] proposed algorithm for identifying overlapping communities based on difference between edges that are inward and outward. This approach is based on ranking of partitions of network nodes in overlapping communities. It uses optimization methods for finding the overlapping community structure. Here non-fuzzy measure is used that quantifies cluster structures. The algorithm is designed to discover the overlapping community-structure of networks and it uses tuning-parameter (k). The modules which are formed are called “k-clique communities” and is union of k-cliques that can be formed by group of k-cliques that are adjacent. Theoretically k can be any positive integer ranging from 3, but it is usually smaller than ten practically. Coscia et al. in 2012 [14] presented an algorithm to identify communities that are overlapping. Label propagation algorithm is used for voting of nodes in a community and then the communities found locally are merged to form global. This is achieved by allowing every node to vote for the communities and to limit the global view. This method can be used for large scale network as it has limited time complexity. The experiment is done on the state-of-the-art overlapping and non-overlapping community discovery methods, and it is found that the algorithm outperforms the others in the quality of the communities discovered. Li et al. in 2013 [15] proposed algorithm for discovering overlapping communities which is applied to weighted networks. In the first step, seed communities are detected. Then degree function is used for finding more communities. This algorithm is successful in detecting the nodes belonging to multiple communities. Badie et al. in 2013 [16] proposed algorithm for finding both disjoint and overlapping communities. It takes the concept of closenses of the node and improving the label propagation algorithm result. He defines Communities as groups of nodes which are strongly connected units in networks. First we compute link to link algorithm for all links in netwwork and then assign initial labels to nodes of the network graph. Then we select the hubs and assign agents to the nodes in the selected hubs. Agents explore the input network. Then community cores are determined based on the labels and overlapping nodes through thresholding are identified and add overlapping nodes to appropriate cores. Finally we delete communities that are completely nested inside a bigger community. Yaozu Cui et al. in 2014 [17] presented an algorithm for overlapping communities detection in complex network. The author utilized different types of theories i.e. subgraph maximum between the two neighboring communities and clustering coefficient. The sub-graphs maximum are removed in the first step from the original networks and then they are combined on the basis of clustering coefficient of two neighboring maximal sub-graphs. To enhance the algorithm, new extended modularity is used as an additional feature. This algorithm helps in discovering the overlapping vertex. This paper also covers the comparison of results of the algorithm is with other correlated algorithms. Zhou et al. in 2015 [18] presented a hierarchical gamma process is proposed to detect unweighted community. To factorize the binary adjacency matrix undirected network it uses infinite edge partition model. The model is scalable to big sparse networks and computes on pair of linked nodes. It predicts missing edges as well as

A Study on Overlapping Community Detection for Multimedia Social Network

577

discover overlapping communities. Data augmentation techniques are used for revealing the number of communities. Whang et al. in 2016 [19] presented an efficient overlapping community detection algorithm. It used an approach based on a seed expansion. The basic idea is to find good seeds, and then based on community metric expands the seeds greedily. This method is used to determine the good seeds in graph. Community score is optimized by the developed seeding stratergies. Neighborhood inflation step is an important step in this method where to represent their entire vertex, neighborhood seeds are modified. Wen et al. in 2017 [20] presented an algorithm for overlapping community detection. This is maximum cliques algorithm which is based on multiobjective evolutionary algorithm (MOEA) where a maximal-clique graph is presented as new representation scheme. the same nodes are allowed to share two maximum cliques of the original graph and Set of maximal cliques of original graph as nodes defines maximal clique graph.

3 Conclusion Overlapping community detection approaches have attracted a lot of attention of researchers in recent years. Analyzing community structure in social network has emerged as a topic of growing interest as it shows the interplay between the structures of the network and its functioning. This paper tries to review different papers for overlapping community detection.

References 1. Newman MEJ (2003) The structure and function of complex network, vol 2, pp 167–256 2. Steinhaeuser K, Chawla NV (2008) Community detection in large real world networks 3. Gregory S (2009) Finding overlapping communities using disjoint community detection algorithms. In: Complex networks, Springer, pp 47–61 4. Xie J, Kelley S, Szymanski BK (2012) Overlapping community detection in networks: the state of the art and comparative study, arXiv preprint arXiv 5. Charu C Aggarwal (2011) Social network data analytics 6. Palla G, Derényi I, Farkas I, Vicsek T (2005) Uncovering the overlapping community structure of complex networks in nature and society. Nature 435(7043):814 7. Raghavan UN, Albert R, Kumara S (2007) Near linear time algorithm to detect community structures in largescale networks. Phys Rev E 76(3):036106 8. Shen H, Cheng X, Cai K, Hu M-B (2008) Detect overlapping and hierarchical community structure in networks, November 2008 9. Lancichinetti A, Fortunato S, Kertész J (2009) Detecting the overlapping and hierarchical community structure in complex networks. New J Phys 11(3):033015 10. Shen H-W, Cheng X-Q, Guo J-F (2009) Quantifying and identifying the overlapping community structure in networks. J Stat Mech: Theory Exp 11. Ahn Y-Y, Bagrow JP, Lehmann S (2010) Link communities reveal multiscale complexity in networks. Nature 466(7307):761–764

578

S. Fatima and S. Badugu

12. Chen D, Shang M, Lv Z, Fu Y (2010) Detecting overlapping communities of weighted networks via a local algorithm. Phys A: Stat Mech Appl 389(19):4177–4187 13. Lázár A, Ábel D, Vicsek T (2010) Modularity measure of networks with overlapping communities. EPL (Europhys Lett) 90(1):18001 14. Coscia M, Rossetti G, Giannotti F, Pedreschi D (2012) Demon: a local-first discovery method for overlapping communities, ACM 15. Li J, Wang X, Eustace J (2013) Detecting overlapping communities by seed community in weighted complex networks, 1 December 16. Badie R, Aleahmad A, Asadpour M, Rahgozar M (2013) An efficient agent-based algorithm for overlapping community detection using nodes closeness. Phys A: Stat Mech Appl 392 (20):5231–5247 17. Cui Y, Wang X, Li J (2014) Detecting overlapping communities in networks using the maximal subgraph and the clustering coefficient. Phys A 405:85–91 18. Zhou M (2015) Infinite edge partition models for overlapping community detection and link prediction. Comput Sci 1135–1143 19. Whang JJ, Gleich DF, Dhillon IS (2016) Overlapping community detection using neighborhood-inflated seed expansion. IEEE Trans Knowl Data Eng 28(5):1272–1284 20. Wen X et al (2017) A maximal clique based multiobjective evolutionary algorithm for overlapping community detection. IEEE Trans Evol Comput 21(3):363–377

A Review on Different Question Answering System Approaches Tahseen Sultana(&) and Srinivasu Badugu Department of Computer Science and Engineering, Stanley College of Engineering and Technology, Hyderabad, India [email protected], [email protected]

Abstract. Question Answering systems (QASs) is a system that provide answers to the question or query asked by the user in the natural language. It retrieves small portion of text from the collection of document which contains the answer of the user’s question. Therefore to retrieve such an accurate and precise answer from the collection of document, Information Retrieval (IR) Techniques are required and to process or understand the user’s question posed in the natural language (NLP) Natural Language Techniques are used.In this survey paper we will see what exactly a Question Answering System is, previous work done on such Question Answering system and we will also compare research against each other with respect to the different approaches that were followed and components that were used. At the end, the survey gives a clear comparison between the different QASs and idea of the our proposed QAS model. Keywords: QAS (Question Answering System)
NLP (Natural Language Processing)
Information Retrieval (IR) Passage retrieval
TF-IDF




1 Introduction Question answering (QA) System is a computer science discipline which belong to the fields of information retrieval and natural language processing [18] (NLP). These systems are concerned about the creation of the systems that can automatically answer questions posed by humans in a natural language [1]. Question Answering System is a multi-disciplinary field. That means it is a collection of several academic discipline such as Artificial Intelligence, Natural Language Processing, Information Retrieval (Fig. 1). A QA implementation, is a computer program, that construct the required answers by querying or scanning a structured database of knowledge or information, which is called as a ‘knowledge base’. Usually, QA systems can retrieve answers from an unstructured collection of natural language documents [18]. Some of the examples of Collection of documents in Natural Language used for QA systems include:

© Springer Nature Switzerland AG 2020 S. C. Satapathy et al. (Eds.): ICETE 2019, LAIS 4, pp. 579–586, 2020. https://doi.org/10.1007/978-3-030-24318-0_67

580

T. Sultana and S. Badugu

Fig. 1. General Question Answering system (source: [2])

• • • • •

A subset of World Wide Web pages A local collection of reference texts Compiled newswire reports Internal organization documents and web pages A set of Wikipedia pages [18].

The Text Retrieval Conference (TREC), a conference series which is co-sponsored by NIST, They has initiated the Question-Answering Track early in 1999, which used to test the systems’ ability to retrieve short or concise text snippets in response to the factoid questions shows a growing need for It revealed more sophisticated search engines able to extract the specific piece of information that could be considered as the best possible answer for the user question [3]. Question Answering (QA) is a rapidly growing research area that combines research from different, but related, academics such as Information Extraction (IE), Natural Language Processing (NLP). Information Retrieval (IR). Question Answering (QA), in Information Retrieval (IR), is a task of automatically retrieving the answers to a user’s question posed in natural language (NL) using either a pre-structured database or collection of natural language documents. However, there are many search engines available [19]. All these search engines have great success and have remarkable capabilities, but the problem with these search engines is that instead of giving a direct accurate and precise answer to the user’s query or question they usually provide list of document related to websites which might contain the answer of that question so, in order to achieve the required information or answer the user have to go through all the website, document or file listed by the search engine. QA research is concerned with a board range of question types including: fact, definition, list, Why, How, semantically constrained, hypothetical, and cross-lingual question

A Review on Different Question Answering System Approaches

1.1

581

Types of Question Answering System (QASs)

• Closed-domain question answering systems deals with specific questions which belong to a specific domain (for example, Sport, Medicine, Education, Entertainment etc.), and can be seen as an simpler task as NLP systems can only exploit domain-specific knowledge which is frequently formalized in ontologies. ClosedDomain QASs will be restricted to the specific domain such system cannot answer the question asked beyond the selected domain. On the other hand, closed-domain can also be referred as a condition where only a limited type of questions are considered, such as questions asking for descriptive rather than procedural information. • Open-domain question answering systems deals with general Question. It can answer the questions within any domain, unlike the Closed-Domain QAS which is restricted to single domain, This system allows user to ask question from any domain. for example, The user can ask one question related to sports and the another question can be related to politics or education etc.. Such systems make use of world knowledge and general ontologies and these system will usually have huge amount of data available from which it retrieves the accurate answer. 1.2

Different Question Answering System Approaches

(1) Frequently Asked Question & Answer (FAQs) Approach: Question Answering system using FAQs is very simple approach, In which we collection of Question & Answer pairs are collected and stored as our dataset. Where Answers are produced by simply searching the given question from the stored Question Answer pair when the required query is found its respective answer is given back to the User. (2) Information Retrieval Approach: Information Retrieval Techniques for QAS is widely used concept. The idea of Question Answering system is to retrieve precise and accurate answer from collection of document, In order to extract such accurate answer people employ different IR techniques in many different ways. General steps include: A. B. C. D.

Preprocessing Question Analysis Document Retrieval Answer Extraction

(3) Machine Learning Approach: This Approach for the creation of QAS is quite similar to the previous approach which is Information Retrieval Approach. As, here we will be using few IR Techniques along with some Machine learning Classification Algorithm such as SVM (Support Vector Machine) to classify the question to its types.

582

T. Sultana and S. Badugu

2 Literature Survey Bhardwaj et al. in [4] discussed about a Question Answering System for Frequently Asked Questions in which they have created an Open domain Question Answering system that uses FAQs to answer the user’s question. They have implemented their approach using QA4FAQ from website which present in csv format as their dataset. In there paper [4] they have proposed an approach which combines two techniques such as Orthodox AND/OR searching with Combinatorics searching for searching the user’s posted query in the question answers pair list store as their dataset to retrieved the answer with respect to the given question. Fu in the [5] Introduce a QA system on Music using Database Ontology Knowledge where a user can ask any question about music. Here in the paper [5] the author has given two approaches to retrieve an answer they are First is FAQ module and Ontology Knowledge. Whenever a system gets an user’s query it will first check that query in FAQ module and if it present, then the paired answer with the matched question is extracted and given to the user else the system will go for the second method where it has to scan the ontology knowledge base for the appropriate answer for that it performs the following steps they are: Question Classification, Question Analysis and Answer Extraction. The author in [5] has stated that the first approach which is FAQ has given good result then the second approach. Ketsmur1 et al. in [6], have planned a brand new design to make a factoid question answering system which gives answer to the user’s question by scanning or interrogating a Knowledge Database. The database they have used is DBpedia ontology. DBpedia is a project aims at extracting structured content from the information created in the Wikipedia article [17]. The author in [6] has explained their approach which leads to the creation of QA system, i.e. first they have created an SPARL query. Now this query will be send to DBpedia server to go through the data present in the DBpedia and select the appropriate answer which will be send back to the user. Now to create that SPARQL query they have classify the question by creating Decision model using SVM (Support Vector Machine) algorithm [6, 14] which decide the type of the answer. Next, After determining the required Answer type from the Question next they will determine the main focus of the question for which they have used Resource Extraction and Keywords Extraction techniques. SPARQL query is created using Resources, Keyword set and Ontology classes. Resources, Keyword set they have already got from the previous step for the Ontology classes and properties of the given question is determine by executing a query. Now, Finally this query is send to the DBpedia server the result of query execution is an RDF file which is parsed to get our answer. Pragisha et al. in [7] has proposed their Question Answering System in Malayalam Language that means their system can answer any question asked by user in Malayalam. This is a closed domain QA system where their domain is Kerela sport for that they have stored collection of Malayalam document about Kerela Sport as their dataset. The implementation of this system begins with the Question Type Analysis module where they identify the Malayalam question word and their meaning. Next they perform Document Processing for that they have used Sentence tokenizer to split the document into sentences and stored in an array, Then they have perform ranking of the

A Review on Different Question Answering System Approaches

583

sentence based on its similarity with the user’s question after that the topped rank sentences called as Answer Candidate are selected. On that Answer candidate the task of Name Entity Recognizer is performed using TnT tagger. Now, Finally the last step is Answer Extraction where we identify the expected tag of the question word and that tag is consider as Answer key. Now if the Name Entity or Tag of Answer Candidate matches with the Answer key then the Answer Candidate with maximum matches will be selected as an Final Answer. Ryu et al. in [8] proposed a “Open domain question answering using Wikipediabased knowledge model” Which describes the use of Wikipedia as a huge source of knowledge for a question answering (QA) system with a multiple answer matching modules which is based on the different types of semi-structured knowledge sources of Wikipedia, Such as article structure, article content, category structure, info boxes, and definitions. These semi-structured knowledge sources have their own strengths in finding the accurate and precise answers for specific question types, such as info-boxes for factoid questions, definitions for descriptive questions and category structure for list questions. First it perform the Question analysis to analyze the user’s question to identify the nature of the answer being sought. Then it describe the Wikipedia QA System (Wikipedia Question Answering System) which comprises of several module namely Article content module, Article structure module, Category structure module, Info-box module, Definition module and finally Answer merging module. Zhang and Lee in their [9] proposed “A Web-based Question Answering system”. In which they have created an Open domain QA system which they named as LAMP. Here the author has utilizes the concept of Snippet present in result returned from search engine. Next the author in [9] has explained the steps, their proposed system LAMP takes the following steps when it receive a question from the user. Firstly the system will send the question to Search engine and takes top 100 search result. For each result it create a Snippet which is nothing but Title, URL, String segment of every search result given by the search engine. The next step they performed is to classify the question to identify the expected answer type for that they have used Support Vector Machine (SVM) Algorithm [15]. After that they have used HMM (Hidden Markov Model)- based named entity recognizer [16] to identify the type or name entity of the Snippets then only the Snippets whose name Entity or type matches with the type of the question are seleted as a Candidate Answer. Then they have created Vector Space Model where the Candidate Answer is considered as one Vector and User Question word are considered another Vector using these two vectors they calculate the similarity score to see which candidate answer is more related to the query and that Candidate Answer is finally given to the user. Sahu et al. (2012) [10] have proposed Question Answering system which provide answers for the questions asked in Hindi Language in which they have used collection of hindi document about some specific topic. Here first they have converted the given user’s hindi question in Query logic language (QLL) which is a subset of Prolog using developed rules. This query is send to the database which interrogate the stored information to extract an answer which they will be converting into hindi before sending the answer back to the user. Kumar et al. (2005) [11] discussed about a Hindi Question Answering system by using Information Retrieval technique. To create this QAS they first collected all the

584

T. Sultana and S. Badugu

entity present corpus in Automatic Entity Generator module. Then on receiving a question from the user the system first classify the question in Question Classification module. In Query formation module the keyword from the question is extracted by filtering the stop words using this query keyword and Entity which is extracted from the corpus in the first nodule they retrieved the related document. In addition to this module we have Query Expansion module where synonyms of the hindi words are present since there is no Hindi Wordnet available they have created they own lexical database of synonyms. After collecting related document important passages are selected in Passage Selection module using Locality-based similarity heuristic these topped passages are called as Candidate passages. From the Candidate passages the answer is selected based on the given question type. Bhoir V and M. A. Potey, in [12] has discussed about “Question answering system: A heuristic approach” This is a closed domain Question Answering System whose selected domain is “Tourism”. For that the Author has used collection of information about Pune tourism as their corpus. In this paper [12] Author has taken following step where firstly they have created a web Crawler using Java and to that they have given list of website which contains the information related to pune tourism. After that the information collected by the crawler is preprocessed to get the keyword, Then whenever the QAS get a user question, that question will also get preprocessed to remove stop-word or noise and only important keywords should be their, After extracting keyword from both question and corpus they have used the concept of Procedure programming language to retrieve the final answer. In order to exceed the time required for retrieving the answer they have also utilize the concept of master sentence where from the data, the sentence in which there is any number followed by “km”, “miles” etc will be considered as master sentence and if the question is related to the distance then the answer will be given more accurately. Moussa, and Abdel-Kader (2011) in [13] proposed “QASYO: A Question Answering System for YAGO Ontology”. In this paper the author has used YAGO ontology as their dataset. QASYO is nothing but a Sentence level- Question Answering System which combine NLP (Natural language processing), Ontologies and IR (Information retrieval) Techniques. Here in [13] to create this QASYO model the author has taken the following steps. Where, The first step is to classify the question in different types such as “what”, “when“, “who”, “where”, “which”. Next step is to create a logical representation of the query for that they have used the concept of mapping the input query into the linguistic-triple form that is nothing but subject, relation, object triple. Now, next step they have discussed is Query Generator and Query simplication, where firstly the query triple is created which is a simplified version of input query so that we can easily manipulate the given input query. After that query simplification is done using Functional/Special words deletion and Word form modification these steps are taken basically to reduce the computing time. Now the final step is called as Query Processor which receives the simplified query-triple as its input by using that input it will scan the YAGO ontology to retrieved the desired Answer.

A Review on Different Question Answering System Approaches

585

3 Comparison of Question Answering System Methods Following table shows details about the Question Answering System and the approaches they used (Table 1): Table 1. Comparison of question answering system methods S. Title no 1

2

3

4

5

6 7 8

9

Domain Corpus

Question Answering System for Open Frequently Frequently Asked Questions Domain Asked Question (FAQ) Music knowledge question Closed FAQs and answering system on the Domain Music ontology knowledge base ontology Malayalam question answering Closed Kerala system Domain sports Related Document DBPEDIA Open DBPEDIA BASED FACTOIDQUESTION Domain ANSWERING SYSTEM Open domain question Open Wikipedia answering using WikipediaDomain Articles based knowledge model

A Web-based Question Answering System A Hindi Question Answering System A Hindi Question Answering System for E-learning documents

Open Domain Closed Domain Closed Domain

Question answering system: A heuristic approach

Closed Domain

10 QASYO: A Question Answering Open System for YAGO Ontology Domain

Question type

Answer extraction approach

Factoid

Question Matching techniques and Ranking Algorithm

Factoid

Malayalam factual questions

Question matching and Question Analyzing techniques Named Entity snippe Recognition for t extraction

Factoid

SPRQL queries

Factoid List Factoid- TREC ‘‘exact Descriptive answer” criterion. Descriptive- ‘‘key phrases,” similar to the TREC ‘‘nugget” criterion Web Pages factual HMM-based named entity questions recognizer Hindi text factual Hindi shallow parser and Documents questions Some Rules locality-based similarity Agriculture Factoid, heuristic, syntactic and and Science List, and partial semantic information document Causal questions Tourism Factoid Procedure Programming domain And List Style URL Questions YAGO Factoid, Forming of Query-Triple Ontology List and Yes/No question

4 Conclusion We can conclude by observing the previous work done on the Question Answering System that most of the people have employed Information Retrieval Approach for Question Answering System. However for our project we would like to use both the

586

T. Sultana and S. Badugu

concept of Information Retrieval Techniques and Machine learning for creation of our QAS. and we can also see that almost equal amount of work is done on both Closed domain QAS and Open domain QAS as when we have compare these ‘11’ papers in which ‘5’ papers belongs to Closed and the rest of the paper belong to Open domain QAS, However in term of accuracy of the system. Closed domain QASs are more accurate as the system has to work on single domain which reduces the amount of processing or working required to retrieve a precise answers to the users questions.

References 1. Cimiano P, Unger C, McCrae J (2014) Ontology-based interpretation of natural language, Morgan & Claypool Publishers. Accessed 1 Mar 2014. ISBN 978-1-60845-990-2 2. Saini A, Yadav PK (2017) A survey on question–answering system. Int J Eng Comput Sci 6 (3):20453–20457. https://doi.org/10.18535/ijecs/v6i3.09. ISSN:2319-7242 3. Tirpude S, Alvi AS (2015) Department of computer science & engineering closed domain question answering system: a survey. IJIFR/ V2/ E9/ 065 4. Bhardwaj D, Pakray P, Bentham J, Saha S (2016) Question answering system for frequently asked questions, Department of CSE NIT, Mizoram, India 5. Fu J (2009) Domain ontology based automatic question answering 6. Tahri A, Tibermacine O (2013) DBPedia based factoid question answering system. Int J Web Semant Technol (IJWesT) 4(3):23 7. Pragisha K, Reghuraj PC (2014) A natural language question answering system in malayalam using domain dependent document collection as repository. Int J Comput Linguist Nat Lang Process 3(3):0756–2279 8. Ryu P-M, Jang M-G, Kim H-K (2014) Open domain question answering using wikipediabased knowledge model. Inf Process Manag 50:683–692 9. Zhang D, Lee WS (2003) A web-based question answering system 10. Sahu S, Vashnik N, Roy D (2012) Prashnottar: a Hindi question answering system. Int J Comput Sci Inf Technol (IJCSIT) 4(2):149–158 11. Kumar P, Kashyap S, Mittal A, Gupta S (2005) A Hindi question answering system for Elearning documents. In: Proceedings of IEEE international conference on intelligent sensing and information processing, Bangalore, India, pp 80–85 12. Bhoir VM, Potey A (2014) Question answering system: a heuristic approach. In: 2014, IEEE fifth international conference on applications of digital information and web technologies 13. Moussa AM, Abdel-Kader R (2011) QASYO: a question answering system for YAGO ontology. Int J Database Theory Appl 4(2):99 14. Li Y, Bontcheva K, Cunningham H (2004) SVM based learning system for information extraction, Department of Computer Science, the University of Sheffield, Sheffield, S1 4DP, UK 15. Cristianini C, Shawe-Taylor J (2000) An introduction to support vector machines. Cambridge University Press, Cambridge 16. Bikel D, Schwartz R, Weischedel R (1999) An algorithm that learns what’s in a name. Mach Learn 34(1–3):211–231 17. Bizer C, Lehmann J, Kobilarov G, Auer S, Becker C, Cyganiak R, Hellmann S, (2009) DBPedia - a crystallization point for the web of data (PDF). Web Semant: Sci Serv Agents World Wide Web 7(3):154–165 https://doi.org/10.1016/j.websem.2009.07.002. ISSN 15708268, CiteSeerX 10.1.1.150.4898 18. https://en.wikipedia.org/wiki/Question_answering 19. https://en.wikipedia.org/wiki/Web_search_engine

A Study of Malicious URL Detection Using Machine Learning and Heuristic Approaches Aliya Begum(&) and Srinivasu Badugu Department of Computer Science and Engineering, Stanley College of Engineering and Technology for Women, Abids, Hyderabad 500 001, India [email protected], [email protected]

Abstract. Malicious URL is a typical and genuine threat to cybersecurity. A Malicious URL has an assortment of spontaneous content in the form of phishing, spam in order to launch attacks. Innocent users visit such web sites move toward becoming casualties of various sorts of scams, including monetary loss, theft of private information (identity, credit-cards, etc.). It is essential to identify and follow up on such dangers in an opportune way. In this paper we had studies different techniques for detecting malicious URL and discussing each and every technique their merits and demerits. Keywords: Malicious URL
Machine Learning Internet security
Cybersecurity


Support vector machines


1 Introduction Machine learning (ML) is a branch of Artificial Intelligence that pushes forward the idea that, by giving access to the right data, machines can learn by themselves how to solve a specific problem [1]. The methods of new communication technologies have extremely large impact in the growth and promotion of businesses spanning across many applications like online banking, social networking sites, and electronics commerce. In fact, in today’s age it is almost compulsory to have an online exposure to run a successful venture. As a result, the importance of the World Wide Web has continuously been increasing. Unfortunately, the technological improvement come integrated with new advance techniques to attack and scam users. Such attacks include crook websites that sell fake goods, financial fraud by tricking users into revealing sensitive information which eventually lead to theft of money or identity, or even installing malware in the user’s system. There are a wide assortment of strategies to actualize drive-by exploits, watering hole, phishing, denial of service, distributed denial of service, man-in-the-middle, social engineering, explicit hacking, endeavors attacks, and numerous others [29]. Considering the diverse sorts of attacks, conceivably new attack types, and the countless settings in which such attacks can show up, it is difficult to plan vigorous frameworks to detect cyber-security breaches. The confinements of customary security management technologies are becoming more and more serious given this exponential development © Springer Nature Switzerland AG 2020 S. C. Satapathy et al. (Eds.): ICETE 2019, LAIS 4, pp. 587–597, 2020. https://doi.org/10.1007/978-3-030-24318-0_68

588

A. Begum and S. Badugu

of new security dangers, quick changes of new IT technologies, and significant lack of security experts. A significant number of these attacking strategies are acknowledged through spreading compromised URLs [3]. There has been a great deal of research to prevent users from visiting malicious websites so as to lessen Internet violations [28]. Well known sorts of attacks utilizing malicious URLs include: Spam, Phishing and Social Engineering [2]. URL is the abbreviation of Uniform Resource Locator, which is the www address of reports and different assets on the World Wide Web. The URL contains protocol, host name, directory and path is shown in Fig. 1.

Fig. 1. Example of a URL - “Uniform Resource Locator” Source: adapted from [25]

The protocol refers to a communication protocol for exchanging information between information devices; e.g., HTTP, FTP, HTTPS, and so on. Protocols are of different kinds and are utilized as per the ideal specialized strategy. The fully qualified domain name distinguishes the server who hosts the web page. It contains a registered domain name i.e. second-level domain and suffix i.e. top-level domain (TLD). The domain name portion has to be registered with a domain name registrar. A host name comprises of a sub domain name and domain name [26].

2 Literature Survey Over the decade, many strategies have been proposed for malicious URLs detection. In this section, we will review few state of the art methods briefly. Following are some of the techniques that researchers have utilized for the malicious URLs detection, which are described below. • • • • 2.1

Machine Learning-based Approaches Non-machine Learning-based Approaches Neural Network-based Approaches Behaviour-based Detection Approaches Machine Learning-Based Approaches

Babagoli et al. [4] have introduced a strategy for phishing website detection which uses a meta-heuristic based non-linear regression algorithm along with a feature selection approach. So as to approve the proposed strategy, they have utilized a dataset which

A Study of Malicious URL Detection Using Machine Learning

589

involves 11055 legitimate and phishing WebPages and then, They have chosen 20 features to be extracted from the listed websites. They have utilized two feature selection strategies that are, First is Decision tree and other is Wrapper to choose the best feature subset and they accomplish the high detection accuracy rate as 96.32% by utilizing wrapper strategy. After the feature strategy is finished. At that point, In request to predict and identify the fake websites they implemented two meta-heuristic algorithms such as harmony search (HS) which was deployed based on SVM and nonlinear regression techniques. As indicated by them, for classifying the websites, Nonlinear regression approach was used, where they got the parameters of the proposed regression model using HS algorithm. The results of the experiment demonstrate that, the non-linear regression based on HS has given high exactness rates of 94.13% for training and 92.80% for testing processes. Finally, after looking at the execution of both the procedures, the non-linear regression based HS result is better compared with SVM. Zuhair et al. [5] have portrayed about a set of 58 new Webpage hybrid features which they have refined to couple of least redundant, most extreme relevant and robust features however much as could be expected. These features were taken from two unique sources: web pages content and URL. In this manner, they have consider two feature categories through the experiment analysis. The first feature category is a group of 48 features which extracted from the source code and HTML tags of collected webpages. Whereas, The second feature category is a group of 10 URL features from extracted the web pages URL. They utilized a specific criterion i.e. mRMR to find a optimal feature subset for effective phishing detection. Since, as per them, mRMR removes excess and insignificant features simultaneously over a high dimensional feature space. They have used SVM machine learning classifier and assessment criteria such as TP, FP, FN, Precision, Recall and F-measure to assess their methodology. The result of experimental analysis demonstrates that, their methodology can be utilized to enhance a phish detection model for any anti-phishing scheme later on. Choi et al. [6] have proposed a technique to recognize malicious URLs utilizing machine learning of every single distinctive kind of attacks, for example, spam, phishing, malware and to identify the kind of attack a malicious URL attempts to launch. They have utilized features for example, lexical, Webpage content, link popularity, DNS information and network traffic. They have employed 3 Machine Learning algorithms such as Support Vector Machine to detect malicious URLs, RAkEL and ML-k Nearest Neighbour learning algorithms for multi-label classification problem to recognize attack type. They tested their technique on forty thousand benign URLs and thirty-two thousand malicious URLs acquired from real-life Internet sources, for example, Malware URLs from DNS-BH, Phishing URLs from PhishTank, Spam URLs from jwSpamSpy, Web spam dataset, and benign URLs from DMOZ Open Directory Project, Yahoo!s directory. Finally, They demonstrates that They accomplished the exactness of 98% in detection of malicious URLs and 93% in identification of attack types. They likewise provide details regarding the adequacy of each gathering of discriminative features, and examine their evadability. Canali et al. [7] have proposed a filter, called Prophiler that utilizes static analysis techniques to rapidly look at a Web page for malicious content. Prophiler is fed by a modified instance of Heritrix which crawls a list of seed URLs fetched daily from three

590

A. Begum and S. Badugu

search engines namely, Google, Yahoo, and Bing. They have utilized features gotten from the HTML contents of a page, from the associated JavaScript code and from the corresponding URL. They utilized diverse machine learning algorithms like Random Tree, Random Forest, Naive Bayes, Logistic, J48 Bayes Net and Logistic regression for evaluation. As indicated by them, their filtering approach is able to reduce the load on a more exorbitant dynamic analysis tools i.e. Wepawet by over 85%, with a negligible amount of missed malicious. Ma et al. [8] have proposed a methodology depend on automated URL classification, utilizing statistical methods to discover the lexical and host-based properties of malicious Web site URLs. They have extracted the Lexical features and Host-based features. The host-based features contain IP address properties, WHOIS properties, domain name properties and geographic properties [27]. They have utilized machine learning algorithms like Naive Bayes, SVM and Logistic Regression for evaluation. As per them, the resulting classifiers obtain 95–99% accuracy, identifying expansive quantities of malicious Web sites from URLs, with just humble false positives. Thomas et al. [9] have presented Monarch, a real-time system that slithers URLs as they are submitted to Web services and decides if the URLs direct to spam. They demonstrate that Monarch can give precise, ongoing assurance. They have used Monarchs feature collection infrastructure through the span of two months to creep 1.25 million spam email URLs, approximately 567,000 blacklisted Twitter URLs and more than 9 million non-spam Twitter URLs. They have assessed the accuracy of Logistic Regression with L1-regularization classifier and its run- time execution. Their trial results demonstrate that they can recognize Web service spam with 90.78% precision and 0.87% false positives, with a median feature collection and classification time of 5.54 s. Eshete et al. [10] have proposed a insignificant concept, called BINSPECT that joins static analysis and emulation. They have utilized Supervised Learning method of Machine Learning (ML) in detection of malicious Web pages that may launch phishing, drive-by- download, injection and malware distribution attacks. They have collected features like page-source features, social-reputation features, and URL features. They collected a malicious dataset of 71,919 URLs from the malware and phishing blacklist of Google, Phishtank database and the malware and injection attack URL list of Malware URL. The benign dataset of 414,000 benign URLs is gathered from 3 prevalent sources like the Alexa Top sites, the Yahoo random URL generation service and the DMOZ directory. As indicated by their exploratory assessment, BINSPECT accomplished 97% precision with low false flags. Basnet et al. [11] have proposed a machine learning based method to deal with distinguish detect phishing Web pages. They used numerous unique content based features and applied cutting-edge machine learning techniques such as batch learning algorithms, Random Forests, SVM with rbf linear kernels, C4.5, LR and a set of online learning algorithms: Perceptron, PA, NB-U, CW algorithms, and LB-U. They have utilized 179 Web page features for example, keyword based features, reputation based features, lexical based features, and search engine based feature to show their methodology. To perform all the experiments, they utilized WEKA and CW libraries. The experimental results demonstrate that their proposed methodology can detect

A Study of Malicious URL Detection Using Machine Learning

591

phishing WebPages with an exactness of 99.9%, false positive rate of as low as 0.00% and false negative rate of 0.06%. HajianNezhad et al. [12] examined a novel set of features including HTML, JavaScript (jQuery library) and XSS attacks. They have assessed the proposed features on a data set assembled by a crawler from malicious web domains, IP address and black lists. They utilized various machine learning algorithms with the end goal of assessment. As indicated by them, the trial results demonstrate that by utilizing the proposed set of features, the C4.5-Tree algorithm offers the best execution with 97.61% precision, and F1-measure has 96.75% exactness. Likewise, they performed ranking of the features. As indicated their ranking outcomes, they recommended that nine of the proposed features are among the twenty best discriminative features. Lee et al. [13] have proposed a heuristic-based phishing detection method that utilizes uniform resource locator (URL) features and recognized features that phishing site URLs contain. The proposed strategy utilizes those features for phishing detection. The method was performed with a dataset of three thousand phishing site URLs and legitimate site URLs. Their experiment results analysis demonstrate that the proposed procedure can detect more than 98.23% of phishing sites. Altaher [14] have proposed a hybrid approach for classifying the websites as Legitimate or Suspicious websites, Phishing, the proposed methodology keenly joins the K-nearest neighbors (KNN) with the Support Vector Machine (SVM) algorithms at two stages. Right off the bat, the KNN was used as a robust to noisy data and powerful classifier. Also, SVM is an amazing classifier. The proposed methodology coordinates the effortlessness of KNN with the adequacy of SVM. The experimental results show that the proposed hybrid approach accomplished the most elevated precision of 90.04% when compared with different methodologies. 2.2

Non-machine Learning-Based Approaches

Dewald et al. [15] have described about ‘ADSandbox’, an analysis system for malicious Websites that focuses on detecting attacks through JavaScript. They have utilized a novel concept of a client-side JavaScript sandbox. Apparently, this methodology consolidates generality with usability, since the system is executed directly on the client running the Web program before the Web page is shown. As per them, the experimental results demonstrate that, they can accomplish false positive rates near 0% and false negative rates beneath 15% with an execution overhead of just a couple of moments. Zhang et al. [16] have proposed another technique to decide the malware distribution network (MDN) from the secondary URLs and re-direct chains recorded by a high-collaboration client honeypot. They have proposed a novel drive-by download detection strategy. As per them, rather than relying upon the malicious contentused by previous techniques, their algorithm initially recognizes and after that use the URLs of the MDN’s central servers, where a focal server is a typical server shared by a substantial level of the drive-by download attacks in the same MDN. A set of regular expression based signatures are formed based on URLs of every focal server. This technique enables extra malicious webpages to be identified which launched but failed to execute a successful drive-by download attack. The new drive by detection system

592

A. Begum and S. Badugu

named arrow has been executed, and then they provide a large-scale assessment on the output of a production drive-by detection system. The experimental results shows the effectiveness of their technique, where the detection coverage has been helped by 96% with a very low false positive rate. Lee et al. [17] have proposed a “WARNINGBIRD”, a suspicious URL detection system for Twitter [31]. Rather than concentrating on the landing pages of individual URLs in each tweet, They consider related divert chains of URLs in various tweets. Since attackers have limited resources and in this manner need to reuse them, a segment of their divert chains will be shared. They concentrated on these shared resources in order to detect suspicious URLs. They have gathered an extensive number of tweets from the Twitter public timeline and trained a statistical classifier with features taken from corresponded URLs and tweet context information. Their classifier has high accuracy and low false-positive and false-negative rates. They additionally exhibits WARNINGBIRD as a realtime system for classifying suspicious URLs in the Twitter stream. Sonowal et al. [18] have proposed a multilayer model to recognize phishing, PhiDMA (Phishing Detection using Multi-filter Approach). As indicated by them PhiDMA model include five layers: Lexical signature layer, Auto upgrade whitelist layer, Lexical signature layer, URL features layer, Lexical signature layer, String matching layer and Accessibility Score comparison layer. They have executed a model of PhiDMA show for people with visual weaknesses. As indicated by their experiment results, they demonstrates that the model is competent to detect phishing sites with a precision of 92.72%. 2.3

Neural Network-Based Approaches

Vinayakumar et al. [19] have assessed different deep learning architectures exceptionally RNN, I-RNN, LSTM, CNN, and CNN-LSTM architectures for the task of malicious URLs detection. They have conducted different experiments with different configurations of network parameters and network structures, to locate the optimal parameters for deep learning architecture. All the experiments run till a thousand epochs at a learning rate in the range 0.01–0.5 [30]. As per them, deep learning mechanisms outperformed the hand created feature mechanism. Specifically, LSTM and hybrid network of CNN and LSTM have accomplished most elevated exactness as 0.9996 and 0.9995 individually. Smadi et al. [20] a novel system is proposed which consolidates a neural network with reinforcement learning to recognize phishing attacks n the online mode for the first time. The proposed model can adjust to create another phishing email detection framework that reflects changes in recently investigated practices, which is accomplished by adopting the idea of reinforcement learning to upgrade the framework progressively after some time. This proposed model take care of the issue of constrained dataset by automatically add more emails to the offline dataset in the online mode. An algorithm is proposed to investigate new phishing behaviors in a new dataset. Through precise testing using the well-known data sets, They demonstrate that the proposed technique can deal with zero-day phishing attacks with superior dimensions accomplishing high precision, TPR, and TNR at 98.63%, 99.07%, and 98.19% separately. What’s more, it indicates low FPR and FNR, at 1.81% and 0.93%

A Study of Malicious URL Detection Using Machine Learning

593

individually. Comparison with other similar techniques on the equivalent dataset demonstrates that the proposed model outflanks the current techniques. 2.4

Behaviour-Based Detection Approaches

Kim et al. [21] conducted detailed study on 1,529,433 malicious URLs gathered in the course of recent years. They analyzed attackers’ strategic behavior with respect to URLs and extracted common features. And after that isolate them into three diverse feature pools to decide the dimension of trade off of obscure URLs. To use detection rates, they utilized a similarity matching technique. They trust that new URLs can be identified through attackers’ routine URL manipulation behaviors. This approach covers an expansive arrangement of malicious URLs with small feature sets. The precision of the proposed methodology (up to 70%) is sensible and the methodology requires just the attributes of URLs to be examined. This model can be used amid preprocessing to decide if input URLs are benign, and as a web filter or a risk-level scaler to appraise whether a URL is malicious. Considering the classification based malicious URLs detection techniques, Table 1 gives the comparative study of previous work done on malicious URLs detection. Table 2 gives the list of acronyms used in Table 1. Table 1. Comparative study of prior work Authors

Classification/regression technique

Feature selection techniques

Accuracy rate

Detected classes

Vinayakumar et al. [19] Kim et al. [21] Babagoli et al. [4] Smadi et al. [20] Zuhair et al. [5] Choi et al. [6]

RNN, I-RNN, LSTM, CNN, CNN-LSTM Fuzzy HS, SVM

NA

99%

Malicious/Benign URLs

NA DT, WFS

70% 92.80%

Malicious/Benign URLs Phish/Benign Websites

NN

NA

98.63%

Phish/Benign Websites

SVM

mRMR

98%

Phish/Benign Websites

SVM, RAkEL, ML-kNN

NA

98%

RT, RF, NB, J48, Bayes Net, LR String matching, Scoring Algorithm NB, SVM, LR LR

NA

90%

Malicious/Phishing/Spam/Benign URLs Malicious/Benign Websites

NA

92.72%

Phish/Benign Websites

NA NA

95–99% 90.78%

Malicious/Benign URLs Spam/non-spam URLs

NA

97%

Malicious/Benign Websites

NA

99.9%

Phish/Benign Websites

Canali et al. [7] Sonowal et al. [18] Ma et al. [8] Thomas et al. [9] Eshete et al. [10] Basnet et al. [11] Lee et al. [17] HajianNezhad et al. [12]

J48, RT, RF, NB, Bayes Net, SVM, LR, CW Majority Vote RF, SVM, NB, C4.5, LR, NB-U, LB-U, Perceptron, PA, CW Statistical classifier Multi-layer perceptron, NB, SVM, KNN, ADTree, BFTree, C4.5

NA 86.3% Entropy Based, Gain 97.61% Ratio, correlation coefficient square, TOPSIS

Malicious/Benign URLs Malicious/Benign Websites

594

A. Begum and S. Badugu Table 2. List of acronyms used in Table 1 RNN: I-RNN: LSTM: CNN: CNN-LSTM: HS: SVM: DT: WFS: NN: mRMR: RAkEL: RT: ML-kNN: RF: NB: J48: LR: CW: NB-U: LB-U: PA: KNN: AD-Tree: BFTree: TOPSIS: WFLE: AP: AROW: NA:

Recurrent Neural Network Identity-Recurrent Neural Network Long Short-term Memory Convolution Neural Network Convolutional Neural Network-long Short-term Memory Harmony Search Support Vector Machine Decision Tree Wrapper-based Feature Selection Neural Network Maximum Relevant Minimum Redundant RAndom k-labELsets Random Tree Multi-label K-nearest neighbor Random Forest Naive Bayes J48 Decision Tree Logistic Regression Confidence Weighted Naive Bayes Updatable LogitBoost Updatable Passive Aggressive K-nearest neighbor Alternating Decision Tree Best First Tree Technique for Order Preference by Similarity to Ideal Solution Weighted Feature Line Embedding Averaged Perceptron Adaptive Regularization of Weight Vectors Not Applicable

A comparative study of previous work for malicious URLs detection based on key features like URL features, URL source features, domain name features and short URLs features and their merits and demerits are given in Table 3.

A Study of Malicious URL Detection Using Machine Learning

595

Table 3. Comparison of Different Feature Types Feature Type

Merits

• Easy feature extraction • Fast processing • Less system overhead • Better for static analysis of URLs • Widely used in the literature URL source • Deep analysis of webpages features possible [4, 5, 7, 10–12, 18, • Suitable for malware and embedded code detection 20, 22, 24] • Attacker behavior detection possible • Able to detect attacks on the fly • Widely considered in the prior work Domain Name • Easy feature extraction Features [6–8, 10, • Useful in different attack 11] detection • Widely considered in the prior work for attack types identification Short URLs • High comprehensiveness Features [22] URL features [6, 8, 9, 17, 19, 21–23]

Demerits • Not able to detect todays ever- changing attacks • Not able to identify other attacks only suitable for phishing attacks • Not able to detect run-time behavior of attacker • Require more time for webpage rendering and processing • Code obfuscation and injection becomes challenging • System design becomes cumbersome

• Prone to typosquatting or URL hijacking • Domain name resolution may becomes cumbersome • Domain registration period play an vital role because malicious domains have very short lifespan • Time consuming extraction process • Most of the URLs were offline

3 Conclusion and Future Work Malicious URL detection plays a critical role for many cybersecurity applications. Automated detection of malicious URLs remains a very challenging open issue as URLs are dynamic in nature. Since new URLs can be generated everyday very easily. Attackers can create a technique to evade a blacklist and fool users by modifying the URL to appear legitimate. Once the URLs appear legitimate and user’s visit them, an attack can be launched. Nowadays machine learning and deep learning approaches are used for malicious URL detection and clearly deep learning approaches are a promising direction. Future directions incorporate more effective feature extraction and representation learning (e.g., via deep learning approaches).

References 1. Marsland S (2015) Machine learning: an algorithmic perspective. CRC Press, Boca Raton 2. Patil DR, Patil J (2015) Survey on malicious web pages detection techniques. Int J u-and eServ Sci Technol 8(5):195–206 3. Hong J (2012) The state of phishing attacks. Commun ACM 55(1):74–81

596

A. Begum and S. Badugu

4. Babagoli M, Aghababa MP Solouk V (2018) Heuristic nonlinear regression strategy for detecting phishing websites. Soft Computing, pp 1–13. https://doi.org/10.1007/s005000183084-2 5. Zuhair H, Selamat A, Salleh M (2015) Selection of robust feature subsets for phish webpage prediction using maximum relevance and minimum redundancy criterion. J Theor Appl Inf Technol 81(2):188–205 6. Choi H, Zhu BB, Lee H (2011) Detecting malicious web links and identifying their attack types. In: 2nd USENIX conference on web application development (WebApps 2011), pp 1– 12 7. Canali D, Cova M, Vigna G, Kruegel C (2011) Prophiler: a fast filter for the large-scale detection of malicious web pages. In: 20th international conference on world wide web (WWW11), pp 197–206 8. Ma J, Saul LK, Savage S, Voelker GM (2011) Learning to detect malicious urls. ACM Trans Intell Syst Technol 3(2):1–24 https://doi.org/10.1145/1961189. 1961202 9. Thomas K, Grier C, Ma J, Paxson V, Song D (2011) Design and evaluation of a realtime URL spam filtering service. In: IEEE symposium on security and privacy (SP), pp 447–462 10. Eshete B, Villafiorita A, Weldemariam K (2012) BINSPECT: holistic analysis and detection of malicious web pages. In: SecureComm, pp 149–166 11. Basnet RB, Mukkamala S, Sung AH (2008) Detection of phishing attacks: a machine learning approach. In: Soft computing applications in industry, pp 373–383 12. Nezhad JH, Jahan MV, Tayarani-NM, Sadrnezhad Z (2017) Analyzing new features of infected web content in detection of malicious web pages. ISC Int J Inf Secur 9(2):63–83 13. Lee JL, Kim DH, Chang-hoon, L (2015) Heuristic-based approach for phishing site detection using url features 14. Altaher A (2017) Phishing websites classification using hybrid SVM and KNN approach. Int J Adv Comput Sci Appl 8(6):90–95 15. Dewald A, Holz T, Freiling FC (2010) ADSandbox: sandboxing javascript to fight malicious websites. In: ACM symposium on applied computing, pp 1859–1864 16. Zhang J, Seifert C, Stokes JW Lee W (2011) Arrow: generating signatures to detect drive-by downloads. In: 20th international conference on world wide web, pp 187–196 17. Lee S, Kim J (2013) WarningBird: detecting suspicious URLs in Twitter stream. In: Network and distributed system security symposium (NDSS12), pp 1–13 18. Sonowal G, Kuppusamy KS (2017) PhiDMA - a phishing detection model with multi-filter approach. J King Saud Univ Comput Inf Sci, 1–14. https://doi.org/10.1016/j.jksuci.2017.07. 005 19. Vinayakumar R, Soman KP, Poornachandran P (2018) Evaluating deep learning approaches to characterize and classify malicious URLs. J Intell Fuzzy Syst 34(3):1333– 1343. https:// doi.org/10.3233/jifs-169429 20. Smadi S, Aslam N, Zhang L (2018) Detection of online phishing email using dynamic evolving neural network based on reinforcement learning. Decis Support Syst 107:88–102. https://doi.org/10.1016/j.dss.2018.01.001 21. Kim S, Kim J, Kang BB (2018) Malicious URL protection based on attackers’ habitual behavioral analysis. Comput Secur 77:790–806 22. Nepali RK, Wang Y (2016) You look suspicious: leveraging visible attributes to classify malicious short urls on Twitter. In: 49th Hawaii international conference on system sciences (HICSS), pp 2648–2655 23. Patil DR, Patil JB (2016) Malicious web pages detection using static analysis of URLs. Int J Inf Secur Cybercrime 5(2):57–70. https://doi.org/10.19107/IJISC.2016.02.06 24. Patil DR, Patil JB (2017) Detection of malicious JavaScript code in web pages. Indian J Sci Technol 10(19):1–12. https://doi.org/10.17485/ijst/2017/v10i19/114828

A Study of Malicious URL Detection Using Machine Learning

597

25. https://towardsdatascience.com/phishing-domain-detection-with-ml-5be9c99293e5 26. Buber E, Demir O, Sahingoz OK (2017) Feature selections for the machine learning based detection of phishing websites. In: 2017 international artificial intelligence and data processing symposium (IDAP) 27. Ali G, Li KF (2014) Consumer transactions on the web. In: 2014 28th international conference on advanced information networking and applications workshops 28. Jha PK, Shanker P, Sujadevi VG, Prabhaharam P (2019) Deepmal4J: Java malware detection employing deep learning. In: Springer 6th International Symposium, SSCC 2019 29. Kumar PR, Raj PH, Jelciana, P (2019) A framework to detect compromised websites using link structure anomalies, chap 7. Springer, America Inc. 30. Swapna G, Soman KP, Vinayakumar R (2018) Automated detection of cardiac arrhythmia using deep learning techniques. Proc Comput Sci 132:1192–1201 31. (2019) International conference on computer networks and communication technologies, Springer, America Inc.

A Review on Network Intrusion Detection System Using Machine Learning T. Rupa Devi(&) and Srinivasu Badugu Department of Computer Science and Engineering, Stanley College of Engineering and Technology for Women, Abids, Hyderabad 500 001, India [email protected], [email protected]

Abstract. After digital revolution, large amount of data are produced from diverse networks from time to time. Hence security of this data is more important. So, there is a need to automate this security system. Intrusion detection systems are considered as the best solution to detect intrusions. Network intrusion detection systems (NIDS) are hired as a defense system to protect networks. Numerous techniques for the development of these defense systems are found in the literature. However, study on the enhancement of datasets used to train and test such security systems is also important. Improved datasets progress the detection capabilities for both offline and online intrusion detection models. Standard datasets like KDD 99, NSL-KDD cup 99 and DARPA 1999 are outdated and they don’t contain data of present attacks such as Denial of Service, therefore they are not suitable for evaluation. In this paper, in depth analysis of CIDDS-001 dataset is shown and the sightings are presented. In this paper, a gist of different papers available related to NIDS are given. This paper even compares all the research papers specifying their merits and demerits. This paper is concluded by providing a research method that can be applied to develop a better network intrusion detection system. Keywords: Anomaly CIDDS – 001 dataset


K-Nearest neighbor
KDD
Metrics
Signature


1 Introduction Network security has become one of the most concerning problems for internet users and service providers with drastic increase in the internet usage [1, 2]. A secure network is defined in terms of the protection of its software and hardware in contrast to different types of intrusions. A network is secured by applying a robust observation, analysis and defense mechanisms. As the world has become more connected over the Internet, computer networks are more prone to malicious attacks [1, 3]. “Intrusion is an attempt to compromise CIA (Confidentiality, Integrity, Availability), or to bypass the security mechanisms of a computer or network” [4, 5]. “Intrusion detection is the process of monitoring the events occurring in a computer system or network, and analysing them for signs of intrusion” [4, 5]. NIDS is one of the main tools used to report network attacks. © Springer Nature Switzerland AG 2020 S. C. Satapathy et al. (Eds.): ICETE 2019, LAIS 4, pp. 598–607, 2020. https://doi.org/10.1007/978-3-030-24318-0_69

A Review on Network Intrusion Detection System Using Machine Learning

599

A network intrusion detection system (NIDS) monitors the network traffic by identifying suspicious activity, which may represent an attack or illegal access. NIDS are tools which implement such mechanisms so as to protect a network from intrusions which may be from within the network or from outside the network. These systems observes the incoming and also outgoing traffic of a network, perform analysis periodically and report when an intrusion is detected. The major components of this system include traffic collector, analysis engine, signature database and alarm storage, as shown in Fig. 1 [6].

Fig. 1. Intrusion detection system components. Source: adapted from [6]

The role of every component is important in intrusion detection. Network traffic is sniffed by traffic collector which are in the form of packet traces, then analysis engine performs a profound analysis of the sniffed traffic data and directs the alarm signals to alarm storage once intrusion is identified. The patterns or signatures of known intruders are stored in signature database, and then matching is done using these signatures. A typical NIDS is shown in Fig. 2 [6].

Fig. 2. Design of network intrusion detection system. Source: adapted from [6]

600

T. Rupa Devi and S. Badugu

NIDS is classified into Misuse detection (MD) [1, 7], Anomaly Detection (AD) [1, 8]. MD based NIDS use patterns or signatures of already surviving attacks to identify intrusions. Whereas AD based NIDS checks firm deviations from normal profiles of the network traffic and report it as an attack. MD based NIDS’ Detection Rate (DR) is high while False Positive Rate (FPR) being low when compared to AD. But AD based NIDS detects novel attacks in networks and so, this property overtakes them from MD based NIDS. MD works better on offline data while AD works better on online data. Machine Learning (ML) [1, 9] plays a major role in building a better NIDS. It makes a system to learn from the already existing traffic patterns or signatures and act accordingly for the upcoming traffic patterns. Training and testing are the two important jobs in the ML. ML needs huge and complex datasets consisting of distinct types of normal and abnormal traffic patterns. There is also a necessity to use ML algorithms to NIDS which is low in computational time and space complexity for an improved learning. In this work we have analyzed CIDDS-001 dataset using some prominent NIDS evaluation metrics like DR, FPR, Accuracy, Precision and F-measure [1, 10]. We have used ML models like KNN classification algorithm [1, 11] due to its better DR and Naive Bayes classification as it can often outperform most of the sophisticated classification methods despite its simplicity. CIDDS-001 Dataset CIDDS-001 (“CIDDS-001”, 2017) [12] is a labelled flow-based dataset (Ring, Wunderlich et al. 2017). It was developed mainly to evaluate AD based NIDS. The dataset contains traffic from both OpenStack and External Servers. CIDDS-001 dataset comprises of 13 features and a class attribute. Out of them 11 features were used for this study. The features like Attack ID and Attack Description are ignored because they just give elaborated information of the executed attacks. So, these attributes did not contribute to the analysis significantly. Almost 153,026 instances of external servers and 172,839 instances of OpenStack server were gathered for analysis. Every instance was labelled as classes namely, normal, victim, attacker, suspicious and unknown.

2 Analysis on Network Intrusion Detection System Using Machine Learning Techniques In this section, detailed overview of the different research techniques are given with their working procedure. Verma et al. [1] has found that to perform training and testing, the datasets used are of equal importance as better datasets can progress offline Intrusion Detection. This paper tells that datasets like KDD99 and NSL-KDD cup 99 are outdated and are not suitable for assessing Anomaly based NIDS. This paper also presents a new dataset CIDDS-001. The methods used by them are distance based machine learning algorithms like k-nearest neighbour classification (KNN) and k-means clustering algorithms using Weka tool. They analysis is done using prominent evaluation metrics which are used for evaluating NIDS including Detection Rate, Accuracy and False Positive Rate.

A Review on Network Intrusion Detection System Using Machine Learning

601

Hsiao et al. [13] presents a study of unsupervised machine learning based approaches like stationary and non-stationary models for NIDS. The methods used are Gaussian Mixture Model as stationary model and PHAD (Packet Header Anomaly Detection) as non-stationary model. They used 1999 DARPA dataset. They illustrate that a non-stationary model (PHAD) achieves more than 35  higher quality than the simple stationary model (GMM) for the NIDS which is a sniffer in a network. The results of PHAD for detecting potential attacks in network traffic are reproduced and detects 62 of 201 attacks with a total of 86 false positives (an average of under 9 per day) on the 1999 DARPA dataset. Alsallal [14] believes that a sophisticated attacker can easily evade the procedures where known threats are detected based on defined rules (i.e. signature based techniques) or behavioral analysis by base lining the network. Hence there is a need for more intelligent intrusion detection. So, researchers are trying to apply machine learning techniques to this area of cybersecurity. As data is very important for implementing ML techniques he created and experimented on a small, data set with 9 distinct features selected out of 33 attributes which helps analysts to train computers to detect attacks like zero-day attacks. Different classification algorithms like Multilayer perceptron, SMO, SVM, FT, Naïve Bayes, J48, Multinomial logistic regression and Bayesian network were implemented on this dataset using Weka tool and its accuracy was calculated to be 100%, 100%, 87.5%, 100%, 82.5%, 90%, 97.5%, 95% respectively. KarsligEl et al. [15] implemented a new semi-supervised anomaly detection system. They used k-means clustering algorithm and separated normal samples into clusters in the training phase. Then, a threshold value was calculated to differentiate normal and abnormal samples. The samples that are distant from the clusters’ centers more than the threshold value are termed as anomalies. They used NSL-KDD dataset to test the effectiveness of their system. They attained an accuracy of 80.119%. Shaya et al. [4] evaluated the usage of machine learning techniques in intrusion detection systems. He presented the three main intrusion detection methodologies such as Anomaly-based Detection (AD), Signature-based Detection (SD) and Stateful Protocol Analysis (SPA) in computer networks according to Liao et al. He inspected how machine learning is used under the Anomaly-based Detection methodology. He also presented a machine learning based intrusion detection system from Pasocal et al. and went through the system’s methods and techniques in addition to presenting the performance results of the system as described in [4, 16] and he discussed these results. Finally, he introduced some of the challenges that face using machine learning techniques in intrusion detection systems as characterized by Sommer et al. in [4, 17]. Buczak et al. [18] surveyed on machine learning (ML) and data mining (DM) methods for cyber analytics to support intrusion detection. They provided short descriptions of different ML/DM methods like ANN, Sequence Mining, KNN, Association rules, Decision trees, Bayesian networks, GA, HMM, Naïve Bayes, Random forest and clustering algorithms like K-Means, Hierarchical, DBSCAN. Different types of datasets like Packet level, Net flow and Public data for ML/DM methods used in cyber are described. The complexity and challenges of ML/DM algorithms are addressed and recommended when to use which method.

602

T. Rupa Devi and S. Badugu

Biswas et al. [19] presents a combination of classifiers and feature selection techniques. A group of more relevant features are selected from the original set of features using feature selection techniques and then they are used to train different types of classifiers to make the IDS. They used Minimum redundancy maximum- relevance, CFS, PCA and IGR feature selection techniques. Classifiers used by them are SVM, KNN, NN, DT and NB. To find results NSL-KDD dataset is used on which five folds cross validation is done. In this paper it is observed that KNN classifier performs better than other classifiers and also concludes that information gain ratio based feature selection method is better than other feature selection methods. Aggarwal et al. (2015; “NSL-KDD cup 99 dataset”, 1999) [1, 20] conducted analytical study on NSL-KDD cup 99 dataset. The dataset attributes were classified into four classes like basic, content, traffic and host. Random Tree algorithm was used in Weka tool. The results were analyzed for each class of attributes to improve the Detection Rate (DR) and minimize False Alarm Rate (FAR). Siddiqui et al. [1, 21] used NSL-KDD dataset designed for Intrusion Detection (ID). K-means clustering was used to construct 1000 clusters from 494,020 records and established a relationship between types of attacks and protocols which are used in intrusion. Ingre et al. [1, 22] used Artificial Neural Network (ANN) to analyze NSL-KDD dataset. DR of intrusion detection was 81.2% and attack type classification were 72.9%. Moustafa et al. [1, 23], used the irrelevant features from KDD99 (“KDD 99 dataset”, 1999) [1, 24] and UNSW-NB15 (“UNSW-NB15 dataset”, 2017) [1, 25] and concluded that they reduce the efficiency of NIDS. Feature selection from two datasets was done by using association rule mining algorithm. False alarm rate (FAR) and Accuracy were calculated using classifiers. The results show that UNSW-NB15 features are more efficient than KDD 99 dataset. Saboori et al. [26] proposed a feature selection method which identifies features which are of low-quality in the dataset. The random variable variance is used to measure quality of a feature. They compared existing similarity-based algorithms like maximal information compression index, correlation coefficient and least square regression error. The outputs of these algorithms identified some features which are given to naive bayes and k-nearest neighbour classifiers to test the suggested method. This suggested technique performed better than similarity-based algorithms which are already existing based on computational cost. Rampure et al. [27] recommended feature selection based on rough set theory performed on KDD Cup99 dataset. This is built on the idea that there is an increase in data pattern visibility if there is a decrease in degree of precision in the data. Based on this idea, facts from data which was imperfect were exposed. Hasan et al. [28] suggested feature selection by Random Forests. The researchers removed repeated records from KDD99Train + KDD99Test + sets of NSL-KDD dataset and derived a dataset, RRE-KDD which is then used for evaluation of Random Forest (RF). RF chooses the most significant features desired for classification and increases accuracy by reducing time complexity. Janarthanan et al. [29] used Weka tool to examine UNSW-NB15 dataset. Techniques to select attributes like InfoGainAttibuteEval (attribute evaluator) with Ranker

A Review on Network Intrusion Detection System Using Machine Learning

603

method and CfsSubsetEval (attribute evaluator) with Greedy Stepwise method were used for choosing important features. The subset of attributes which are best was used for classification using a few ML algorithms including RF. So, it was found that kappa statistics enhanced due to classification using selected features. Aminanto, Choi et al. [30] used AWID (“AWID dataset”, 2018) [31] dataset for detection of Wi-Fi impersonation which is weighted feature selection method. They used deep feature selection and extraction for feature reduction in the dataset. The FAR of 0.012% and the accuracy of 99.918% was achieved using this approach.

3 Comparison of Network Intrusion Detection System Methods Using Machine Learning Techniques This section provides the overall comparison view of the different research techniques in terms of varying merits and demerits has been given. The following Table 1 provides the merits and demerits of the different research techniques which are discussed above.

Table 1. Comparison of different research methodologies S. no. Authors 1.

2.

3.

Merits

Verma et al. Used latest dataset CIDDS-001 for training and testing. Evaluation metrics used are Detection Rate, Accuracy and False Positive Rate Hsiao et al. Illustrates that a non-stationary model (PHAD) achieves more than 35  higher quality than the simple stationary model (GMM) for the NIDS which is a sniffer in a network. The results of PHAD for detecting potential attacks in network traffic are reproduced and detects 62 of 201 attacks with a total of 86 false positives (an average of under 9 per day) on the 1999 DARPA dataset Alsallal Helps analysts train computers to detect attacks like zero-day attacks. Different classification algorithms like Multilayer perceptron, SMO, SVM, FT, Naïve Bayes, J48, Multinomial logistic regression and Bayesian network were implemented on this dataset using Weka tool and its accuracy was calculated to be 100%, 100%, 87.5%, 100%, 82.5%, 90%, 97.5%, 95% respectively

Demerits Used only distance based machine learning algorithms like KNN and KMeans hence computation time taken is more Used 1999 DARPA dataset which is very old

Compared only the accuracy of different algorithms which were run using Weka tool. Limited dataset is used

(continued)

604

T. Rupa Devi and S. Badugu Table 1. (continued)

S. no. Authors 4.

5.

6.

7.

8.

9.

10.

KarsligЕl et al.

Merits

They implemented a new semisupervised anomaly detection system using k-means clustering algorithm and separated normal samples into clusters in the training phase. They attained an accuracy of 80.119% Shaya et al. They aimed at evaluating the use of ML techniques in intrusion detection systems. They presented the performance results of different systems and discussed their results Buczak The complexity and challenges of et al. ML/DM algorithms like ANN, Sequence Mining, KNN, Association rules, Decision trees, Bayesian networks, GA, HMM, Naive Bayes, Random forest and clustering algorithms like K-Means, Hierarchical, DBSCAN are addressed and recommended when to use which method Biswas et al. They used Minimum redundancy maximum- relevance, CFS, PCA and IGR feature selection techniques. Classifiers used by them are SVM, KNN, NN, DT and NB. They observed that KNN classifier performs better than other classifiers and also concludes that information gain ratio (IGR) based feature selection method is better than other feature selection methods Aggarwal Random Tree algorithm was used in et al. Weka tool. The results were analyzed for each class of attributes to improve the Detection Rate (DR) and minimize the False Alarm Rate (FAR) Siddiqui K-means clustering was used to et al. construct 1000 clusters from 494,020 records and established a relationship between types of attacks and protocols which are used in intrusion Ingre et al. Used Artificial Neural Network (ANN) to analyze NSL-KDD dataset. DR for intrusion detection was found to be 81.2% and attack type classification was 72.9%

Demerits They used labelled NSL-KDD - a dataset for testing their method’s effectiveness

Just introduced some of the challenges that face using ML techniques in intrusion detection systems

Different types of datasets like Packet level, Net flow and Public data (like KDD 1999 and DARPA 1999 datasets) are used

Results were found by performing Five folds cross validation on NSL-KDD dataset

NSL-KDD cup 99 dataset was used

NSL-KDD dataset was used

NSL-KDD dataset was used

(continued)

A Review on Network Intrusion Detection System Using Machine Learning

605

Table 1. (continued) S. no. Authors

Merits

Demerits

11.

Moustafa et al.

The datasets, KDD 99 and UNSW-NB15 are used

12.

Parsazad et al.

13.

Rampure et al.

14.

Hasan et al.

15.

Janarthanan et al.

16.

Aminanto et al.

Used association rule mining algorithm. Accuracy and false alarm rate (FAR) were calculated using classifiers. The results show that features of UNSWNB15 are much more efficient than the KDD 99 dataset This suggested technique performed better than similarity-based algorithms which are already existing based on computational cost. Recommended feature selection based on rough set theory performed on KDD Cup99 dataset. Facts from data which was imperfect were exposed Random Forest chooses the most significant features desired for classification and increases accuracy by reducing time complexity It was found that kappa statistics enhanced due to classification using selected features An accuracy of 99.918% and a FAR of 0.012% was achieved

KDD 99 dataset was used

KDD Cup99 dataset was used

RRE-KDD is used for the evaluation of Random Forest (RF)

UNSW-NB15 dataset was used

AWID dataset is available only after registration

4 Proposed System and Conclusion Numerous methods for the effective advancement of security systems are documented in the literature. However, study on the enhancement of datasets used for training and testing purposes of such security systems is also important. In this paper, in depth analysis of CIDDS-001 dataset is shown and the sightings are presented. This paper provides the detailed overview about the working procedure of multiple research techniques along with the merits and demerits. From the comparison of the research papers it is proved that the most recent method by Verma et al. [1] of using machine learning algorithms on CIDDS-001 dataset provide the better results than the existing research method. To improve the computational time and cost we can implement Deep Learning Algorithms on the latest datasets like CIDDS-001 and CIDDS-002. If we can train the system using live data or online data by capturing them from real time networks, the results can be more accurate.

606

T. Rupa Devi and S. Badugu

References 1. Verma A, Ranga V Statistical analysis of CIDDS-001 dataset for network intrusion detection systems using distance-based machine learning, Department of Computer Engineering, NIT Kurukshetra, India 2. Medaglia CM, Serbanati A (2010) An overview of privacy and security issues in the Internet of Things. In: The Internet of Things, Springer, New York pp 389–395 3. Machine learning for network intrusion detection, Luke Hsiao, Stanford University. [email protected], Stephen Ibanez, [email protected] 4. Shaya O (2008) Using machine learning in networks intrusion detection. https://doi.org/10. 13140/rg.2.1.2586.0321 5. Liao H, Lin C, Lin Y, Tung K (2013) Intrusion detection system: a comprehensive review. J Netw Comput Appl 16–24. https://www.researchgate.net/publication/281451813_Using_ Machine_Learning_in_Networks_Intrusion_Detection. Accessed 01 Jan 2019. 2013 (PDF) Using Machine Learning in Networks Intrusion Detection 6. Verma A, Ranga V (2018) On evaluation of network intrusion detection systems: statistical analysis of CIDDS-001 dataset using machine learning techniques, Department of Computer Engineering, National Institute of Technology, Kurukshetra, Haryana, India 7. Garcia-Teodoro P, Diaz-Verdejo J, Maciá-Fernández G, Vázquez E (2009) Anomaly-based network intrusion detection: techniques, systems and challenges. Comput Secur, 28(1): 18–28 8. Sommer R, Paxson V (2010) Outside the closed world: on using machine learning for network intrusion detection. In: 2010 IEEE symposium on security and privacy (SP). IEEE, pp 305–316 9. Lippmann RP, Fried DJ, Graf I, Haines JW, Kendall KR, McClung D, Zissman MA (2000) Evaluating intrusion detection systems: the 1998 DARPA off-line intrusion detection evaluation. In: Proceedings of the DARPA information survivability conference and exposition, 2000, DISCEX 2000, vol 2. IEEE, pp 12–26 10. Flach P (2012) Machine learning: the art and science of algorithms that make sense of data. Cambridge University Press, Cambridge 11. Cover T, Hart P (1967) Nearest neighbor pattern classification. IEEE Trans Inf Theory 13 (1):21–27 12. CIDDS-001 dataset (2017). https://www.hs-coburg.de/forschung-kooperation/forschungs projekte-oeffentlich/ingenieurwissenschaften/cidds-coburg-intrusion-detection-data-sets.html 13. Hsiao L, Ibanez S. Machine Learning for Network Intrusion Detection, Stanford University 14. Alsallal M (2017) Applying machine learning to improve your intrusion detection system, 17 January 15. KarsligЕl ME, Yavuz AG, Güvensan MA, Hanifi K, Bank H (2017) Network intrusion detection using machine learning anomaly detection algorithms. In: 2017 25th signal processing and communications applications conference (SIU), Antalya. http://ieeexplore. ieee.org/stamp/stamp.jsp?tp=&arnumber=7960616&isnumber=7960135 16. Pasocal C, Oliveira M, Valdas R, Filzmoser P, Salvador P, Pacheco A (2012) Robust feature selection and robust PCA for Internet traffic anomaly detection. In Proceedings IEEE INFOCOM, pp 1755–1763 17. Sommer R, Paxson V (2010) Outside the closed world: on using machine learning for network intrusion detection. In: IEEE symposium on security and privacy, pp 305–316

A Review on Network Intrusion Detection System Using Machine Learning

607

18. Buczak AL, Guven E (2016) A survey of data mining and machine learning methods for cyber security intrusion detection. In: IEEE communications surveys & tutorials, vol 18, no. 2, pp 1153–1176. http://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=7307098& isnumber=7475979 19. Biswas S (2018) Intrusion detection using machine learning: a comparison study. Int J Pure Appl Math 118:101–114 20. Aggarwal P, Sharma SK (2015) Analysis of KDD dataset attributes-class wise for intrusion detection. Proc Comput Sci 57:842–851 21. Siddiqui MK, Naahid S (2013) Analysis of KDD CUP 99 dataset using clustering based data mining. Int J Database Theor Appl 6(5):23–34 22. Ingre B, Yadav A (2015) Performance analysis of NSL-KDD dataset using ANN. In: 2015 International conference on signal processing and communication engineering systems (SPACES). IEEE, pp 92–96 23. Moustafa N, Slay J (2015) The significant features of the UNSW-NB15 and the KDD99 data sets for network intrusion detection systems. In: 2015 4th International workshop on building analysis datasets and gathering experience returns for security (BADGERS). IEEE, pp. 25–31 24. KDD Cup 1999 (2014). http://kdd.ics.uci.edu/databases/kddcup99/ 25. UNSW-NB15 dataset (2017). https://www.unsw.adfa.edu.au/australian-centre-forcybersecurity/cybersecurity/ADFA-NB15-Datasets/ 26. Parsazad S, Saboori E, Allahyar A (2012) Fast feature reduction in intrusion detection datasets. In: Proceedings of the 35th international convention MIPRO Opatija, Croatia. IEEE, pp 1023–1029 27. Rampure V, Tiwari A (2015) A rough set based feature selection on KDD CUP 99 data set. Int J Database Theor Appl 8(1):149–156 28. Hasan MAM, Nasser M, Ahmad S, Molla KI (2016) Feature selection for intrusion detection using random forest. J Inf Secur 7(03):129–140 29. Janarthanan T, Zargari S (2017) Feature selection in UNSW-NB15 and KDDCUP’99 datasets. In: Proceedings of 26th international symposium on industrial electronics (ISIE), Edinburgh, UK. IEEE, pp 1881–1886 30. Aminanto ME, Choi R, Tanuwidjaja HC, Yoo PD, Kim K (2018) Deep abstraction and weighted feature selection for Wi-Fi impersonation detection. IEEE Trans Inf Forensics Secur 13(3):621–636 31. AWID (2018) AWID dataset. Accessed 2 Jan 2018. http://icsdweb.aegean.gr/awid/ download.html

Review on Facial Expression Recognition System Using Machine Learning Techniques Amreen Fathima(&) and K. Vaidehi Department of Computer Science and Engineering, Stanley College of Engineering and Technology for Women, Abids, Hyderabad 500 001, India [email protected], [email protected]

Abstract. Facial expressions are the most convenient way of expressing one’s thoughts. The aim facial expression recognition (FER) algorithm is to extort the discriminative and distinguishable characteristic of a face. Multiple methods have been devised to identify face and facial expression. Facial expressions not only depict the feelings of any individual but it is also used to judge his/her intellectual views. This paper not only includes the introduction of the face detection and facial expression recognition but also provide an exploration on the recent previous researches to extract the useful and capable method for facial expression recognition. The identification of various facial expressions are done through geometric features, appearance features and hybrid features. This paper presents a literature summary of the various strategies used for facial expression reputation. The comparative study is also carried out using various preprocessing, feature extraction and classification techniques used for facial expression recognition. Keywords: Face detection
Facial expression recognition Feature extraction
Classification




1 Introduction Sirovich and Kirby introduced the foremost facial recognition method in 1988. Initially they apply linear algebra to the crisis of facial identification, which is known as the Eigen face method and started as a research for a low-dimensional representation of facial images. Facial expression is most commanding, likely and extreme means of human beings to communicate their feelings and intentions. The facial expression contains one or more action or position of the muscles beneath the skin of the face [1]. The one set of divisive theory, these actions expresses the emotional state of an human being to observe. Facial expression is one of the most frequent non-verbal ways that humans exploit to convey their internal emotional states and consequentially, plays a major role in interpersonal communications. The Facial Expression is a visible manifestation of the affective state, cognitive activity, intention, individuality and psychopathology of a person. The non-verbal statement technique by which one can understand the state of a human being is the expression of face like happy, sad, fear, disgust, surprise and anger. © Springer Nature Switzerland AG 2020 S. C. Satapathy et al. (Eds.): ICETE 2019, LAIS 4, pp. 608–618, 2020. https://doi.org/10.1007/978-3-030-24318-0_70

Review on Facial Expression Recognition System

609

Automatic facial expression recognition (FER) has become an attractive and demanding area of the computer visualization field and its application areas are not limited to intellectual state identification, security, regular therapy system, face expression fusion, lie detection, music for mood, automated training systems, operator fatigue detection etc. The preliminary form of a computer application is seen in portable platform and in other forms of technology, such as robotics. It is classically used as access control in security systems and can be compare to other biometrics such as fingerprint, iris recognition systems, and gait. The truth of the face identification system as a biometric technology is lesser than iris recognition and fingerprint recognition, it is widely taken due to its contactless and non-invasive process. The facial expression becomes popular as a commercial detection and a marketing device. The further applications include advanced human-computer interaction, video surveillance, automatic indexing of images, and video database. The key benefit of FER is capable to know person mass classification as it does not require the collaboration of the test subject to work. The system installed in the air-port, banks and other public places can recognize individuals among the crowd, without passerby even being aware of the system. Emotions have a great control on every human being in one or additional way. The feelings of humans are represented in many ways, such as facial expression, voice and body gesture. The facial expressions are one of the most significant methods of nonverbal communication especially in human. The movement of one or more than one muscles underneath the skin constitutes facial expression. The facial expressions are the facial change in response to a person’s internal emotional state, intention or communications [2]. It is the most observable and powerful sign of emotional state of mind. FER plays a vital role in Human Computer Interaction. Computer visualization system can interrelate with a human by interpreting facial expression in a normal way. Six major facial expressions are accepted universally such as sad, fear, disgust, happy, surprise neutral and anger which is shown in Fig. 1. In the facial recognition system quality measures are the important factors as large degrees of variations are possible in face images. The factors such as illumination, expression, pose and noise, affects the performance during the capturing of the image in facial recognition systems [3]. The facial recognition is the most accurate system among all the biometric system which used in railway stations and airports. The FER system is a technology which is capable of identification or verification of a person from a digital image. There are multiple methods in which facial recognition systems work, but generally they work by comparing the selected facial features from given image with the faces of the person within the database. It is also called as a biometric artificial intelligence based application that can individually identify a person by evaluating patterns based on the person’s facial textures and shape. It is a computer application for recognizing the facial expressions of any person either using an image or a video clip or the person itself. A Facial recognition is generally used for security purposes. Facial expressions recognition is still an activeopen research field of machine learning. It has several existing applications in various areas. The fully automatic and real time facial expression systems helps in understanding non-verbal facial gestures used in different applications like behavioral research, video-calling, computer vision systems health care, games and e-learning.

610

A. Fathima and K. Vaidehi

Fig. 1. Different facial expressions

The technique used for recognizing a facial expression are face detection, feature extraction and expression classification [4]. Face Detection: The preprocessing step for identifying facial expression is face detection. The steps implicated in converting an image to a normalize facial image for facial characteristics extraction which is used for detecting characteristic points, locating, rotating to line up and cropping the face region using a rectangle, according to the face model. The face detection involves methods for classifying faces in a single image. Feature Extraction: Feature extraction change pixel data into a higher level representation of shape, motion, color, texture, and spatial input space. The reduction process should uphold essential information as it is an important task in pattern identification system. Various techniques are used for feature extraction. Expression Classification: Expression classification is performing by a classifier, which consists of model of pattern sharing which is attached to a decision procedure. To recognize expressions various classification techniques are used.

2 Review on Databases Used for Facial Expression Recognition FER Database In the field of FER [5], frequent databases have been used for relative and extensive experiments. The 2D static images or 2D video sequence are used to calculate human

Review on Facial Expression Recognition System

611

facial emotions. A 2D- large pose variation and facial behaviors has the complexity in handling 2D based examinations. It briefly introduces some accepted databases related to FER consisting of 2D and 3D video sequence and motionless images. The Extended Cohn-kanade Dataset (CK+): CK+ [6] contains both posed and nonposed emotion, and 593 video sequences along with added types of metadata. The database consist of 123 subjects from 18–30 years, most of them are female. Prototypes and action units are used to measure and analyze the image. It provides results for protocol and baseline emotion recognition, AUs, and facial feature tracking. The image has a pixel resolution of 640  480 and 640  490. Japanese Female Facial Expressions (JAFFE): The JAFFE database [7] contains 213 images of seven facial emotions (six basic facial emotion and one neutral) posed by ten different female Japanese models. Each image is based on six emotions using 60 Japanese person images. The original size of each facial image is 256 pixels  256 pixels (Fig. 2).

Fig. 2. Sample Images from JAFFE Database

3 Literature Review Banu, Danciu et al. [8], implemented a novel approach for face expression recognition. In this work the face features are extracted by means of Haar classifiers with openCV library. The faces are rotated so that the lines which are connected to the eyes are kept parallel. The exact eye curve is recognized and approximated this curve by using bezier curve. The pixels are eliminated on behalf of the skin. The three features for each eye and two feature of mouth is extracted. The facial appearance is extracted and classified by means of neural networks. The facial expressions are separated into moduleby using K-means categorization. Zhen, Zilu et al. [9] FER based on adaptive local binary pattern and sparse representation (SRC) approaches are used in this work. SRC algorithm is helpful to both the gray expression images and ALBP features of appearance images. The facial appearance is detected by using GRAY + SRC and ALBP + SRC methods. Initially input the matrix of instruction samples for k classes. Reshape the instructionsample by using GRAY + SRC and test the sample of the vectors by stacking its column, then calculate the categorization by using SRC. When the categorization results of GRAY + SRC and ALBP + SRC algorithms are the same then the final class label remains unaffected. The categorization is done based on the SVM, LDA, KPCA classifiers.

612

A. Fathima and K. Vaidehi

Deepthi, Archana et al. [10] implemented FER using Artificial Neural Networks. Image processing techniques is used to improve, develop or modify an image and to get ready it for image analysis. It is divided into many sub process, including histogram analysis, thresholding, masking, edge detection, segmentation and others. A 2D DCT is used for feature extraction and neural network is used like a classifier by using JAFFE database. Liu, Song et al. [11] implemented FER based on discriminative dictionary learning. Preprocessing is done by using gray value feature and local binary patterns. Gray value features are applied in the conventional SRC based face detection. FER experiments are also conducted by means of gray value facial appearance as a baseline. A local binary pattern (LBP) is an efficient texture explanation operator and can measure and extort the texture information on the local neighborhood in gray images. The LBP methods calculate each pixel of the image and the binary association of local neighborhood points on the grayscale. Then the binary relationship weighted to form the LBP code. Finally, the LBP histogram series of image sub-region can be regard as the image characterization. Gabor has applied widely in image analysis for its outstanding description of texture. Hence, in this work used five-scale, eight-direction Gabor filters. It is also used to extract facial features. Gabor filter is used for feature extraction and SRC (D-KSVD) is used as a classifier by using JAFFE database. Punitha, Geetha et al. [12] implemented “HMM based real time facial expression recognition”. Face region isidentified by using HMM such as mouth which plays an important role in expressing feeling and its facial appearance which is used for classifying expressions. The mouth concentration code value (MICV) extracted from the mouth region is used. The MICV difference between the first and the greatest facial appearance intensity frame is used as an input to a Hidden Markov Model (HMM) and HMM is used as a classifier, with the own created dataset and achieved 94% accuracy. Zhang, Liu et al. [13] implemented the FER using LBP and LPQ based on the Gabor wavelet transform. The pre-processing is done by using LBP and LQP. Gabor filter is used for feature extraction. PCA and LDA is used to reduce dimensions of features by Gabor LBP feature and Gabor LPQ features Multi class SVM classifier is used for classification by using JAFFE database. The accuracy obtained is 98%. Shah, Khanna et al. [14] implemented FER for Color Images using Gabor, Log Gabor Filters and PCA.. Gabor filters are used to extract the features from the images. The features are detected to extract the feature vector by using Gabor and Log Gabor filter. Dimensions are reduced to extract features by using Principle Component Analysis (PCA). The Euclidian distance is used to classify the reduced features. The self-database is used for testing with an accuracy of 86.7%. Lajevardi, Husain et al. [15] implemented “Feature extraction for facial expression recognition based on hybrid face regions”. A FER system is built based on hybrid face regions (HFR). Using Log Gabor filter features are extracted based on whole face image and face region. Principle component analysis PCA, mutual information MI and independent component analysis ICA is used for feature extraction. Naïve Bayes is used as a classifier. JAFFE database and Cohn-kanade database are used for testing with an accuracy of 97% and 91% respectively. Rejila, Menon et al. [16] implemented “Automatic Facial Expression Recognition based on the Salient Facial Patches”. PCA and LDA are used for feature extraction by

Review on Facial Expression Recognition System

613

generating a high dimensional feature vector. The database used inthis work is JAFFE database. SVM and ANN are used for classification of the data. The accuracy obtained is 97%. The low resolution images give the best images. ELLaban, Ewees, Elsaeed et al. [17] implemented “A Real-Time System for Facial Expression Recognition using Support Vector Machines and k-Nearest Neighbor Classifier”. The pre-processing is done by using Viola-Jones approach. Gabor, PCA are used for feature extraction. SVM and KNN classifiers are used for classification of the features extracted. The accuracy we achieve by testing the self-database using SVM is 90%. SVM outperformed than KNN. Hernansez Matamoros et al. [18] implemented facial expression recognition with automatic segmentation of face regions”. Appearance based is done for preprocessing and gabor functions are used to extract the features. Classification is done by using SVM classifier and achieved 99% accuracy. Sumathi, Santhanam and Mahadevi et al. [19] implemented, “Automatic Facial expression analysis” using Facial Action Coding System(FACS) action units and the methods which recognize the action units parameter using facial expression data that are extracted, various kinds of human facial expressions are recognized based on their geometric facial appearance, and hybrid features. The two essential concepts of extracting features are based on facial deformation and facial motion by using RUFACS record achieved good accuracy. Siddiqi, Ali et al. [20] implemented “Depth Camera-Based Facial Expression Recognition System Using Multilayer Scheme”, Depth camera, Principal component analysis, Independent component analysis, Linear discriminate analysis, Hidden Markov model are the techniques used. A hierarchical classifier was used, where the expression group was recognized at the first level, followed by the actual expression recognition at the second level, achieved animportant improvement in accuracy 98.0%. Lee, Uddin and Kim et al. [21] implemented Spatiotemporal Human Facial Expression Recognition Using Fisher Independent Component Analysis and Hidden Markov Model. Cohn-kanade database is used to detect features. The FICA Fisher Linear Discriminant (FLD) is used for feature extraction based on a class specific learning algorithm. The idea of this method is to find a best local presentation of face images in a low dimensional space and to acquire the feature space having also temporally evolving shape. The hidden markov model is used for classification by using cohn-kanade database and achieved 92% accuracy.

4 Proposed System The methodology requires preprocessing, feature extraction, classification after face detection and using a strong classifier. Steps: 1. Image preprocessing: Image pre-processing implies the operation done on any image prior to using it for either training or testing. This process is essential in the view to remove the local variations present in an image and thereby improving feature extraction process.

614

A. Fathima and K. Vaidehi

2. Feature extraction: Every specific patterns has some invariable features. It includes color, shape, dimension and many others attributes which can be easily seen. Features of every patterns are different from others because of these features that every patterns. 3. Classification: SVM is used as a classifier. It is based on simple ideas of hyper plane and leads to high performance in practical application. SVM classifier

Preprocessing

Input image

Feature Extraction

Training fer

Testing image

face recognizied

5 Comparitive Study on Various Methods for FER S.no Author Title 1

[8]

2

[9]

3

[10]

4

[11]

5

[12]

6

[13]

Expressions Face detection

A novel approach for face Angry, expression recognition disgust, fear, happy, neutral, sad Facial expression Angry, recognition based on disgust, adaptive local binary fear, happy, pattern and sparse neutral, sad, representation surprise Facial expression Happy, sad, recognition using ANN normal Facial expression Angry, recognition based on disgust, discriminative distance fear, happy, learning Neutral, sadnes, surprise HMM based Real time Happy, facial expression surprise, recognition disgust and normal FER using LBP and LPQ Angry, based on gabor wavelet disgust, transform based on gabor fear, happy, face image neutral, sad and surprise

Appearance based

Fusion approach

Feature Expression Accuracy extraction classification (%) FeedBeziercurve, K- forward mean neural network Gabor SRC filter

85%

70%

DCT

NN

–

Gabor filter

SRC (DKSVD)

94.3%

Self database MICV

MICV

94%

JAFFE database

SVM

82% 98%

Appearance based (1) Gray (2) LBP (3) Gabor

LBP, LPQ, Gabor wavelet, PCA, LDA

(continued)

Review on Facial Expression Recognition System

615

(continued) S.no Author Title

Expressions Face detection

7

[14]

FER for color images using gabor, loggabor filter and PCA

Happy, neutral, surprise

8

[15]

Feature Extraction for facial expression recognition based on hybrid face region

9

[16]

10

[17]

11

[18]

12

[19]

13

[20]

14

[21]

Anger, disgust, happy, sad, fear, surprise Automatic facial Anger, fear, expression recognition disgust, based on the salient facial happiness, patches sadness, and surprise A Real time system for Anger, facial expression disgust, recognition using support fear, happy, vector machines and knervous, nearest neighbor classifier sad, surprise A Facial Expression Disgust, Recognition with sad, smile, Automatic Segmentation surprise, of Face Regions anger, fear, neutral Automatic facial Happy, expression analysis using Angry, Sad, facial action coding Fear, system (FACS) Disgust, and Surprise Depth camera-based facial Anger, expression recognition disgust, system using multilayer fear, happy, nervous, scheme sad, surprise Spatiotemporal human Angry, facial expression disgust, recognition using fisher fear, happy, neutral, sad independent component analysis and hidden markov model

Feature Expression Accuracy extraction classification (%)

Self database PCA, SVM LDA, Gabor wavelet Cohn-kanade PCA, MI, NB database ICA

86%

91% 98%

Appearance based

LDA, LBP, PC

SVM ANN

97%

Viola-Jones approaches

Gabor Feature

SVM KNN

90%

Appearance based

Gabor Function

SVM

99%

Musclebased approach

–

–

–

Appearance based

LDA, ICA, PCA

HMM

98%

FICA

FLDA

HMM

92%

616

A. Fathima and K. Vaidehi

Comparision Between Merits and Demerits of the Related Work S.no 1

Author [8]

2

[9]

3

[10]

4

[11]

5

[12]

6

[13]

7

[14]

8

[15]

9

[16]

Title A Novel Approach on Face Expression Recognition A Facial Expression Recognition Based on Adaptive Local Binary Pattern And Sparse Representation Facial Expression Recognition Using ANN Facial Expression Recognition Based on Discriminative Distance Learning HMM Based Real Time Facial Expression Recognition FER Using LBP And LPQ Based on Gabor Wavelet Transform Based on Gabor Face Image FER For Color Images Using Gabor, Loggabor Filter And PCA

Merits Feed-forward achieved high accuracy It uses a new algorithm which solves sparse representations on raw images Neural network achieves good accuracy Gabor filter Works good and gains a good accuracy

Demerits It is not able to find the expression when the Eyes are closed Achieved less accuracy

Accuracy gets increased with different database

Low recognition rate

It shows impressive performance and strong Robustness

Feature Extraction For Facial Expression Recognition Based on Hybrid Face Region Automatic Facial Expression Based on The Sailent Facial Patches

Accuracy gets increased with different database and is robust Detects some facial points accurately with less cost. Expression recognition accuracy is high and computational cost is significantly less

In low recognition rate, it is still a challenge issue in the person independent Case Low resolution images are not detected correctly. Databases are very limited and are not easily available Faces some difficulty in using different classifier

Log Gabor filters performs better than Gabor filters

It is not able to find all the expressions in DCT Low recognition rate

Performed better in low resolution images

(continued)

Review on Facial Expression Recognition System

617

(continued) S.no 10

Author [17]

11

[18]

12

[19]

13

[20]

14

[21]

Title A Real Time System For Facial Expression Recognition Using Support Vector Machine And KNearest Neighboring Classifier Facial Expression Recognition with Automatic Segmentation of Face Regions Automatic Facial Expression Analysis using Facial Action Coding System Depth Camera-Based Facial Expression Recognition System Using Multilayer Schemes Spatiotemporal Human Facial Expression Recognition using Fisher Independent Component Analysis And Hidden Markov Model

Merits The SVM classifier performance is the highest recognition rate for facial expressions and image recognition

Demerits K-NN classifier performance is the lowest recognition rate

Achieves good ROI under varying illumination conditions

Low complexity classifier

Attains good accuracy

Faces some difficulty in different classifier

It uses depth camera which hides sensitive information

No online validation is done

High accuracy in the presence of various algorithms

Low recognition rate

6 Conclusion This paper has reviewed on facial expression recognition system. FER used in many applications such as medical, lie detection, cognitive activity, robotics interaction, forensic section, automated training systems, security, intellectual state identification, music for mood, operator fatigue detection, etc. The publically available FER databases are explained in this paper. The technique used for recognizing a facial expression are face detection, feature extraction and expression classification. Hence, various feature extraction and classification techniques used by researchers for FER is compared. According to the comparative analysis Gabor function with SVM classification method provides 99% accuracy and it recognized the several expressions such as happy, smile, sad, anger, fear, neutral. Most of the previous work has done with CK + and JAFFE databases.

618

A. Fathima and K. Vaidehi

References 1. De A, Saha A, Pal MC (2015) A human facial expression recognition model based on Eigen face approach. Proc Comput Sci 45:282–289 2. Siddiqi MH, Alruwaili M, Bang J, Lee, S (2017) Real time human facial expression recognition system using smartphone. Int J Comput Sci Netw Secur 17(10):223–230 3. Teo WK, De Silva LC, Vadakkepat P (2004) Facial expression detection and recognition system. J Inst Eng 44(3) 4. Kumar S, Gupta A Facial expression recognition. In: Special conference issue: national conference on cloud computing & big data 5. FER dataset. https://www.kaggle.com/ 6. Ck + dataset. http://www.consortium.ri.cmu.edu/data/ck/ 7. JAFFE database. https://www.kaggle.com/c/emotion-detection-from-facial-expressions 8. Banu SM, Danciu GM, Boboc RG, Moga H, Bălany C (2012) A novel approach for face expressions recognition. In: 2012 IEEE 10th jubilee international symposium on intelligent systems and informatics (SISY), IEEE, pp. 537–541 9. Zhen W, Zilu Y (2012). Facial expression recognition based on adaptive local binary pattern and sparse representation. In: 2012 IEEE international conference on computer science and automation engineering (CSAE), vol. 2, IEEE, pp 440–444 10. Deepthi S, Archana GS, JagathyRaj VP (2013) Facial expression recognition using artificial neural networks. IOSR J Comput Eng (IOSRJCE), 8(4):01–06. ISSN 2278-0661, ISBN 2278-8727 11. Liu W, Song C, Wang Y (2012). Facial expression recognition based on discriminative dictionary learning. In: Proceedings of the 21st international conference on pattern recognition (ICPR2012), IEEE, pp 1839–1842 12. Punitha A, Geetha MK (2013) HMM based real time facial expression recognition. Int J Emerg Technol Adv Eng 3(1):180–185 13. Zhang B, Liu G (2016) Facial expression recognition using LBP and LPQ based on gabor wavelet transform. In: IEEE international conference on computer and communications 14. Shah SK, Khanna V (2015) Facial expression recognition for color images using Gabor, log Gabor filters and PCA. Int J Comput Appl 113(4) 15. Lajevardi SM, Hussain ZM (2009) Feature extraction for facial expression recognition based on hybrid face regions. Adv Electr Comput Eng 9(3):63–67 16. Rejila RC, Menon M (2016) Automatic facial expression recognition based on the salient facial patches. Int J Sci Technol Eng 2(11) 17. ELLaban HA, Ewees AA, Elsaeed AE (2017) A real-time system for facial expression recognition using support vector machines and k-nearest neighbor classifier. Int J Comput Appl 159(8):0975–8887 18. Hernandez-matamoros A, Bonarini A, Escamilla-Hernandez E, Nakano-miyatake M (2015) A facial expression recognition with automatic segmentation of face regions. In: International Conference on Intelligent Software Methodologies, Tools, and Techniques, pp 529–540. https://doi.org/10.1007/978-3-319-22689-7 19. Sumathi CP, Santhanam T, Mahadevi M (2012) Automatic facial expression analysis a survey. IEEE Int J Comput Sci Eng Surv 3(6):47 20. Siddiqi MH, Ali R, Sattar A, Khan MA, Lee S (2014) Depth camera-based facial expression recognition system using multilayer scheme. IETE Tech Rev 31(4):277–286 21. Lee JJ, Uddin MZ, Kim TS (2008) Spatiotemporal human facial expression recognition using fisher independent component analysis and hidden markov model. In: 2008 30th annual international conference of the IEEE engineering in medicine and biology society, EMBS 2008, IEEE, pp 2546–2549

Document Clustering Using Different Unsupervised Learning Approaches: A Survey Munazza Afreen(&) and Srinivasu Badugu Department of Computer Science and Engineering, Stanley College of Engineering and Technology for Women, Abids 500 001, Hyderabad, India [email protected], [email protected]

Abstract. One of the fastest growing research fields in recent times is Document Clustering. It gained its importance in text mining due to the tremendous increase of documents on internet. Textual data management is the need of every organization and clustering the documents is one of the fastest and widely used techniques. Document clustering is an unsupervised technique that organizes similar documents into classes in order to improve information retrieval. The overall evaluation of the research work is performed by comparing the working procedure and merits of each method with other in terms of some performance metrics. This research work concluded with the better research method which can be applied to cluster the similar documents on the basis of text. The comparative analysis shows the accuracy of documents clustered using K-Means are high compare to other approaches of unsupervised learning in terms of FMeasure, Precision, recall and time complexity. Keywords: Unsupervised learning
Document clustering
Cosine similarity
Euclidean measure
F-measure
K-Means

1 Introduction Clustering is the unsupervised classification of data or text that has no predefined class. It is a technique that groups the data or text that are similar to one another into a class or clusters without any prior Knowledge. Clustering is useful in Machine Learning, information retrieval, Pattern recognition and Pattern classification. Document Clustering is the process of organizing documents together in a cluster or class based on their text or their content. The document clustering involves data that is unlabelled, different types of data sets are used like news paper dataset, web pages, links, Microsoft datasets, IBM datasets, general datasets involving emails from an organization to clients, complaint letters and so on. The main aim of document clustering is to cluster the documents that are similar to one another in a single label or class or a cluster by finding out how similar the words in the documents are to each other using different approaches thereby increasing the inter cluster distance and minimizing the intra cluster distance in unsupervised learning. The Unsupervised Learning algorithms of Machine Learning are used in clustering the data whose class or clusters are unknown. The algorithms helps in finding a pattern, © Springer Nature Switzerland AG 2020 S. C. Satapathy et al. (Eds.): ICETE 2019, LAIS 4, pp. 619–629, 2020. https://doi.org/10.1007/978-3-030-24318-0_71

620

M. Afreen and S. Badugu

structure or a similar group in dataset that can be used as a base to classify the data or text. The different types of approaches in unsupervised clustering are Euclidean measure, K-means, Pearson Correlation, improved K-Means, Jaccard co-efficient, density based clustering, class based clustering, soft clustering and hard clustering, partitional hierarchical clustering, different weight based algorithms, cosine similarity, Mahalanobis Distance to name a few. These approaches are used to find a pattern or similarity in a document and cluster them together or label them. The accuracy is calculated in terms of recall, entropy, F-measure.

2 Analysis of Different Mahcine Learning Algorithms for Clustering In this section, detailed overview of the different research techniques are given with their working procedure. The author paper [1] performed a comparative study for traditional K-Means and Improved K-Means algorithm on different data set, in which traditional K-Means has the problem of choosing initial clustering centers that results in more number of iterations than compared to Improved K-Means that uses the concept of Kruskal’s Algorithm and Minimum Spanning Tree technique that results in increase in accuracy and less number of iterations. It was found there is an increase in accuracy using improved K-Means which was Optimal (Fig. 1).

Fig. 1. Comparison between K-Means and improved K-Means. Source adapted from [1]

The author of Paper [2] performed a comparative study on different distance measures or similarity measures like Euclidean distance, Jaccard, Pearson, KLD, cosine measures and concluded that Euclidean gives the worst results in terms of purity and entropy and Jaccard, Pearson better and slightly higher purity measures i.e., coherent clusters. Different types of dataset were used to perform the comparative study and the Pearson measure has the best result on any given data set.

Document Clustering Using Different Unsupervised Learning Approaches

621

The paper [3] surveyed clustering algorithms for text data. Among different classes of algorithms, distance based methods are popular and widely used. The main area of the study was the types of datasets, Dynamic Applications and Heterogeneous Applications. Dynamic application consists of data from social networks or online chat applications that requires cleaning before clustering and heterogeneous applications the text is available in context of links and other heterogeneous multimedia scenarios. In paper [4] the classification is done on non-linear dataset using different weighing attributing methods namely, attribute weighting based K-Means clustering (KMCBAW), weighted attribute using fuzzy c means (FCMCBAW) and subtractive clustering based attribute weighting (SCBAW) and applied to discriminate two medical datasets ANFIS and C4.5 decision tree classifiers were used to classify the weighted datasets. The high classification performance was obtained from Subtractive clustering based attributed weighing Algorithm. The author of [5] attempted to improve the document clustering problem in web domain in four different components. The entire process consists taking web documents as input, and then document structure identification is done. Well structured XML documents are expected if they are not those then Document Index Graph Representation is done that result in phrase matching, Document similarity calculation is carried out and incremental clustering is done resulting in document Clustering. First component consists of breaking down the web documents into sentences and identifying their weights. The Second component is based on index web documents using their phrases and significance, the phrase matching, quality of clustering and similarity is robust, accurate and effective. The Third component is phrase-based similarity measure for accurate calculation of pair-wise document similarity. The Fourth component is maintaining high cluster cohesiveness by improving pair-wise document similarity distribution inside each cluster. These four components are combined together and performed on a dataset yields a better and accurate result than the traditional clustering methods (Figs. 2, 3 and 4).

Fig. 2. Web document clustering system design. Source adapted from [5]

622

M. Afreen and S. Badugu

Fig. 3. Incremental construction of the document index graph. Source adapted from [5]

Fig. 4. Cluster similarity histogram. Source adapted from [5]

In [6] the method, ZOOM-IN, a local feature selection method for partitional hierarchy clustering, in which all documents are classified into, sub-clusters. It automatically defines the number of selected features in each iteration. The results were satisfactory and it also showed the benefits of selecting features locally. The following graphs shows the results of Precision for Reuters-Global and Reuters-Local (Fig. 5). The paper [8] discusses about the different techniques and improvements of KMeans clustering algorithm based on different research papers referred like K-Means, refined initial centroid selection method, parallel K - Means in which Modified KMeans algorithm with dynamic clustering of data, MRT for parallel K-Means. Implementation of K-Means clustering algorithm gives best result in terms of speed up, scale up, size up for a large data from this review paper and their limitations are discussed. Hierarchical clustering is done on the documents and diagrammatically it is represented as shown in Fig. 6.

Document Clustering Using Different Unsupervised Learning Approaches

623

Fig. 5. Micro-averaged precision for the Reuters collection, with local and global feature selection. Source adapted from [6]

Fig. 6. Block diagram of document clustering. Source adapted from [8]

The author of [9] performed a detailed analytical on the comparison of well known clustering algorithms like K- Means, PAM, CLARA, CLARANS, DBSCAN etc. to make it easy for a user to select clustering algorithm as per their specifications. The clustering process consists of Data step followed by Feature selection or feature extraction, clustering algorithm design or selection is performed that get validated and results are interpreted (Fig. 7).

Fig. 7. Clustering process. Source adapted from [9]

624

M. Afreen and S. Badugu

The paper [10] performed a comparative study using five different similarity measures namely Euclidean Measure, Cosine Similarity, Mahalanobis Distance, Pearson Correlation Coefficient using k-means clustering algorithms and Jaccard Coefficient on seven different datasets with different characteristics. It was concluded that the Euclidean measure which is the most popular and used as a default distance measure performs the worst whereas Pearson Correlation and Jaccard Coefficient gives most coherent clusters with a high purity value. The overall system Architecture consists of Plain Text documents as input, pre-processing techniques like tokenization, stemming and stop words removal are performed, tf-idf is calculated and similarity measure is applied that has two approaches, K- Means and K-Means++ are performed that results in Clustering documents as output (Fig. 8).

Fig. 8. Overall system architecture. Source adapted from [10]

Tatsunori et al. [16] “Term Weighting Method based on Information Gain Ratio for Summarizing Documents retrieved by IR systems” introduced a method utilize the similarity information among original documents by hierarchical clustering, information gain ratio of a word as a term weight and this method proves to be effective in providing summarization of retrieved documents. Clustering retrieved documents consists of all documents grouped into retrieved document that can be further classified into sub clusters. Figure 10 shows the word distribution and portioning of clusters into different sub clusters. Information gain ratio is used to weigh the terms in the documents. IR tasks summaries are evaluated. The overview of consists of giving retrieved document as input, tokenize and extract noun followed by calculation of tf-idf for each word. The documents are mapped into vectors and hierarchical clustering is performed followed by calculation of information gain ratio of each word and weight of each sentence that results in set of summarize of documents as output (Figs. 9, 11, 12 and 13).

Document Clustering Using Different Unsupervised Learning Approaches

Fig. 9. Clustering retrieved documents. Source adapted from [16]

Fig. 10. Word distribution and partitioning of clustering. Source adapted from [16]

Fig. 11. Weighing terms by information gain ratio. Source adapted from [16]

625

626

M. Afreen and S. Badugu

Fig. 12. Evaluation of Summaries in IR task. Source adapted from [16]

Fig. 13. Scheme overview. Source adapted from [16]

Document Clustering Using Different Unsupervised Learning Approaches

627

3 Comparison of Research Methodologies This section provides the overall comparison view of the different research techniques in terms of varying merits and demerits. The following Table 1 provides the merits and demerits of the different research techniques which are discussed above.

Table 1. Comparison evaluation of the research methodologies S. No. Author

Title

Method

Merits

Demerits

1.

Abhilash CB

A Comparative Study on Clustering of Data using Improved K- means Algorithms

Improved K-Means: Less number of iterations and Higher accuracy

K-means: More number of iterations and problem of choosing initial clustering centroid

2

Anna Huang

Similarity Measures for Text Document Clustering

Charu C A

A Survey of Text Clustering Algorithms

4

Kemal P

Pearson correlation gives the best performance with high purity and entropy The text data in Dynamic Application requires cleaning Different weighing Subtractive attributes methods clustering based attributed weighing Algorithm gives the highest Performance

Euclidean gives the least results in purity and entropy

3

K-Means and Improved K-means using concept of Kruskal’s algorithm and Minimum spanning Tree A comparative study of different similarity methods in terms of purity and entropy Term strength and entropy based ranking

Breaks a single process into four different components

Traditional method The four different produces results component when combined together with low accuracy gives a higher result

5

6

7

Application of Attribute Weighting Method Based on Clustering Centers to Discrimination of Linearly NonSeparable Medical Datasets Khaled M Efficient PhraseH Based Document Indexing for Web Document Clustering Marcelo Local feature NR selection in text clustering Vrinda K

ZOOM-IN method Automatically defines the number of selected feature in each iteration K-Means proves to Efficient clustering K-Means,Parallel K-Means, modified be the best among of data using improved K-means K-Means with all algorithms algorithm: A dynamic clustering implemented with Review of data high accuracy and precision rate

The data is present in text links in Heterogeneous applications Lower Accuracy compared to Subtracting based algorithm

The results were only satisfactory

The introduced algorithms Failed to provide better results in the High dimensional dataset

(continued)

628

M. Afreen and S. Badugu Table 1. (continued)

S. No. Author

Title

Method

Merits

Demerits

8

Neha S

Detailed analysis on comparison if different clustering algorithms

User can choose a single algorithm based on their specifications

The desired features may not be present in a single algorithm

9

Pranjul S

Categorization of Several Clustering Algorithms from Different Perspective: A Review Text Document Clustering and Similarity Measures

10

Tatsunori M

Term Weighting Method based Information Gain Ratio for Summarizing Documents retrieved by IR systems

Comparative study Pearson Correlation of different and Jaccard similarity measures Coefficient produce a high purity value giving most coherent clusters Information Gain Ratio

The Euclidean similarity measure failed to provide better results High dimensional dataset with different characteristics Provides effective The user needs to summarization of selects one subretrieved documents cluster recursively to reach a desired document

4 Conclusion and Proposed System The better knowledge delivery can be guaranteed by extracting the association relationship between the different data. It is resolved by various researchers by introducing the context aware of different clustering techniques in unsupervised Learning. In this analysis work discussion about the various clustering techniques has been discussed. This work provides the detailed overview about the working procedure of multiple research techniques along with the merits and demerits. The overall evaluation of the research work is performed by comparing the working procedure and merits each method with other in terms of some performance metrics. This research work concluded with the better research method which can be applied to extract the useful information with the concern of context. From the numerical comparison of the research work it is proved that the most recent method namely improved K- Means leads to provide the better decision making outcome than the existing research methods. The Proposed system of this which is done by us is that it yields better results when weighted K-Means are implied on the documents to cluster them based on their similarity. The Precision, Recall and Accuracy deliver a better performance rate and there is always room for further improvement.

Document Clustering Using Different Unsupervised Learning Approaches

629

References 1. Abhilash CB, Basavanagowda S (2013) A comparative study on clustering of data using improved K- means algorithms. Int J Comput Trends Technol (IJCTT), 4(4). ISSN: 22312803 2. Huang A Similarity measures for text document clustering, Department of Computer Science 3. Aggarwal CC, Zhai CX A survey of text clustering algorithms. IBM T J Watson Res Cent Yorktown Heights 4. Polat K (2011) Application of attribute weighting method based on clustering centers to discrimination of linearly non-separable medical datasets. J. Med Syst 5. Hammouda KM, Kamel MS (2004) Efficient phrase-based document indexing for web document clustering. IEEE Trans Knowl Data Eng 16(10) 6. Ribeiro MN, Neto MJR, Prudˆencio RBC Local feature selection in text clustering, Universidade Federal de Pernambuco 7. Law MHC, Figueiredo MAT, Jain AK (2004) Simultaneous feature selection and clustering using mixture models. IEEE Trans Pattern Anal Mach Intell 6(9) 8. khairnar V, Patil S (2016) Efficient clustering of data using improved K-means algorithm: a review. Imperial J Interdisc Res (IJIR), 2(1). ISSN 2454-1362 9. Soni N, Ganatra A (2012) Categorization of several clustering algorithms from different perspective: a review. Int J Adv Res Comput Sci Softw Eng, 2(8). ISSN 2277 128X 10. Singh P, Sharma M (2013) Text document clustering and similarity measures, Department of Computer Science and Engineering 11. Alelyani S, Tang J, Liu H Feature selection for clustering: a review 12. Kunwar S (2013) Text documents clustering using K- means algorithm 13. Sharmila P, Shanthalakshmi Revathy J (2013) An efficient clustering algorithm for spam mail detection. Int J Adv Res Comput Sci Softw Eng 3(3). ISSN 2277 128X 14. Kolhe S, Sawarkar S (2015) Review of document clustering techniques: issues, challenges and feasible avenue. Int J Adv Res Comput Sci Softw Eng, 5(4). ISSN 2277 128X 15. Bisht S, Paul A (2013) Document clustering: a review. Int J Comput Appl 73(11) 16. Mori T, Kikuchi M, Yoshida K Term weighting method based on information gain ratio for summarizing documents retrieved by IR systems. Div Electr Comput Eng 17. Yafooz WMS, Abidin SZZ, Omar N, Halim RA (2013) Dynamic semantic textual document clustering using frequent terms and named entity. In: IEEE 3rd international conference on system engineering and technology 18. Jain Y, Nandanwar AK (2014) A theoretical study of text document clustering. Int J Comput Sci Inf Technol 5(2):2246–2251. ISSN 0975 9646 19. https://kdd.ics.uci.edu/databases/20newsgroups/20newsgroups.html 20. https://www.cs.cornell.edu/people/pabo/movie-review-data/

A Study of Liver Disease Classification Using Data Mining and Machine Learning Algorithms Hajera Subhani and Srinivasu Badugu(&) Department of Computer Science and Engineering, Stanley College of Engineering and Technology for Women, Abids, Hyderabad 500001, India [email protected], [email protected]

Abstract. The overall evaluation of the research work is based on the liver disease classification and prediction performed by comparing different working procedures and merits of each method with other works in terms of performance metrics. This research work concluded that SVM have the better performance rate when compared to different techniques for classification of the patients with the liver disease using different datasets. Keywords: SVM


Classification
Prediction
Liver

1 Introduction According to WHO globally second leading cause of death is cancer, in 2018 cancer is responsible for an estimated of 9.6 million deaths. Globally, cancer is causing about 1 in 6 death every day. Around the world 5 leading behavioral and dietary risks for causing cancer are: high mass index in human body, low intake of fruit and vegetable, no physical activity in daily life, high rate of tobacco intake, and alcohol use. In 2018 liver cancer leaded the cause of death worldwide with 782,000 deaths. According to times of India, in India one in five persons is affected by liver disease. It is expected by the experts that by 2025 India may become the ‘World Capital for Liver cancer’. According to the American Cancer Society, hepatitis C is the most common cause of liver cancer in the U.S. People with both hepatitis B or C have a significantly higher risk of developing liver cancer than other healthy individuals, as both forms of the disease can result in cirrhosis. Early identification of the liver cancer will play vital role for the survival of the patients. Automatic tools may reduce the burden on doctors Data mining and Machine learning has various algorithms for classification, such as SVM, NB, C4.5 Decision Tree, Random Forest, J48, MLP and Bayesian Network. In this paper, we performing the comparative study of liver disease classification using the above all techniques.

© Springer Nature Switzerland AG 2020 S. C. Satapathy et al. (Eds.): ICETE 2019, LAIS 4, pp. 630–640, 2020. https://doi.org/10.1007/978-3-030-24318-0_72

A Study of Liver Disease Classification Using Data Mining

631

2 Analysis Research Methodologies In this section, detailed overview of the different research techniques are given with their working procedure. Kefelegn [1] Prediction and Analysis of Liver Disorder Diseases by using Data Mining Technique: Survey. In this paper, dataset is collected from the uci data repository. Data cannot be given to the algorithm for training for this purpose data preprocessing is required. After preprocessing of data features are selected for the partitioning of the data in to training and testing data. Data mining technique for classification of patients with the liver disease and the patients who are not suffering from liver disease is applied such as SVM and NB. C4.5 Decision Tree considered for performance evaluation of liver disease classification using uci dataset. In this paper, classification experiments are performed using the Weka tool on the dataset. By comparing data mining classification algorithm, we got to know SVM accuracy score is 94.04% which is greater than NB using data mining approaches and 97.13% in machine learning which is also greater than NB approach. In this paper we got to know that SVM have the high performance rate when compared to NB while classification done using different domains such as machine learning and data mining (Fig. 1).

Fig. 1. Methodology work flow diagram

Nagaraj and Sridhar [2] NeuroSVM: A Graphical User Interface for Identification of Liver Patients. In this paper, Indian liver patient dataset is collected from uci data repository. After the collection of data from uci data repository dual feature selection process is carried out. The features which are not in used at the time of training are removes which is know as correlated features. In this paper, based on the applications of the different algorithm feature are selected by using boruta package for classification are used such as NB, Bagging, Random forest and SMV. It is implemented using R platform. All the above algorithm are used for classification of the patients with liver disease. Classification is performed on ILPD dataset using the above algorithms. The classification here classifies the patients with the liver disease and patients who are not

632

H. Subhani and S. Badugu

suffering from liver disease. Based on the performance rate of all the algorithms used in this paper, we got to know that SVM algorithm has the highest performance rate for classification when compared to all the above algorithm so in this paper hybrid neuroSVM model is developed and is constructed as a GUI using R platform. The overall workflow of this paper is shown below in the Fig. 2.

Fig. 2. Flowchart for development of NeuroSVM model

Pakhale and Xaxa [3] A Survey on Diagnosis of Liver Disease Classification. In this paper, ILPD (Indian liver Patient disease) is collected for training the model. As we cannot directly provide our data to our algorithm so preprocessing is required and integration of the dataset is done to improve the performance of our model the next step is feature selection and feature transformation is done for training our model. Data mining techniques have been used for classification such as Decision tree, SVM, NB and Artificial Neural Networks for liver patient dataset. In the evaluation and presentation phase SVM and Artificial Neural rate has a highest performance rate in their patterns when compared to NB, Decision tress. Classification is done of patients with liver disease and no diagnosis is done (Figs. 3 and 4).

Fig. 3. KDD process

A Study of Liver Disease Classification Using Data Mining

633

Fig. 4. General process of classification of data

Kiran Kumar, Sreedevi, Padmanabha Reddy et al [4] Survey on machine learning algorithms for liver disease diagnosis and prediction. In this paper, problem is defined for which the solution is shown in this paper. Data identification is done which is known as feature selection. Dataset splitting is done into two files such as training and tuning data. Training is used for training this model and tuning is used for testing the model. Based on features been selected machine learning algorithm such as supervised learning algorithms such as K-Nearest Neighbour and Support Vector Machine is used for classification of the liver disease patients. In the evaluation phase we got to know the performance rate of SVM is when compared to KNN (Fig. 5).

Fig. 5. Supervised learning process

Jatav and Sharma [5] An algorithm for classification data mining approach in medical diagnosis. In this paper, first phase of this model is preprocessing of the dataset. Data capturing, storing, Analyzing are the step of feature extraction. Data

634

H. Subhani and S. Badugu

capturing is the collection of data which are necessary for our model. Storing data means features of the data which are important for model to train to analyze it. Based on the analysis Features are extracted from the dataset to train the dataset of this model. This research paper is mainly focused to classification of patients with liver disease possibility using data mining and machine learning approach in order to enhance the accuracy and precision describes the performance rate of this model. This paper shows the related work study of different algorithms such as neural network, naïve bayes, SVM, KNN, FCN, etc. and it is concluded that SVM gives the best performance as compared to the other existing techniques. These paper has designed using SVM and RF algorithms. The algorithms results with accuracy of 99.35%, 99.37 and 99.14 on diabetes, kidney and liver disease respectively. SVM have the high experimental results when compared to RF in this work (Fig. 6).

Fig. 6. A typical health informatics processing

Swapna, Prasad Babu et al. [6] Critical Analysis of Indian Liver Patients Dataset using ANOVA Method. In this paper, ILPD data set is used for are Analysis of Variance (ANOVA). Analysis of different attributes such as gender, age and different liver function test such as SGOT, SGPT and Alkaline Phosphates. Each and every instance variable are analyzed in this project. Null values cannot be rejected in this method. Each value of every instance of the dataset is analyzed and studied carefully. Hashem and Mabrouk [7]. A Study of SVM Algorithm for Liver Disease Diagnosis. Here in this paper, two datasets are collected from two different data repository such as uci and kaggle. After collection of the datasets SVM algorithm is used for classification of patients with liver disease using MATLAB software, in the evaluation phase performance of the two different dataset are evaluated. Accuracy, Error rate, sensitivity, Prevalence and Specificity shows the performance. Based on the evaluation phase we got to know the ilpd dataset attributes has the high performance when compared to bupa dataset attributes. This paper is the comparison of performance of

A Study of Liver Disease Classification Using Data Mining

635

two datasets such as BUPA dataset and ILPD dataset for classification using the svm algorithm. After classification evaluation of the classification is done based on confusion matrix. we get to know precision, recall, error rate and sensevity which describes the performance rate of both the datasets, in which we got to know that ILPD dataset performance rate is high (Fig. 7).

Fig. 7. The overall process

Nancy, Sudha, Akiladevi et al. [8] Analysis of feature Selection and Classification algorithms on Hepatitis Data. This research paper mainly aims comparison of classification techniques on predicting the survival of the patients was done using feature selection algorithms. Hcc dataset is collected from the kaggle website. Data preprocessing is the essential step in every data mining project in order to provide the correct type of data to train the algorithm. Feature selection classification is carried out in the next phase. Evaluation of the performance of the different algorithms is done based on confusion matrix. The needs for classification We have analyzed the impact three feature selection algorithms towards classification efficiency and found that the algorithms towards feature selection did not yield promising results in classifying the hepatitis dataset. Analysis provide a discovering of new patterns of disease in many medical datasets of disease occurrence and unearth the effect of data mining techniques in the medical arena (Figs. 8 and 9).

636

H. Subhani and S. Badugu

Fig. 8. Framework of data mining process

Fig. 9. Performance of classification algorithms with feature selection

A Study of Liver Disease Classification Using Data Mining

637

Vijayarani and Dhayanand [9] Liver Disease Prediction using SVM and Naïve Bayes Algorithms. In this paper dataset is collected from uci repository. Data mining techniques such as Classification is used here such as Naïve bayes and Support Vector Machine (SVM) for classifification of patients with the liver disease and the patients who are not suffering from liver disease are classified.performance accuracy is checked of both the algorithms. The experimental results of this work concludes, the highest accuracy is observed in SVM when compared to Naïve Bayes classifier based on the execution time (Fig. 10).

Fig. 10. System architecture

Jacob, Mathew, Mathew, Issac et al. [10]. Diagnosis of Liver Disease Using Machine Learning Techniques. In this paper, they are using machine learning techniques such as SVM, Logistic Regression, KNN and Artificial Neural Network. The system was implemented using all the models and their performance was evaluated. Performance evaluation was based on certain performance metrics. ANN was the model that resulted in the highest accuracy with an accuracy of 98%. Comparing this work with the previous research works, it was discovered that ANN proved highly efficient. A GUI, which can be used as a medical tool by hospitals and medical staff was implemented using ANN. Banu Priya, Laura Juliet, Tamilselvi et al. [11]. Performance Analysis of Liver Disease Prediction Using Machine Learning Algorithms. In this paper, structured and unstructured patient data is collected including all symptoms like fever, pain, pain in abdomen etc., neural network is applied, PSO (particle swarm optimization algorithm) feature selection methods is used for Indian Liver Patient Dataset. Decision support system and final diagnosis is done. The algorithms used in this paper are J48, MLP, SVM, Random Forest, and Bayes network Classification. Particle swarm optimization algorithm (PSO) various result for feature selection. It is observed that bayes network and J48 Classification gives better results compare to other classification algorithms. After which user interactive screen is developed (Fig. 11).

638

H. Subhani and S. Badugu

Fig. 11. Overall architecture of proposed system

3 Comparison of Research Methodologies See Table 1. Table 1. Comparison evaluation of the research methodologies S. no. Author 1

2

3

4

Shambel Kefelegn

Title

Prediction and Analysis of Liver Disorder Diseases by using Data Mining Technique: Survey Kalyan Nagaraj NeuroSVM: A and Graphical User Amulyashree Interface for Sridhar Identification of Liver Patients Harsha Pakhale A Survey on and Deepak Diagnosis of Liver Kumar Xaxa Disease Classification

M. Kiran Kumar, M. Sreedevi and Y. C. A. Padmanabha Reddy

Methodologies

Merits

Demerits

SVM, NB and C4.5 Decision Tree

SVM is eager learner; hence it has a higher performance rate

NB, Bagging, Random forest and SMV using R platform

SVM is eager learner, hence it has a higher performance rate

C4.5 and NB are lazy learners as generalization takes time because of iterations NB and random forest are lazy learners

Data Mining techniques – Decision tree, SVM, NB, Artificial Neural Networks Survey on machine Various Machine learning algorithms learning for liver disease algorithms diagnosis and prediction

SVM and Artificial Neural rate has a higher performance rate

The Performance of NB and Decision tree have low performance rate Accuracy of SVM Low accuracy is 97.07% when rate compared to compared to the SVM rest algorithms

5

(continued)

A Study of Liver Disease Classification Using Data Mining

639

Table 1. (continued) S. no. Author

6

7

Methodologies

Merits

Demerits

Shakuntala An algorithm for Jatav and Vivek predictive data Sharma mining approach in medical diagnosis

Title

Data Mining techniques like SVM, Random forest

SVM has high accuracy

Low accuracy rate of random forest

K. Swapna and Critical Analysis of Prof. M. S. Indian Liver Prasad Babu Patients Dataset using ANOVA Method Esraa M. A Study of Support Hashem and Vector Machine Mai S. Algorithm for Liver Mabrouk Disease Diagnosis Analysis of feature Nancy, P., Sudha, V. and Selection and Akiladevi, R. Classification algorithms on Hepatitis Data

ANOVA (Analysis of variance)

Analysis of different attributes

Null values cannot be rejected

Performance rate and accuracy of SVM is done

MATLAB is expensive to access, license is required It does not predict other diseases and percentage of Hepatitis in patients NB has low performance rate compared to SVM

9

Dr. S. Vijayarani and Mr. S. Dhayanand

Liver Disease Prediction using SVM and Naïve Bayes Algorithms

SVM implemented using MATLAB software Data mining techniques – RND tree, Quinlan decision tree (c4.5), KNN algorithm etc. NB, SVM implemented using MATLAB tool

10

Joel Jacob, Joseph Chakkalakal Mathew, Johns Mathew and Elizabeth Issac M. Banu Priya, P. Laura Juliet and P.R. Tamilselvi

Diagnosis of Liver Disease Using Machine Learning Techniques

Machine leaning techniques – SVM, C4.5, KNN, ANN

Performance Analysis of Liver Disease Prediction Using Machine Learning Algorithms

Random Forest, SVM, J48, MLP, Bayesian Network

8

11

Classifies patients with Hepatitis disease

SVM is an eager learner and has higher performance rate. Highest classification accuracy Highest accuracy rate GUI using tkinter packaging python

Classification is done of the patients

No prediction is done, just classification is done

There is no prediction

4 Conclusion and Proposed Work The better classification can be guaranteed by comparing the different machine learning and data mining algorithm on different dataset. It is resolved by various researchers by introducing the different machine learning and data mining algorithm such as SVM, NB, C4.5 Decision Tree, Random Forest, J48, MLP and Bayesian Network are discussed. This work provides the detailed overview about the working procedure of multiple research techniques along with the merits and demerits. The overall evaluation

640

H. Subhani and S. Badugu

of the research work is performed by comparing the working procedure and merits each method with other in terms of some performance metrics. This research work concluded with the better research method which can be applied to extract the useful information with the concern of context. Based on the performance rate SVM and KNN has high accuracy and is suitable classification of the patients with liver than the existing research methods. The proposed system of this work which is done by us is predicting different types of liver disease by using five different datasets for training our model and for classification we are using svm, knn and naïve bayes algorithms and for predicting different liver diseases linear regression is used.

References 1. Kefelegn S (2018) Prediction and analysis of liver disorder diseases by using data mining technique: survey. Int J Pure Appl Math 118(9):765–770 2. Nagaraj K, Sridhar A (2014) NeuroSVM: a graphical user interface for identification of liver patients. Int J Comput Sci Inf Technol 5(6):8280–8284 3. Pakhale H, Xaxa DK (2016) A survey on diagnosis of liver disease classification. Int J Eng Tech 2(3) 4. Kiran Kumar M, Sreedevi M, Padmanabha Reddy YCA (2018) Survey on machine learning algorithms for liver disease diagnosis and prediction. Int J Eng Technol 7(1.8):99–102 5. Jatav S, Sharma V (2018) An algorithm for predictive data mining approach in medical diagnosis. Int J Comput Sci Inf Technol 10(1) 6. K Swapna, Prasad Babu MS et al (2017) Critical analysis of Indian liver patients dataset using ANOVA method. Int J Eng Technol 17(03) 7. Hashem EM, Mabrouk MS (2014) A study of support vector machine algorithm for liver disease diagnosis. Am J Intell Syst 4(1):9–14 8. Nancy P, Sudha V, Akiladevi R (2017) Analysis of feature selection and classification algorithms on hepatitis data. Int J Adv Res Comput Eng Technol 6(1). ISSN: 2278 – 1323 9. Vijayarani S, Dhayanand S (2015) Liver disease prediction using SVM and Naïve Bayes algorithms. Int J Sci Eng Technol Res 4(4) 10. Jacob J, Mathew JC, Mathew J, Issac E (2018) Diagnosis of liver disease using machine learning techniques. Int Res J Eng Technol 05(04) 11. Banu Priya M, Laura Juliet P, Tamilselvi PR (2018) Performance analysis of liver disease prediction using machine learning algorithms. Int Res J Eng Technol 05(01) 12. Vaidya H, Chaudhari SK, Ingale HT (2017) Literature review on liver disease classification. Int J Adv Res Innovative Ideas Educ 3(3) 13. Friedman LS, Keeffe EB, Dienstag JL. Handbook of liver disease 14. Liver function testing in primary care (2007) Developed by bpac, George St Dunedin, New Zealand. www.bpac.org.nz 15. Vett-Joice C (2012) Capital pathology 16. Thapa BR, Walia A (2007) Liver function tests and their interpretation. Indian J Pediatr 74 (July)

A Study of Routing Protocols for Energy Conservation in Manets Aqsa Parveen(&) and Y. V. S. Sai Pragathi Department of Computer Science and Engineering, Stanley College of Engineering and Technology for Women, Abids, Hyderabad 500001, India [email protected], [email protected]

Abstract. Today, we all are living in the technological era and in the few past years we have seen the speedy increase in the usage of mobile devices that work without base station, moreover, all these devices are movable that means they are not fixed. The mobiles devices can be cell phones, PDA, Laptop, Tablet PCs etc. All the wireless devices communicate with neighboring node through the multiple hops as they are infrastructure-less network All the mobile nodes are battery oriented and energy is a limited resource in MANETs. One of the considerable issue in mobile ad hoc network (MANETs) communication is energy consumption. To overcome this dispute many algorithms has been proposed that preserved the energy of the devices as far as possible without affecting the performance and other characteristics of the network. Keywords: Manets


Energy consumption
AODV
AOMDV

1 Introduction The design goal of MANETs is to provide access to the information anytime characterized by the lack of infrastructure and no show of base station, for this goal to be achieved we required an adequate routing protocol. In multi-hop wireless networking ad hoc wireless network is the new paradigm that is rapidly become popular and become an necessary part of the networking environment, comprising of infra-structure and infra-structure less mobile network MANETs are rapidly increasing in wireless networking because of its self organizing and self configuring properties. They are convenient for applications involves in special outdoor events, communication in areas with no wireless infrastructure, emergency, natural disasters, military operations, mine site operations, urgent business meetings and robust data acquisition (Fig. 1). Advantages of MANETs: Accessibility: MANET provides access to the information and services irrespective of the geographic position. Deployment: The network can establish within no time and at any location. Infrastructure-less: The network works without any pre-existing base station and allow people and devices to work in areas with no supporting infrastructure. Dynamic: Can dynamically arrange into arbitrary and temporary network topologies. © Springer Nature Switzerland AG 2020 S. C. Satapathy et al. (Eds.): ICETE 2019, LAIS 4, pp. 641–647, 2020. https://doi.org/10.1007/978-3-030-24318-0_73

642

A. Parveen and Y. V. S. Sai Pragathi

Fig. 1. Structure of MANETs

The goal is to design a reliable and efficient routing strategy that can consume a minimum amount of battery to provide the services and access to the information. Basically, there are two kind of routing protocol (Fig. 2).

Fig. 2. Types of routing protocol in MANETs

Proactive Routing Protocols: Proactive routing protocol stores the information of the routing and updates the information periodically by exchanging the control packets with their neighbors. This kind of routing algorithm make use of Link-State routing algorithm that floods link information about its neighbors. Proactive protocols are also referred to as table driven protocols because they need to maintain node information for every node present in the routing table.

A Study of Routing Protocols for Energy Conservation in Manets

643

Reactive Routing Protocols: In this type protocols the route is established from source to destination only when, it is needed. Nodes launch route discovery process on demand basis. In this manner it reduces the overhead that is present in the proactive routing protocols. This kind of protocols uses distance-vector routing algorithm for the discovery of the route. Reactive protocols composed of: (1) Route Discovery (2) Route Maintenance. Hybrid Routing Protocols: Hybrid Protocols combines the features of both proactive and reactive routing protocols. It can achieve by storing routes of nearby nodes and finding the routes that are far away nodes using a route discovery phase.

2 Literature Survey Ravi and Kashwan [1] proposed an algorithm known as Energy Aware Span Routing Protocol (EASRP) that conserve the energy equivalent to the fidelity Energy Conservation Algorithm (FECA). In this algorithm packet delivery ratio is improved and it also saves the energy of the entire network. Wang suggested [2] an energy management model. In this model each node in the network can be in two of the modes, i.e., Active Mode (AM) and Power Save Mode (PS). In an active mode, a node is always active and can receive data at any time irrespective of the location. In power save mode of energy model management the node is sleeping and roll out periodically to check the pending messages. Any mobile node within the network can switch from an active mode to power save mode or vice versa whenever packet arrives and keep alive timer gets expires. Divya, Subasree, Sakthivel et al. [3] investigated the performance of Efficient Routing Protocol in MANETs that shows the evolution only in mobile computing field. They also find that performance of the network degrades if the size of the network increases. Rajaram and Sugesh [4] provides a power aware ad hoc on demand multipath distance vector routing protocol for preserving the energy of node in the network. In PAAOMDV (power aware ad hoc on demand multipath distance vector) instead of route cache which is present in the every on-demand protocols each node will maintain Energy Reservation Technique (ERT). ERT contains the following entries source id, destination id, request id, amount of energy reserved and route. In PAAOMDV source node starts sending the packets that containing the information to the destination after the amount of energy just consumed from the energy reserved. Whenever the node finds fault in transferring the packets it will generate Error Packet (RERR) and send that packet back to it (source node) the node that receives the RERR packet will use expiration time out is used to switch to the alternate path. Fatima et al. [5] proposed Reactive Congestion Multipath Routing Protocol (RCRP). They have transformed AOMDV RREQ packet with delay and energy parameters. Every node have to maintain a table with an entry of Energy Reduction Rate (ERR). They also assume the threshold value of ERR that is termed as TERR when the node value of ERR reach to the greater than TERR then that node is avoided

644

A. Parveen and Y. V. S. Sai Pragathi

to be participate in the route discovery process. The RREQ packet contain an additional field known as DEDR (Delay Energy Drain Rate) compare to AMODV and again DEDR contain two subfields known as ERR and pdt. Before transferring the packets the value of DEDR is set to 0 and it will get incremented as it proceeds through the intermediate nodes, The destination will select the route based on less value of DEDR as primary route and the route with more DEDR value will be set as secondary path. Vidwans, Shrivastava et al. [6]. They have improved the performance of AMODV protocol in QOS with the help of queue length of AOMDV protocol and called this protocol as Enhanced AOMDV (EAOMDV). It is one of the easy implementation in real MANETs but this protocol is inefficient to implement. Suresh et al. [7] proposed an Efficient Power Aware Routing Protocol (EPAR), that enhanced the lifetime of the network in MANETs. EPAR identifies the capacity of the node by its residual energy and by the expected energy spent in reliably forwarding data packets through the specific route by using min–max formulation. EPAR choose the path that has the largest packet capacity at the smallest residual packet transmission capacity. This protocol handles high mobility of the nodes in the MANETs that change the network topology. This protocol reduces 20% the total energy consumption. Siddiqui, Afroz [8] They provided Intelligent Systems Design, Minimum Delay Routing Protocol with Enhanced Multimedia Transmission over MANETs. They analyzed various delays of packet transmission along with mathematical assumption and condition subjected to certain assumption. Their proposed work is suitable for multimedia transmission over parallel links between two approaches with different data rates. They have considered end-to-end delay for calculation of the path with maximum throughput that is sum of nodal delays. This model is inefficient as there is a scope for improvement in queue delay and energy of the node. Pooja et al. [9] proposed an Enhancement of Multipath Routing protocol for Route Discovery (EMPRR) to provide multipath discovery. This protocol used sufficient amount of bandwidth. The EMPRR increases delivery ratio of the packet and reduced the end-to-end delay but the problem is the data confidentiality in multi hop delivery application is not centralized control. Dhanalakshmi et al. [10] proposed Optimized link State Routing Protocol (OLSR). In this protocol energy cost is calculated by forecasting the energy consumption level of the node using ECAO model. Initially network formation was carried out and node location is updated along with the neighbors table and the distance from the source to the neighbor node is also calculated. Based on the routing table update for high energy cost and updating of neighborhood table low value of energy consumption route discovery is performd and destination is check if the destination is checked successfully multipath analysis is carried out otherwise route discovery is performed again, optimal path is selected based on Modified Dijikstra’s algorithm after that failure of the link is checked if there is any failure link, neighbor table is updated otherwise, processing of the data is carried out.

A Study of Routing Protocols for Energy Conservation in Manets

645

3 Comparison of Energy Consumption Methods S. no Authors 1 Ravi and Kashwan [1]

2

3 4

5

6

7

8

9

10

Merits Improves packet delivery ratio and saves the energy of the entire network Wang [2] Energy management Achieve higher performance and model: Nodes can be in two extend the life-span of network with energy modes: (1) Active mode (AM) control (2) Power Save mode (PS) Divya et al. [3] Efficient routing Preserves the energy of protocol in MANETs the node Rajaram et al. Power aware ad hoc Saves the energy of the [4] on demand multipath network distance vector (PAAOMDV) Fatima et al. Reactive congestion This protocol avoids the node to participate [5] multipath routing in the transferring of protocol (RCRP) the packets which is having less amount of energy Vidwans et al. Enhanced AOMDV Easy implementation [6] (EAOMDV) in real MANETs Suresh et al. [7]

Algorithm Energy aware span routing protocol

Power aware routing protocol (EPAR)

This protocol reduces 20% the total energy consumption and achieved good packet delivery ratio Siddiqui et al. Minimum delay For multimedia [8] routing protocol with transmission over enhanced multimedia parallel links between two approaches with transmission over different data rates MANETs Increased the packet Pooja et al. [9] Enhancement of delivery ratio and multipath routing reduced the end-to-end protocol for route discovery (EMPRR) delay Dhanalakshmi Optimized link state Achieved better PDR et al. [10] routing protocol and end-to-end delay (OLSR)

Demerits Suitable for small networks

Switching the nodes from AM to PS is one of the overhead

Not suitable for large network Route cache is not available

DEDR field increased the processing overhead

This protocol is not efficiently implemented Processing overhead and network traffic is increased because of min–max formulation No improvement in queue delay and energy of the node

Data confidentiality in multi hop delivery application is not centralized control Security measures issues are noticed

646

A. Parveen and Y. V. S. Sai Pragathi

4 Proposed System In our approach we are going to alter the AODV protocol RREQ packet by adding energy, delay and throughput parameters, where every node will have table with energy reduction rate we are going to have one extra value which is called as threshold value (TV) wherever ERR value reach to TV value than that node will be avoided to take participate in the route discovery phase. Every RREQ packet will have an extra field which will be refer to as DEDR. Whenever the source node broadcast RREQ packet DEDR will be set to 0 and it will be incremented by 1 by passing through intermediate nodes. The path which is having less DEDR value will be set as primary path and the next less value of DEDR will be set as secondary path.

5 Conclusion This survey paper addresses the different kind of protocol and they modification that are necessary for conserving energy of the nodes in MANETs network, the performance of every protocol depends on the parameters of the network. From the above discussion we would like to conclude that energy is one of the crucial resources in MANETs. There are many techniques in order to preserves the energy. By preserving the energy, network performance can be increased but there is no protocol that can provide the overall performance without comprising other parameters.

References 1. Ravi G, Kashwan KR (2015) A new routing protocol for energy efficient mobile applications for ad hoc networks. J Comput Electr Eng 48(C):77–85 2. Wang Y (2010) Study on energy conservation in MANET. J Netw 5(6):708–771 3. Divya M, Subasree S, Sakthivel NK (2015) Performance analysis of efficient energy routing protocols in MANET. Proc Comput Sci 57:890–897 4. Rajaram A, Sugesh J (2011) Power aware routing for MANET using on-demand multipath routing protocol. Int J Comput Sci Issues 8(4, no 2):517–522 5. Fatima A, Parveen A, Fatima R, Raziuddin S, Improving the quality of service based on reactive congestion control protocol for multipath routing in MANETS. Int J Adv Res Eng Manag 18–24. ISSN: 2456-2033 6. Vidwans A, Shrivastava AK, Manoria M (2014) QoS enhancement of AOMDV routing protocol using queue length improvement. In: 2014 fourth international conference on communication systems and network technologies. https://doi.org/10.1109/csnt.2014.60 7. Suresh HN, Varaprasad G, Jayanthi G (2014) Designing energy routing protocol with power consumption optimization in MANET. IEEE Trans Emerg Top Comput 2.2:192–197 8. Siddiqui KAA, Afroz YK (2016), Minimum delay routing protocol with enhanced multimedia transmission over heterogeneous MANETs. Int J Comput Appl 139(5) 9. Pooja AD (2013) Enhancement of multipath routing protocol for route recovery in MANET. Eur Sci J 9:270–281

A Study of Routing Protocols for Energy Conservation in Manets

647

10. Dhanalakshmi N, Alli P (2015) Efficient energy conservation in MANET using energy conserving advanced optimised link state routing model. Int J Parallel Emergent Distrib Syst 31(5):469–480 11. Kuri J, Kasera SK (2001) Reliable multicast in multi-access wireless LANs. Wireless Netw 7:359–369 12. Haseeb Zafar DH, Andonovic I, Hasan L, Khattak A (2012) QoS-aware multipath routing scheme for mobile ad hoc networks. Int J Commun Netw Inf Secur 4:1–10 13. Chowdhury T, Mukta RBM (2014) A novel approach to find the complete node- disjoint multipath in AODV. In: 3rd international conference on informatics, electronics & vision 2014 14. Surjeet, Parkash A, Tripathi R (2013) QoS bandwidth estimation scheme for delay sensitive applications in MANETs. Commun Netw 5(1):8 15. Rao M, Rao M, Singh N, Surajmal M (2014) Quality of service enhancement in MANETs with an efficient routing algorithm. In: 2014 IEEE international advance computing conference (IACC), IEEE. 978-1-4799-2572-8/14/$31.00_c 16. Jain SA, Bande A, Deshmukh G, Rade Y, Sandhanshiv M (2012) An improvement in congestion control using multipath routing in MANET. Int J Eng Res Appl 2:509–514 17. Sarkar S, Datta R (2012) A trust based protocol for energy-efficient routing in self-organized MANETs. In: Annual IEEE india conference (INDICON), Kochi, pp 1084–1089 18. Sana AB, Iqbal F (2015) Quality of Service Routing for MultiPath Manets. In: Proceedings of the international conference on signal processing and communication engineering systems (SPACES), IEEE. ISBN:978-1-4799-6108-5

Credit Risk Valuation Using an Efficient Machine Learning Algorithm Ramya Sri Kovvuri(&) and Ramesh Cheripelli Department of IT, G. Narayanamma Institute of Technology and Science, Shaikpet, Hyderabad, Telangana, India [email protected], [email protected]

Abstract. The automation process helps in improving the efficiency of the detection process, and it may also provide higher detection accuracy by removing the internal subjective human factors in the process. If machine learning can automatically identify bad customers, it will provide considerable benefits to the banking and financial system. The goal is to calculate the credit score and categorize customers into good or bad. Algorithms of machine learning library is used to classify the data sets of finance sectors. A large volume of multi structured customer data is generated. When the quality of this data is incomplete the exactness of study is reduced. In the proposed system, we provide machine learning algorithms for effective prediction of various occurrences in societies. We experiment the altered estimate models over real-life bank data collected. Compared to several typical estimate algorithms, the calculation exactness of our proposed algorithm is high. Keywords: Machine learning
Credit scoring Random forest
CRISP DM Framework


Logistic Regression


1 Introduction Hundreds of banks in the United States alone suffer from non-payment or late-payment of loans [1]. Predicting such customers earlier facilitates preventive banking interventions, which in turn can lead to enormous cost savings and improved outcomes [2]. Algorithms are developed for predicting customer behavior by drawing from ideas and techniques in the field of machine learning [3]. A problematic information assortment mechanism is intended and therefore the correlation analysis of this collected knowledge is performed [4]. A stochastic prediction model is designed to foresee the future condition of the most correlated customers based on their current account status [5]. In banking and finance communities, a large volume of multi structured customer data is generated from the transactions, account statements and online purchases.

2 Data Understanding Two data sets are required for the analysis, Demographic data and Credit bureau data. Demographic Data: Demographic data has simple variables. © Springer Nature Switzerland AG 2020 S. C. Satapathy et al. (Eds.): ICETE 2019, LAIS 4, pp. 648–657, 2020. https://doi.org/10.1007/978-3-030-24318-0_74

Credit Risk Valuation Using an Efficient Machine Learning Algorithm

649

Credit Bureau data: Credit bureau data has variables obtained from previous history of the customer. Both datasets are provided by the bank (Tables 1 and 2, Fig 1). Table 1. Demographic data. Variables Application ID Age Gender Marital status No of dependents Income Education Profession Type of residence No of months in current residence No of months in current or any company Performance tag

Description Unique ID of the customers Age of customer Gender of customer Marital status of customer (at the time of application) No. of children’s of customers Income of customers Education of customers Profession of customers Type of residence of customers No of months in current residence of customers No of months in current company of customers Status of customer performance (“1” represents “Default”)

Table 2. Credit bureau data. Application ID No of times 90 DPD or worse in last 6 months No of times 60 DPD or worse in last 6 months No of times 30 DPD or worse in last 6 months No of times 90 DPD or worse in last 12 months No of times 60 DPD or worse in last 12 months No of times 30 DPD or worse in last 12 months Average CC utilization in last 12 months No of trades opened in last 6 months No of trades opened in last 12 months No of PL trades opened in last 6 months

Customer application ID Number of times customer has not payed dues since 90 days in last 6 months Number of times customer has not payed dues since 60 days last 6 months Number of times customer has not payed dues since 30 days last 6 months Number of times customer has not payed dues since 90 days last 12 months Number of times customer has not payed dues since 60 days last 12 months Number of times customer has not payed dues since 30 days last 12 months Average utilization of credit card by customer Number of times the customer has done the trades in last 6 months Number of times the customer has done the trades in last 12 months No of PL trades in last 6 month of customer (continued)

650

R. S. Kovvuri and R. Cheripelli Table 2. (continued)

Application ID No of PL trades opened in last 12 months No of inquiries in last 6 months (excluding home and auto loans) No of inquiries in last 12 months (excluding home and auto loans Presence of open home loan

Customer application ID No of PL trades in last 12 month of customer Number of times the customers has inquired in last 6 months Number of times the customers has inquired in last 12 months Is the customer has home loan (1 represents “Yes”) Outstanding balance Outstanding balance of customer Total no of trades Number of times the customer has done total trades Presence of open auto loan Is the customer has auto loan (1 represents “Yes”) Performance tag Status of customer performance (“ 1 represents “Default”) Data contain a variable performance tag which represents whether the applicant has gone default after getting a credit card. Data is having some records where the performance tag is not present. These records are considered as rejected. After keeping aside rejected records there are 69,867 records remaining. Among these 4% of the records are default. Also, company doesn’t know whether rejected are also contain right customers or not.

Fig. 1. Percentage of non-default and default customers.

3 Data Cleaning and Exploratory Data Analysis Preliminary checks like checking structure, summary of data have been done. Checked for duplicates in data and removed 3 duplicates with same App.ID. Merged Demographic and Credit Bureau data. Missing value treatment is taken care by the WOE analysis which is done further. Outlier treatment has been done for variables Age, Income, No.of.months.in.current.company etc.,. Below charts are an example (Figs. 2 and 3).

Credit Risk Valuation Using an Efficient Machine Learning Algorithm

651

Fig. 2. Data cleaning graph.

Fig. 3. Data cleaning graphs.

EDA has been done on all the variables by deriving a variable called Default Rate (Fig. 4).

652

R. S. Kovvuri and R. Cheripelli

Fig. 4. Plots with all variables of credit bureau data.

Credit Risk Valuation Using an Efficient Machine Learning Algorithm

653

4 Weight of Evidence Analysis WOE analysis on the data has been performed and replaced demographic and credit data with WOE values. Sample plot on demographic data as follows. Similar way for credit data also been done (Fig. 5).

Fig. 5. Plots with all variables of credit bureau data.

654

R. S. Kovvuri and R. Cheripelli

5 Model Building Based on the analysis of data, we tried building models using Logistic Regression, Decision Trees and Random Forests and pick the model which is best for this data (Table 3). Table 3. Result of analysis using different models on data. Model Logistic regression Decision trees

Random forests

Logistic regression Decision trees

Random forests

Data on which model was built Demographic data

Accuracy 53.54

Sensitivity 60.4

Specificity 53.24

Demographic data overbalancing Demographic data - under balancing Demographic data - both Demographic data - balancing with ROSE Demographic data overbalancing Demographic data - under balancing Demographic data - both Demographic data - balancing with ROSE Whole data Whole data - balanced Whole data - overbalancing Whole data - under balancing Whole data - both Whole data - balancing with ROSE Whole data - without balancing Whole data - overbalancing Whole data - under balancing Whole data - both Whole data - balancing with ROSE

52.6

60

59.7

56.6

55.6

55.7

52.6 61.42

60 49.97

59.7 50.46

51.4

56.22

51.18

52.8

53.1

52.8

52 55

54.4 53.5

51.8 55.06

67.49 63.5 50.79 59.9 50.79 73.92

58.71 63.8 76.01 67.3 76.01 47.96

67.87 63.5 49.67 59.57 49.67 75.06

64.5 55.22 61.74 62.2 63.4

57.35 62.33 61.99 57.8 64.06

64.82 54.9 61.72 62.39 63.41

Credit Risk Valuation Using an Efficient Machine Learning Algorithm

655

6 Model Evaluation From the above metrics of all the different models one model must be chosen which is consistent across all the three metrics i.e., Accuracy, Sensitivity and Specificity. Although some models gave 70+ accuracy, they perform poor in Sensitivity. Finally left with Logistic Regression and Random Forest models which has equal numbers for all the three parameters. Chosen Random Forests because of two reasons: Sensitivity is slightly more compared to logistic regression, and as we know Random Forests will perform good on unseen data. So Random Forest with balanced data is our Final Model. 6.1

Important Variables from the Model

From the model which is built only on demographic data, below are the important variables. No.of.months.in.current.residence, Income, and No.of.months.in.current.company. From the final model chosen i.e., Random forest, Important variables can derived from the below plot (Fig. 6).

Fig. 6. Variables plot.

7 Application Score Card 7.1

Implications of Using the Model

From the Model and built scorecard the cut-off score is set to be 355.2808. By applying this cutoff on rejected candidates there are 256 candidates rejected out of 1425 whose score is high. By using the model, the rejected population have been decreased thereby increasing the revenue for the company. Also using the model company can avoid manual process in approving the credit cards. From the predicted model application scorecard has been built. The summary of score card vs log odds as follows (Fig. 7).

656

R. S. Kovvuri and R. Cheripelli

Fig. 7. Score card vs log odds.

8 Conclusion A random forest model based multimodal credit score prediction algorithm using structured data from banking data set. Analyzing the factors based on the banking sector data. Missing values problem can be resolved using machine learning algorithms. Default customer prediction can be done based on the data and type of, region and risk level of the customer’s account status by the availability of the data. Imagine a system where banks can quickly go through millions of anonymized customer records to find people with good credit scores and bank experiences. Through this massive, searchable database, banks could determine whom to offer a loan, based on what has worked effectively for others with similar behavior and characteristics. Acknowledgments. I would like to express my special thanks of gratitude to my guide Ch. Ramesh, Assistant Professor in G. Narayanamma Institute of Technology and Science, as well as our head of the department Information technology Dr I. Ravi Prakash Reddy in G. Narayanamma Institute of Technology and Science, who gave me the golden opportunity to do this wonderful project, which also helped me in doing a lot of Research and I came to know about so many new things I am really thankful to them. Secondly I would also like to thank my parents and friends who helped me a lot in finalizing this paper within the limited time frame.

Credit Risk Valuation Using an Efficient Machine Learning Algorithm

657

References 1. Peng Y, Xu R, Zhao H, Zhou Z, Wu N, Yang Y (2017) Random walk based trade reference computation for personal credit scoring. In: 2017 IEEE 13th international symposium on autonomous decentralized system (ISADS) 2. Li W (2011) An empirical study on credit scoring model for credit card by using data mining technology. In: 2011 seventh international conference on computational intelligence and security 3. Rushin G, Stancil C, Sun M, Adams S, Beling P (2017) Horse race analysis in credit card fraud—deep learning, logistic regression, and gradient boosted tree. In: 2017 systems and information engineering design symposium (SIEDS) 4. Liu Y, Du J, Wang F (2013) Non-negative matrix factorization with sparseness constraints for credit risk assessment. In: Proceedings of 2013 IEEE international conference on grey systems and intelligent services (GSIS) 5. Kraus A (2014) Recent methods from statistics and machine learning for credit scoring

Handwritten Mathematical Symbol Recognition Using Machine Learning Techniques: Review Syeda Aliya Firdaus(&) and K. Vaidehi Department of Computer Science and Engineering, Stanley College of Engineering and Technology for Women, Abids, Hyderabad 500 001, India [email protected], [email protected]

Abstract. Handwritten character/symbol recognition is an important area of research in the present digital world. The problems such as recognition of handwritten characters/symbols, which may be written in different styles when it is recognized can makes job of the human easier. Mathematical expression recognition using machines has become a subject of serious research. The main motivation for this review work is both recognizing of the handwritten mathematical symbol, digits and characters which will be used for mathematical expression recognition. Keywords: Mathematical symbol recognition
Character segmentation Character recognition
Mathematical expression recognition




1 Introduction Handwriting recognition (HWR) [1] is a computer process performed to obtain and understand the handwritten input such as touch-screen, paper documents, photographs and other devices. The images of the written text papers is called as “off line” taken by optical image scanning (or intelligent word recognition). The motion of the pen tip felt generally on a pen-based computer screen surface can be called as “on line”, a easier task as there are more options available. Handwriting recognition primarily follows the process of optical character recognition (OCR). The handwriting recognition system handles and includes formatting of the document, performs the correct segmentation into characters and also finds the most possible words. OCR [2] can be both mechanical or electronic converter. The conversion includes conversion of handwritten image, typed image or printed text into machine-encoded text, taken from a photograph of a document, a scanned document, a scene-photo from subtitle text superimposed on an image. OCR is usually an “offline” [3] process that static document. Handwritten moments are taken as an input to the handwritten recognition system and the input data is the static representation of the handwriting. OCR machines are primarily uses machine printed text and ICR (capital letters)for hand “printed text. The shapes of glyphs and words makes motion capturing easy when © Springer Nature Switzerland AG 2020 S. C. Satapathy et al. (Eds.): ICETE 2019, LAIS 4, pp. 658–671, 2020. https://doi.org/10.1007/978-3-030-24318-0_75

Handwritten Mathematical Symbol Recognition Using Machine Learning Techniques

659

taken as input to the technique. The motions captured are the order of drawing the segments, the direction, and the sequence in which the pen is put down and lifted up. With the help of this additional information the accuracy of an end-to-end process can be increased. This technique can also called the “intelligent character recognition”, “on-line character recognition”, “dynamic or real-time character recognition”. On-line handwriting character recognition [4] takes the input from the special digitizer or PDA. The sensor picks up both the pen-tip movements and pen-up/pen-down switching i.e. lifting and putting down of pen. The data collected by the use of this method is called as digital ink. The ink can be considered as a digital representation of handwriting. The signals are converted to the letter codes and can be used in text-processing applications in the computers. Early versions of character recognition needs to be trained with all images of each character, and has to be operated on one font at a time. But the advanced systems which are used today are capable of achieving a high recognition accuracy for most fonts that are commonly used, and is completed with help of various digital image file format inputs. Some of the systems are also capable of providing outputs of the formatted pages which are approximately same as the original page including images, columns, and other components (non-text). Humans can easily recognize the handwritten document but the recognition of the same by the computer system becomes difficult for it due to the present of random variations in the noise in image, writing size, fonts and styles. OCR is a field of research in artificial intelligence, pattern recognition and computer vision. It is broadly used as a method of entering the information from printed paper or a data records such as computerised receipts, invoices, bank statements, business cards, printouts of static-data, mails or any suitable documentation or a passport documents. Digitising printed texts is done by this process for making them electronically edited, stored more compactly, displayed on-line, searched, and can be used in machine processes such as, machine translation, cognitive computing, text-to-speech and data mining.

2 Methodology Handwritten character or symbol recognition is one of the applications in the pattern classification. Figure 1 shows the block diagram of character recognition.

I/P

pre-processing

Segmentation

Character recognized

Feature Extraction

Classification

Fig. 1. Block diagram for character recognition

660

S. A. Firdaus and K. Vaidehi

3 Different Handwritten Math Symbol Databases Handwritten Math Symbol Dataset This dataset consists of jpg files (45  45), dataset does not contain Hebrew alphabet at all. It consists of basic Greek alphabet symbols included are: alpha, beta, gamma, mu, sigma, phi and theta. English alphanumeric symbols are included. All math operators, set operators. Basic pre-defined math functions like: log, lim, cos, sin, [5] tan. Math symbols included are like: \int, \sum, \sqrt, \delta and more (Fig. 2).

(a)

(b)

(c)

(d)

(e)

Fig. 2. Sample images from the dataset (a): exclamation mark (b): bracket (c): square-root (d): theta (e): lambda

MNIST MNIST [6] consists of 60,000 handwritten digits images (64  64  3) in the training dataset. It also consists of 10,000 images in the test set in gray-scale (28  28  1). The dataset is easily available for research purpose and is free of cost. HWRT Database The dataset consists of 11,081 images. The HWRT database consists of handwritten symbols containing symbols such as all alphanumeric characters, Greek characters, arrows and as well as mathematical symbols like the integral symbol. It is also available easily in jpg format. The size of the jpg images is 156  231. InftyCDB-1 InftyCDB-1 consists of 30 papers in English language which includes of mathematical calculations. It consists of 688,580 alphanumeric character image in 476 pages text document that are recorded along with the character code of the symbol it represents. The links which represent the structure of each word or a mathematical expression are also recorded. InftyCDB-1 [7] can be used as a character database, word database as well as a mathematical expression database. In the InftyCBD-1 database the total number of words images are 108,914 and the total number of images for mathematical expression are 21,056. HASYv2 - Handwritten Symbol Database HASY [8] contains 369 symbol classes images (32  32). HASY consists of 150,000 instances of handwritten symbols. HASY is a easily available and is free of charge dataset of single symbols similar to MNIST. It contains 168233 instance images of 369 classes.

Handwritten Mathematical Symbol Recognition Using Machine Learning Techniques

661

4 Literature Survey Kumar et al. [9] proposed “Analytical review of pre-processing techniques for offline handwritten character recognition”, it explains Fourier transform method for measuring the shape and pattern to obtain relevant information. The method can easily extract features and classify them. The classification is done using feed forward back propagation neural network. Binarization, character recognition, noise removal, normalization are the preprocessing techniques included along with segmentation. Page segmentation, character segmentation techniques are the segmentation methods used in this paper. Local and global features are extracted from the processed image. The NN classifier is used for classification. Shi et al. [10] had proposed “Symbol Graph Based Discriminative Training and Rescoring for Improved Math Symbol Recognition”. The discriminative training of the exponential weights, the insertion of penalty and graph rescoring are included in the work. In this paper, symbol graph theory of training the exponential weights of each model and inserting penalties is followed. The training is considered for two different areas: maximum mutual information (MMI) and minimum symbol error (MSE). In the post-processing (which is done after training step) trigram-based graph rescoring is performed. The database contains 2500 formulas. The accuracy of 97% for symbol recognition is achieved. Kasthuri et al. [11] proposed “Noise Reduction and Pre-processing techniques in Handwritten Character Recognition using Neural Networks”. The Gabor filtering and noise reduction are the two pre-processing techniques used in this work. The aberrations and non-uniformities are considered as the noise in the process of character recognition. To overcome these issues, it is necessary to perform noise reduction. The process of recognition uses statistical analysis to match between the generated pattern and reference pattern. The dataset used contains 7291 combination of handwritten characters (2 languages) 2549 printed characters. The accuracy achieved is 97%. Jubair et al. [12] proposed “A simplified method for handwritten character recognition from document image”. It uses morphological thinning operation as a segmentation technique in this work. The classifier used is KNN. Data-base contains 780 sample images of characters written by different people having different handwritten styles. The accuracy achieved is 95.688% Hu et al. [13] proposed “HMM-Based Recognition of Online Handwritten Mathematical Symbols Using Segmental K-means Initialization and a Modified Penup/down Feature”. The work proposed in this paper is recognition work. Hidden Markov Model (HMM) based recognition system is used for recognizing the isolated online handwritten mathematical symbols. Left to right continuous HMM is designed for each symbol class. The symbol recognition includes two steps: symbol segmentation and isolated symbol recognition. K-means produces better initialization using parameters of Gaussian Mixture Models. The pen-up/down gives less accuracy compared to normalized distance to stroke edge features. The database used in this work consists of 22483 sample images. The recognition accuracy obtained for top-1 dataset is 82.9% and top-5 dataset is 97.8%.

662

S. A. Firdaus and K. Vaidehi

Tian et al. [14] proposed “Research on Symbol Recognition for Mathematical Expressions” for recognition of handwritten math symbol in a mathematical expression. Symbol segmentation and symbol recognition are the two phases of the system. The feedback mechanism is proposed for segmentation and recognition of symbols. Symbol segmentation uses the projective features and connected components labelling method for segmentation of an expressions. The directional line element features and peripheral features are extracted from the input symbols. A coarse-to-fine classification strategy is used to recognize symbols with these features. The accuracy achieved is 97.81%. Wang et al. [15] proposed “The Effectiveness of Data Augmentation in Image Classification using Deep Learning”. Normalization is done in pre-processing step. The dataset used is MNIST dataset. Neural net is trained for augmentation and classification. The process is called neural augmentation. Content loss, Style loss via gram matrix and No loss at the end layer are the three augmentation techniques applied in this paper. CNN is used for classification. Keysers et al. [16] proposed “Multi-Language Online Handwriting Recognition”. Resampling and slope correction are the two pre-processing techniques used in this work. The datasets used are UNIPEN-1, IAM-OnDB and self datsets. The ink is preprocessed. A set of character rules are performed to create a segmentation lattice. Firstly a set of overcomplete potential cut points between the characters are determined using a segmenter. Using the cut points we group set of ink segments into a character hypotheses, this process creates a segmentation lattice. The labelling of the segmentation lattice is accomplished with the help of a classifier by classification of character hypothesis and additional feature functions. The classifier used are HMM and LSTM. The accuracy achieved using UNIPEN-1and IAM-OnDB are 97% and 96%. Davila et al. [17] had proposed “Using Off-line Features and Synthetic Data for Online Handwritten Math Symbol Recognition” on-line recognition system to recognize handwritten math symbols that uses off-line features and synthetic data generation. Global features, crossing feature, 2d fuzzy histogram points, fuzzy histogram orientations are the features extracted. Four different classifiers are used AdaBoost C4.5, Random Forest, SVM linear kernel, SVM RBF kernel. Data-base used in this work have two recognition rates top1 and top5. The top-1 and top-5 acquires accuracy of different percentages using the different databases, for AdaBoost C4.5 are 88.4% & 98.7%, for Random Forest are 87.9% 98.4%, for SVM linear kernel are 88.6% 99.1% and for SVM RBF kernel are 89.8% 99.1% respectively. Rahiman et al. [18] proposed “Recognition of Handwritten Malayalam Characters using Vertical & Horizontal Line Positional Analyzer Algorithm”. Segmentation is done by Line & Character separation. The feature extraction technique used is Horizontal and Vertical Line count and positions. Classifier used is decision Tree for classification. The acquires accuracy is 91%. Sinha et al. [19] proposed “Zone-Based Feature Extraction Techniques and SVM for Handwritten Gurmukhi Character Recognition”. Zone Centroid Zone method provide better recognition accuracy than Image Centroid Zone. The first attribute according to the writer’s variations in writing style, size, shape, ink colour, ink flow and thickness, digitization imperfections etc. The deficiencies that are fond in the particular method are second attribute for feature extraction. The Gurumukhi dataset of 7000 Gurmukhi character sample images is used in this work. Feature extraction

Handwritten Mathematical Symbol Recognition Using Machine Learning Techniques

663

techniques used are Image centroid zone for line segmentation, Zone centroid zone for word segmentation and Hybrid centroid zone for character segmentation. SVM is used for classifier. The accuracy achieved is 95.11%. Kumawat et al. [20] propose “New Approach of Hand writing Recognition using Curvelet Transform and Invariant Statistical Features”. A character recognition system is used for character recognition. The method combines invariant statistical features and curvelet transform. The combined features are used by HMM and SVM to classify the character based on the curvelet transform and invariant statistical features. A 200 samples of two users are taken and the accuracy achieved is 98.92%. Nicolas et al. [21] proposed “Recognition of Handwritten Mathematical Symbols with PHOG Features”. The HOG features are generalized to pyramids of HOG features (PHOG). The classifier used is SVM. The CHROME dataset containing 22000 character sample images gets an accuracy of 96%. And 75 handwritten sample images by a different user gets 92% accuracy. Pradeep et al. [22] proposed “Diagonal Feature Extraction Based Handwritten Character System Using Neural Network”. Diagonal features are used for feature extraction and Feed Forward Back propagation Neural Network for classification. Vertical, diagonal and horizontal directions are the different feature attributes used for extracting 54 features from each input character. The different inputs are tested on the Neural network and it performs well. The highest accuracy 98% is obtained by using diagonal orientation for feature extraction this work. Das et al. [23] propose offline English character recognition model in “HMM based Offline Handwritten Writer Independent English Character Recognition using Global and Local Feature Extraction”, to combine the global and local features for classifying the character using hidden Markov model. Data-base contains 13000 samples images of characters written in five different styles for each character collected from 100 writers. The proposed system acquires 98.26% accuracy of 98.26%. Pirlo et al. [24] proposed “Adaptive Membership Functions for Handwritten Character Recognition by Voronoi-Based Image Zoning”. The handwritten character recognition system that uses static and dynamic zoning topologies is proposed in this work. The segmentation technique used in this work is Optical image segmentation. This is the technique used for feature extraction and can also be known as Voronoi tessellation. Malon et al. [25] had proposed “Mathematical symbol recognition with support vector machines” presents a method for the improving classification process. The SVM is used for classification. Multi-class classification by SVM is done by the system utilizing the ranking of alternatives within InftyReader’s confusion clusters. Misrecognition rate is reduced by 41% overall. 70,637 samples are taken in the database. Recognition rate of the system is 96.10% obtained without using SVM. The recognition accuracy achieved using SVM by this method is 97.70%. Aradhyal et al. [26] proposed “Robust Unconstrained Handwritten Digit Recognition using Radon Transform”. A system is proposed for handwritten digit recognition based on random transform and nearest neighbour algorithm. Two MNIST datasets are used in this work. The first one is of English digit image samples which acquired accuracy of 96.6%, the second one is of Kannada numerals and achieved 91.2%.

664

S. A. Firdaus and K. Vaidehi

Dai Nguyen et al. [27] proposed “Deep Neural Networks for Recognizing Online Handwritten Mathematical Symbols”. The pre-processing technique used is normalization and gradient directional features are extracted. The dataset used is collected from the CHROME which consists of MEs databases containing 120,341 symbol images. Max-out-based CNN is applied to image patterns generated by online patterns. BLSTM is applied to image patterns generated by the original online patterns. Max-out based CNN and BLSTM are also used to combine all the patterns together. Comparing them by MRF and MQDF in a traditional recognition method by experimenting on CROHME database. The classifiers used are CNN and deep max out CNN. The accuracy achieved using BLSTM is 97.61%. Ratnamala et al. [28] proposed “A novel method for handwritten mathematical document based on equation symbols recognition using K-NN and A-NN classifiers”. The dataset used is self-math document. Filtering and ROI selection is used for preprocessing. Features extracted are statistical and LBP features. The two classifiers KNN and ANN are used for classification. Both the KNN and ANN uses statistical and LBP features for classification of the data into Math or Non-math. The accuracy we achieved by KNN is 96% and ANN is 97%. Kulkarni [29] proposed “Handwritten Character Recognition Using HOG, COM by OpenCV & Python”. The dataset used the HASY dataset which contains 168233 instances of 369 classes. Gaussian Blurring and Canny Edge Detector are used for data preparation. Segmentation is done through Otsu’s thresholding. HOG features are extracted by de-skewing the images converted to centre of the images. Accuracy of the descriptor is increased power law compression & Square root of the input image. Linear SVM is the classifier used to train the dataset for recognition in the final step. Accuracy achieved by this HCR is 96.56%. Darmatasia et al. [30] proposed “Handwriting Recognition on Form Document Using Convolutional Neural Network and Support Vector Machines (CNN-SVM)”. The dataset used is NIST Special Database 19. The features are extracted by CNN using feature maps. CNN is used as a feature extracted and is constructed using CNNSVM toolbox. A linear SVM combined with CNN is used for classification by L1 loss function and L2 regularization. The accuracy achieved for CNN+SVM is 91.37% while with the original CNN the accuracy achieved is 88.32%.

5 Comparison of Related Works (See Tables 1 and 2)

Handwritten Mathematical Symbol Recognition Using Machine Learning Techniques

665

Table 1. Comparision between related work for handwritten symbol recognition S. Author Title no.

Preprocessing

1.

[9]

Analytical review of preprocessing techniques for offline handwritten character recognition Symbol graph based discriminative training and rescoring for improved math symbol recognition

2.

[10]

3.

[11]

4.

[12]

5.

[13]

6.

[14]

7.

[15]

8.

[16]

Multi-Language Online Handwriting Recognition

9.

[17]

Binarization, Using off-line Normalization features and synthetic data for on- and edge detection line handwritten math symbol recognition

Binarization, Noise removal, Character Recognition, Normalization Maximum mutual information (MMI) and minimum symbol error (MSE) Noise Reduction and Gabor filtering and noise Pre-processing reduction techniques in Handwritten Character Recognition using Neural Networks A simplified method Normalization for handwritten and character recognition binarization from document image HMM-based Normalization, recognition of online edge detection handwritten and mathematical binarization symbols using segmental k-means initialization and a modified penup/down feature Research on symbol Normalization binarization recognition for and erosion mathematical expressions The effectiveness of Normalization data augmentation in threshold thinning image classification using deep learning Resampling, slope correction

Segmentation

Feature extraction

Classification Accuracy

Page segmentation, character segmentation

Local and global features

NN

–

MMI and MSE Weighting

Trigrambased graph rescoring

97%

Resizing

NN

97

KNN

95.688

Fileting

Morphological thinning operation

Symbol segmentation

Gaussian HMM mixture model, pen up/ pen down features

83%, 97%

Feedback mechanism

Projective features

Corseclassification

97.81

Neural augumentation Content loss, Style loss via gram matrix Character hypotheses, Segmentation Lattice Character segmentation

Gram matrix, mean, variance

CNN

91.5%

Character segmentation

HMMLSTM

97%, 96%

Global features, crossing feature, 2d fuzzy histogram points, fuzzy histogram orientations

AdaBoost C4.5, Random Forest, SVM linear, RBF kernel

98.7% by AdaBootop-5 is highest

(continued)

666

S. A. Firdaus and K. Vaidehi Table 1. (continued)

S. Author Title no.

Preprocessing

10. [18]

11. [19]

12. [20]

13. [21]

14. [22]

15. [23]

16. [24]

17. [25]

18. [26]

Recognition of handwritten Malayalam characters using vertical & horizontal line positional analyzer algorithm Zone-Based Feature Extraction Techniques and SVM for Handwritten Gurmukhi Character Recognition New approach of hand writing recognition using curvelet transform and invariant statistical features Recognition of mathematical symbol using PHOG features Diagonal feature extraction based handwritten character system using neural network HMM based offline handwritten writer independent english character recognition Adaptive membership functions for handwritten character recognition by voronoi-based image zoning Mathematical symbol recognition with support vector machines Robust unconstrained handwritten digit recognition using radon transform

Segmentation

Feature extraction

Classification Accuracy

Normalization Line, character slant correction separation noise removal

Vertical & horizontal line positional analyzer algorithm

Tree classifier

91%

Normalization, binarization and edge detection

Word, Line, character segmentation

Image centroid zone

SVM

95.11%

Normalization noise removal binarization

Character segmentation

Curvelet transform & invariant statistical features

Combined HMM and SVM

98.92%

Binarization thinning edge detection Normalization noise removal binarization

Symbol character segmentation Character segmentation

PHOG features SVM

Binarization Edge detection thining noise removal Noise removal Normalization

Character segmentation

Optical image segmentation

Binarization Edge detection

Symbol segmentation

SVM kernel and confusion cluster

SVM kernel

97.07%

Normalization, binarization

Character segmentation

Local. Global features

Random forest

96.6%,91.2%

Horizontal, vertical, diagonal features

NN

96%

98%

Global and HMM 98.26% location features using HMM Optimal zoning Zoning based – classifier topology by Voronoi tessellation

(continued)

Handwritten Mathematical Symbol Recognition Using Machine Learning Techniques

667

Table 1. (continued) S. Author Title no.

Preprocessing

Segmentation

Feature extraction

Classification Accuracy

19. [27]

Normalization noise removal

Character segmentation

Gradient directional features

CNN, maxout-CNN, BLSTM

97.61% by BLSTM

Filtering thinning

ROI

Statistical, LBP KNN, ANN features

96%, 97%

Gaussian Blurring and Canny Edge Detector

Otsu’s thresholding

HOG feature by deskewing

Linear SVM

96.56%

Binarization, Normalization

Character segmentation

Featuring Mapping

CNN + SVM 91.37%, 88.32%

20. [28]

21. [29]

22. [30]

Deep neural networks for recognizing online handwritten mathematical symbols A novel method for handwritten mathematical document based on equation symbols recognition using KNN and A-NN classifiers Handwritten Character Recognition Using HOG, COM by OpenCV & Python Handwriting recognition on form document using convolutional neural network and support vector machines (CNN-SVM)

Table 2. Comparision between merits and demerits of the related work S. Author Title no.

Database

Merits

Demerits

1.

[9]

Self

Feature extraction step uses the processed image

Need to use all the preprocessing steps to achieve good accuracy

2.

[10]

Self

Trigram rescoring gets the highest symbol recognition rate

3.

[11]

2 self dataset

Multiple algorithms is beneficial for character recognition

Discriminative training and trigram based graph rescoring is done in postprocessing steps Language detection is a costlier process and the accuracy decreases when the quality of the input drops

4.

[12]

Self

Less complex, easily implemented and gives high accuracy

Analytical review of preprocessing techniques for offline handwritten character recognition Symbol graph based discriminative training and rescoring for improved math symbol recognition Noise Reduction and Preprocessing techniques in Handwritten Character Recognition using Neural Networks A simplified method for handwritten character recognition from document image

Cell value calculation is a complex part

(continued)

668

S. A. Firdaus and K. Vaidehi Table 2. (continued)

S. Author Title no.

Database

Merits

Demerits

5.

[13]

Top-1, top5

Good accuracy is obtained

Requires lot of time and mental effort

6.

[14]

7.

[15]

HMM-based recognition of online handwritten mathematical symbols using segmental k-means initialization Research on symbol recognition for mathematical expressions The effectiveness of data augmentation in image classification using deep learning

Selfsymbol dataset MNIST

Feedback mechanism for segmentation and classification Augmentation techniques for ImageNet dataset gives high accuracy

8.

[16]

Multi-Language Online Handwriting Recognition

9.

[17]

UNIPEN, IAMOnDB & self Top-1, top5

Using off-line features and synthetic data for on-line handwritten math symbol recognition Recognition of handwritten Malayalam Malayalam characters using characters vertical & horizontal line positional analyzer algorithm

10. [18]

11. [19]

12. [20]

13. [21]

14. [22]

15. [23]

16. [24]

Zone-Based Feature Extraction Techniques and SVM for Handwritten Gurmukhi Character Recognition New approach of hand writing recognition using curvelet transform and invariant statistical features Recognition of mathematical symbol using PHOG features

Gumukhi characters

Self

CHROME

OCR achieves high accuracy using text but is low for mathematical expressions GAN’s and neural augmentation performance is low and takes 3 times more than the traditional augmentation technique The architecture framework Re-use of components across is flexible various languages and scripts can be a problem Synthetic data generation for Ambiguous classes leads to low global accuracy underrated classes can improve average per class accuracy Identifies both colored Due to the similarity in characters and characters character shapes and with colored background character features in Malayalam language the system gives less accuracy Zone centroid zone and Recognition rate depends on Image centroid zone SVM parameters combination improves the accuracy HMM and SVM kernel are combined to get high efficiency

PHOG feature extraction techniques along with one against one SVM classifier achieves good accuracy Diagonal feature extraction postal Diagonal feature extraction based handwritten character address techniques perform better system using neural network images than the conventional horizontal and vertical feature extraction technique Self English HMM for some specific HMM based offline characters that have wide character handwritten writer range of variation dataset independent english character recognition Adaptive membership CEDAR The segmentation is functions for handwritten and ETL automatic optimal character recognition by segmentation and is proven voronoi-based image zoning an efficient way

Feature vectors affects the performance of the system

Complex expressions are difficult to recognize

Faces difficulty in using different classifier

Low recognition rate for other datasets

No. of zones should be specified as a priori

(continued)

Handwritten Mathematical Symbol Recognition Using Machine Learning Techniques

669

Table 2. (continued) S. Author Title no.

Database

Merits

Demerits

17. [25]

InftyReader

SVM with Inftyreader’s confusion cluster decreases the misrecognition by 14% Efficient and robust randon transform method is used in this paper

Touching and broken characters treatment is required Range selection is unclear for optimal recognition accuracy

In offline method MQDF helps to get wide and specific features whereas the BLSTM can access whole document in online method Initially identifies math symbols and classifies as math and non-math

In the case of mathematical expression the method is difficult to work.

18. [26]

Mathematical symbol recognition with support vector machines Robust unconstrained handwritten digit recognition using radon transform

MNIST English, kannada numeral CHROME

19. [27]

Deep neural networks for recognizing online handwritten mathematical symbols

20. [28]

Self- math a novel method for document handwritten mathematical document based on equation symbols recognition using K-NN and A-NN classifiers HASY Handwritten Character Recognition Using HOG, COM by OpenCV & Python

21. [29]

22. [30]

HOG descriptor using edge detection and normalization helps in extracting the features from images of different styles, size and backgrounds Handwriting recognition on NIST Ten folds Cross-validation form document using special and document forms convolutional neural network database 19 containing boundary box and and support vector machines little noise recognition is the (CNN-SVM) work done

The overall accuracy achieved is less

Improper segmentation leads to unambiguous features in feature extraction step

Connected character is the problem which is difficult to solve

6 Conclusion Handwritten document recognition is a complex task to numerous writing styles for distinct person writing styles. The system first identifies the required segment in a handwritten document of characters for segmentation and features are extracted from the segmented character. Characters are recognized from the extracted features. The paper includes the introduction and review on mathematical handwritten character recognition. A literature Survey for pre-processing, segmentation, feature extraction and classification techniques that are effective and efficient for mathematical symbol recognition is briefly explained. The comparison of different papers based on handwritten math symbol recognition is done vividly.

References 1. MacLean S, Labahn G (2015) A Bayesian model for recognizing handwritten mathematical expressions. Pattern Recogn 48(8):2433–2445 2. Singh D, Khan MA, Bansal A, Bansal N (2015) An application of SVM in character recognition with chain code. In: Communication, control and intelligent systems (CCIS). IEEE, pp 167–171

670

S. A. Firdaus and K. Vaidehi

3. Bharambe M (2015) Recognition of offline handwritten mathematical expressions. Structure 34636:97–98 4. Mouchere H, Viard-Gaudin C, Kim DH, Kim JH, Garain, U (2011) Crohme 2011: competition on recognition of online handwritten mathematical expressions. In: 2011 document analysis and recognition (ICDAR). IEEE, pp 1497–1500 5. https://www.kaggle.com/xainano/handwrittenmathsymbols 6. The mnist database of handwritten digits. http://yann.lecun.com/exdb/mnist. Accessed 19 May 2017 7. http://www.inftyproject.org/en/database.html 8. https://zenodo.org/record/259444#.XEStfVwzZPY 9. Kumar G, Bhatia PK, Banger I (2013) Analytical review of preprocessing techniques for offline handwritten character recognition. Int J Adv Eng Sci 3(3):14–22 10. Luo ZX, Shi Y, Soong FK (2008) Symbol graph based discriminative training and rescoring for improved math symbol recognition. In: 2008 IEEE international conference on acoustics, speech and signal processing, ICASSP 2008. IEEE, pp 1953–1956 11. Kasthuri M, Shanthi V (2014) Noise reduction and pre-processing techniques in handwritten character recognition using neural networks. Technia 6(2):940 12. Jubair MI, Banik P (2012) A simplified method for handwritten character recognition from document image. Int J Comput Appl 51(14) 13. Hu L, Zanibbi R (2011) HMM-based recognition of online handwritten mathematical symbols using segmental k-means initialization and a modified pen-up/down feature. In: 2011 international conference on document analysis and recognition (ICDAR). IEEE, pp 457–462 14. Tian XD, Li HY, Li XF, Zhang LP (2006) Research on symbol recognition for mathematical expressions. In: 2006 first international conference on innovative computing, information and control, ICICIC 2006. IEEE, vol 3, pp 357–360 15. Perez L, Wang J (2017) The effectiveness of data augmentation in image classification using deep learning. arXiv preprint arXiv:1712.04621 16. Keysers D, Deselaers T, Rowley HA, Wang LL, Carbune V (2017) Multi-language online handwriting recognition. IEEE Trans Pattern Anal Mach Intell 39(6):1180–1194 17. Davila K, Ludi S, Zanibbi, R. (2014) Using off-line features and synthetic data for on-line handwritten math symbol recognition. In: 2014 frontiers in handwriting recognition (ICFHR). IEEE, pp 323–328 18. Rahiman MA, Rajasree MS, Masha N, Rema M, Meenakshi R, Kumar GM (2011) Recognition of handwritten Malayalam characters using vertical & horizontal line positional analyzer algorithm. In: 2011 3rd international conference on electronics computer technology (ICECT). IEEE, vol 2, pp 268–274 19. Sinha G, Rani A, Dhir R, Rani MR (2012) Zone-based feature extraction techniques and SVM for handwritten gurmukhi character recognition. Int J Adv Res Comput Sci Softw Eng 2(6) 20. Kumawat P, Khatri A, Nagaria B (2013) New approach of hand writing recognition using curvelet transform and invariant statistical features. Int J Comput Appl 61(18) 21. Jimenez ND, Nguyen L. Recognition of Handwritten Mathematical Symbols with PHOG 22. Pradeep J, Srinivasan E, Himavathi S (2010) Diagonal feature extraction based handwritten character system using neural network. Int J Comput Appl 8(9):17–22 23. Das RL, Prasad BK, Sanyal G (2012) HMM based offline handwritten writer independent english character recognition using global and local feature extraction. Int J Comput Appl 46 (10):45–50

Handwritten Mathematical Symbol Recognition Using Machine Learning Techniques

671

24. Pirlo G, Impedovo D (2012) Adaptive membership functions for handwritten character recognition by voronoi-based image zoning. IEEE Trans Image Process 21(9):3827–3837 25. Malon C, Uchida S, Suzuki M (2008) Mathematical symbol recognition with support vector machines. Pattern Recogn Lett 29(9):1326–1332 26. Aradhya VM, Kumar GH, Noushath S (2007) Robust unconstrained handwritten digit recognition using radon transform. In: 2007 signal processing, communications and networking, ICSCN 2007. IEEE, pp 626–629 27. Dai Nguyen H, Le AD, Nakagawa M (2015) Deep neural networks for recognizing online handwritten mathematical symbols. In: 2015 3rd IAPR Asian conference on pattern recognition (ACPR). IEEE, pp 121–125 28. Ratnamala SP (2016) Shilpa proposed a novel method for handwritten mathematical document based on equation symbols recognition using K-NN and A-NN classifiers. International Journal of Emerging Technology in Computer Science & Electronics (IJETCSE) 23(6). (SPECIAL ISSUE), ISSN 0976-1353 29. Kulkarni RL (2017) Handwritten character recognition using HOG, COM by OpenCV & Python. Int J 5(4) 30. Fanany MI (2017) Handwriting recognition on form document using convolutional neural network and support vector machines (CNN-SVM). In: 2017 information and communication technology (ICoIC 2017). IEEE, pp 1–6

Automatic Water Level Detection Using IoT Ch. V. S. S. Mahalakshmi(&), B. Mridula, and D. Shravani Stanley College of Engineering & Technology for Women, Hyderabad, India [email protected], [email protected], [email protected]

Abstract. This paper implements the design of a water level sensor device called the Aqua Buzzer working under the guidelines of IoT that is able to detect the level of water, when placed in low lying areas or apartments. First, the system senses the level of water by the level detector part comprising the ultrasonic sensor. This sensor works in accordance with the buzzer connected to the Aurdino UNO Board. If the water level rises, the buzzer beeps continuously, alerting people and sends alert messages to the nearby Rescue teams by network. The device is capable of detecting level of rising water continuously and that data is stored in cloud for further references. Keywords: Aqua Buzzer
Water level
Ultrasonic sensor Arduino UNO
Alert
Messages
Data
Cloud


Buzzer


1 Introduction There is excessive water usage either for commercial or domestic works, which is a serious issue, affecting the sustainability of our environment. Shortage or scarcity of water may be due to the climatic changes, like alteration in weather patterns such as drought, increasing pollution and increasing population demand of over usage of water. As water is one of the most essential natural resources, it is important and likely to use it in an efficient way. There is an need to monitor the usage of water across different sectors. From the last decade, there has begun a lot of study and research to conserve natural resources which are water, energy, coal and etc. Energy and water conservation techniques and technology improvements can aid to attain sustainable solutions to our environment that is currently at risk due to excessive use of natural resources due to increase in population, human demand and economic growth. According to United Nations (UN) report, almost half of the world’s workers work in water related sectors showing most of the jobs dependent on water [1]. There is a risk for them as the water’s nature can go unpredictable sometimes. Hence with the invention of new technologies like Internet of Things can be helpful. An electricity consumption campus audit was conducted and reducing energy consumption mechanisms were suggested in [2, 3]. Automatic water pump controlling systems which can be used in irrigation and light intensity controller for energy conservation were designed in [5–7]. Watering of agricultural fields depending on the soil moisture level to avoid water wastage due to improper irrigation system is specifically considered in [4]. Internet of Things (IoT) is an ecosystem of connected physical objects that are accessible through the internet [8]. © Springer Nature Switzerland AG 2020 S. C. Satapathy et al. (Eds.): ICETE 2019, LAIS 4, pp. 672–677, 2020. https://doi.org/10.1007/978-3-030-24318-0_76

Automatic Water Level Detection Using IoT

673

It is also considered as a collection of objects that work together whose main purpose is to serve consumer tasks in a federated manner [9, 10]. In order to deliver data about the surrounding environment, it binds computational data [11, 12]. The word object here represents a person with a heart monitor or an automobile with built-in-sensors, or assigned IP addresess and have the ability to collect and transfer data over a network without human intervention. The objects interact with one another through the embedded technology in them, thus affecting the decisions taken. We can monitor the water level in low lying areas. One of the rapid influences of IoT is on the field of environmental monitoring, especially on disaster management, early warning systems as well as environmental data analytics. Here in this paper, I propose a water level detecting system, which can detect the water level in low lying areas (here apartments is considered as prototype). The device has sensors (ultrasonic sensor) merged in the container and is placed at certain level from water level. The buzzer is connected to the sensor. After a certain decrease in the distance between the sensor and water, the sensor sends signals to the buzzer and the buzzer beeps continuously. This integrated device sends message to the nearby rescue teams through Blynk interface, acting as cloud storage and network propagator.

2 Overview The proposed system works in two phases during hazards namely detecting and alarming. In the detecting phase the sensor is placed at certain level from the ground. The sensor continuously senses the water level. There is a pre-measured distance between the sensor and water level. After a certain acceptable distance called threshold distance if there is any decrease in the distance, the sensor sends signals to the Arduino Board (intermediate) which in turn sends the signals to the buzzer that keeps beeping continuously, thereby alerting people nearby. This happens during the alarming phase. The device sends signals (alert messages) across network to the Rescue teams (Fig. 1).

Fig. 1. Block diagram

674

Ch. V. S. S. Mahalakshmi et al.

3 Methodology 3.1

Water Level Sensor Unit

This unit comprises of the ultrasonic sensor, the Arduino board and the related code. 3.1.1 Ultrasonic Sensor The ultra sonic sensor is used to calculate distance between itself and nearby object. It regularly transmits a small amount of sonic sound to the target and reflects the sound back to the sensor. The circuit then calculates the time for the sound to reach the target and to return to the sensor, thereby calculating the distance between them. Output from the sensor is of variable length, which is the distance between the sensor and obstacle [13]. The sensor is connected digital pin 10–14 and Vcc (Fig. 2).

Fig. 2. Ultrasonic sensor

3.1.2 Arduino UNO Arduino UNO, a microcontroller board is based on AVR microcontroller called Atmega328. The board is made up of 14 digital pins and 6 analog pins. The board allows an easy access to the input and output pins [14]. This Arduino board is connected to any Ethernet of choice. It is a part of raspberry pi, a minicomputer (Fig. 3).

Fig. 3. Arduino Uno Board with ultrasonic sensor

Automatic Water Level Detection Using IoT

675

3.1.3 Code Code comprises of instructions to the ultrasonic sensor which are sent through the Arduino Board. The code is dumped into the memory section of the board for fast access. This section also includes “the message (notification)” that is to be given to the Rescue teams. 3.2

Buzzer

Buzzer is used as an alarm and is connected to analog pin1. It receives signals from the sensor through the board. It receives signals only when the distance between the water and sensor decreases. Then it starts to beep, hereby acting as an alarm indicating danger. 3.3

Blynk Interface (App)

Blynk is an application interface that is responsible for communication between the circuit (water level sensing unit) and electronic gadgets (smartphones, tablets including computers). The signals sent by the sensor are captured by the Blynk interface and these signals (in the form of messages) are transmitted to the Rescue teams, wherein the interface is provided to the teams prior to the setup of device and is to be in active state for reception of signals and then the data is stored in the cloud, which is again provided by Blynk.

4 Implementation and Result In this paper, the implemented distance between the sensor and initial water level is 8 cm and the threshold distance (certain distance which is acceptable) is 6 cm. If the distance between sensor and water is beyond the threshold, the buzzer beeps and immediately signals are sent on network (Figs. 4 and 5).

Fig. 4. A working model of Aqua Buzzer

676

Ch. V. S. S. Mahalakshmi et al.

Fig. 5. Indication of rising water levels

The above figure depicts the outcome. It is a screen indicating the difference between the sensor and the water level. Only 4 cm is displayed continuously on the screen because of the loop in the program, that reads the distance. Here the distance is maintained to be 4 cm for testing purpose. In Fig. 6 X-axis represents the sensor distance from water which is constant and Yaxis represents the water level which is increasing, thereby the reduction in distance between sensor and water indicated by orange line. The blue line represents the threshold (6 cm) which is acceptable level of difference.

Fig. 6. Graph showing constant senor distance with increasing water level

5 Conclusion In this paper, an IoT applied device is designed to monitor and report the level of water in low lying areas or a water detection system in a reservoir. The device is designed to automatically monitor and report water levels from zero to eight where threshold being an acceptable level. The proposed system eliminates human monitoring and reporting for any industrial, or any other purpose. The system works on IoT. The need for data storage about water levels is also considered and is stored in cloud and the information

Automatic Water Level Detection Using IoT

677

about the rising water levels is reported to nearby Rescue teams across network. Thus by automation, this system achieves a proper water level detection helping people to be aware of the upcoming danger.

References 1. Water and Jobs, the United Nations, World Water Development Report 2016. http://unesdoc. unesco.org/images/0024/002439/243938e.pdf 2. Getu BN, Attia HA (2016) Electricity audit and reduction of consumption: campus case study. Int J Appl Eng Res (IJAER) 11(6):4423–4427 3. Attia HA, Getu BN (2016) Authorized timer for reduction of electricity consumption and energy saving in classrooms. Int J Appl Eng Res (IJAER) 11(15):8436–8441 4. Getu BN, Attia HA (2015) Automatic control of agricultural pumps based on soil moisture sensing. In: Proceedings of the IEEE AFRICON Conference on 2015, 14–17 September 2015, pp 667–671 5. Getu BN, Hamad NA, Attia HA (2015) Remote controlling of an agricultural pump system based on the dual tone multifrequency (DTMF) technique. J. Eng Sci Technol (JESTECH) 10(10):1261–1274 6. Getu BN, Attia HA (2015) Remote controlling of light intensity using phone devices. Res J Appl Sci Eng Technol (RJASET) 10(10):1206–1215 7. Attia HA, Getu BN (2015) Design and simulation of remotely power controller. Int J Appl Eng Res (IJAER) 10(12):32609–32626 8. https://www.happiestminds.com/Insights/internet-of-things/ 9. Jie Y, Pei JY, Jun L, Yun G, Wei X (2013) Smart home system based on IOT technologies. In: 2013 fifth international conference on computational and information sciences (ICCIS), 21–23 June 2013, pp. 1789–1791 10. How the Next Evolution of the Internet is Changing Everything. https://www.cisco.com/ web/about/ac79/docs/innov/IoT_IBSG_0411FINAL.pdf 11. Perumal T, Sulaiman MN, Mustapha N, Shahi A, Thinaharan R (2014) Proactive architecture for Internet of Things (IoTs) management in smart homes. In: 2014 IEEE 3rd global conference on consumer electronics (GCCE), 7–10 October 2014, pp 16–17 12. Perumal T, Sulaiman MN, Leong CY (2013) ECA-based interoperability framework for intelligent building. Autom Constr 31:274–280 13. https://www.elprocus.com/ultrasonic-detection-basics-application/ 14. https://www.theengineeringprojects.com/2018/06/introduction-to-arduino-uno.html

Optimal Scheduling of Tasks in Cloud Computing Using Hybrid Firefly-Genetic Algorithm Aravind Rajagopalan(&), Devesh R. Modale, and Radha Senthilkumar Department of Information Technology, Madras Institute of Technology Campus, Anna University, Chennai, India [email protected], [email protected], [email protected]

Abstract. Today cloud computing is an evolved form of utility computing which is widely used for commercial computing needs. The Cloud service provider’s success, profit, and efficiency lie in optimally allocating the computing resources to users from a vast pool of resources. The ability to allocate resources in a ubiquitous, seamless and on-demand connection involves serious challenges. Task scheduling is a variant of job-shop scheduling problem which is categorized as NP-COMPLETE. In this paper, a novel meta-heuristic algorithm of hybrid Firefly-Genetic combination is propounded for scheduling tasks. The proposed algorithm blends benefits of a mathematical optimization algorithm like Firefly with an evolutionary algorithm like Genetic algorithm to form a powerful metaheuristic search algorithm. The proposed hybrid Firefly-Genetic algorithm was able to schedule the tasks with the objective of minimal execution time for all tasks and a swift convergence to the near optimal solution. The proposed algorithm was tested in CloudSim which is a simulator toolkit for testing cloud-based algorithms. The experimental results showed that the proposed algorithm outweighed the performances of traditional First In First Out (FIFO) and Genetic algorithms. Keywords: Cloud computing
Task scheduling
Firefly optimization Genetic algorithm
FIFO
Evolutionary algorithms
Metaheuristic
Virtual machine




1 Introduction Cloud computing is an emerging computing paradigm where users are allocated with resources from a shared pool of resources according to pay as you go strategy. Cloud computing benefits mainly from the technology of virtualization, where the resources are allocated through Virtual Machine (VM) tools. There are many advantages of cloud computing like scalability, elasticity, inexpensive, no pre-investment, ubiquitous and on-demand self-service access, flexibility etc. The computing resources provided by cloud service provider are allocated to end users using task scheduling algorithms. Task

© Springer Nature Switzerland AG 2020 S. C. Satapathy et al. (Eds.): ICETE 2019, LAIS 4, pp. 678–687, 2020. https://doi.org/10.1007/978-3-030-24318-0_77

Optimal Scheduling of Tasks in Cloud Computing

679

scheduling algorithms aim to optimally allocate the resources to vast users as well as manage the load balancing. Since cloud computing is only a recent phenomenon, there is a wide scope of research pertained to the domain of scheduling the tasks in cloud computing. Task scheduling problem has been dealt with many heuristics presented in the recent past by different authors. This paper proposes a hybrid Firefly-Genetic heuristic algorithm to optimally allocate resources and schedule the task in cloud computing. Genetic Algorithm belongs to the class of evolutionary algorithm. It is inspired by the process of natural selection and the theory of evolution. The major advantages of genetic algorithms include its ability to tackle noisy or stochastic objective fitness functions, global search ability, ability to different kinds of encoding of solution set or chromosomes etc. The Genetic algorithm is mostly preferred for problems with multiple local optima values. Firefly algorithm was first introduced by Yang and was influenced by the behavior of fireflies. Yang and He observations about the advantages of Firefly algorithm were its ability of automatic subdivision of the problem and ability to deal with modality constraints [1]. These two advantages combined together made exploration and exploitation of search spaces very balanced and thus resulted in the production of the best results. Thus the Genetic and Firefly algorithms both individually prove to be powerful metaheuristic algorithms and their integration to a single combined hybrid algorithm can outperform both of them individually. This paper utilizes the above fact and introduces a hybrid Firefly-Genetic algorithm with the objective of optimally allocating resources to minimize the total execution time of the tasks in the cloud environment.

2 Literature Survey Ismail and Barua summarized results of scheduling applications with divisible loads in the cloud with the objective of reducing the execution time of tasks [2]. Their real-time applications were tested using the Amazon web services environment. Abadi discussed scheduling and deploying data management models on Amazon web services [3]. He inferred that various tasks like data analysis tasks, decision support systems, and data marts performed better in the cloud environment when compared to the traditional database management systems. Calheiros et al. put forward a simulation toolkit for modeling algorithms for cloud systems called CloudSim [4]. The CloudSim toolkit supports modeling entities like data centers, virtual machines, and resource provisioning policies. Liang et al. proposed bandwidth-aware task-scheduling (BATS) algorithm for divisible task scheduling under bounder multiport criteria [5]. They evaluated their algorithm with CloudSim simulator and compared the BATS algorithm with the fair-based task-scheduling algorithm and found that BATS had better performance. Feng et al. proposed a Particle Swarm Optimisation (PSO) algorithm which is based on small position value rule to schedule tasks in the cloud computing environment [6]. They compared their PSO algorithm with PSO algorithm embedded with crossover and mutation operators and found that their PSO algorithm converged faster than the other two algorithms. Bitam proposed a Bees Life Algorithm

680

A. Rajagopalan et al.

(BLA) as a variant of Bee swarm optimization algorithm for scheduling jobs [7]. BLA was compared with the Genetic algorithm are results were illustrated. Verma and Kaushal discussed Deadline and Budget distribution based Cost-Time Optimization (DBD-CTO) algorithm for scheduling with a dual goal of achieving minimal execution cost and managing deadlines [8]. Xue et al. proposed Load balancing algorithm built on ant colony optimization algorithm (ACO-LB) to solve load imbalance of VM in cloud computing [9]. Their algorithm was able to adapt to the dynamic cloud environment and was aimed to shorten the makespan of tasks. ACO-LB algorithm was simulated using CloudSim tool. Kumar and Verma integrated the Min-Min and Max-Min algorithms with the Genetic algorithm to form an improved Genetic Algorithm for task scheduling. They produced better results with their proposed algorithm than three heuristic algorithms taken individually [10]. The survey of various algorithms suggested that metaheuristic algorithms were most suited for scheduling related optimization problems. The survey helped to form a solid support background for proposing the hybrid Firefly-Genetic algorithm for task scheduling problem.

3 Hybrid Firefly Genetic Metaheuristic The proposed algorithm is a hybrid combination of Firefly Optimization and Genetic Algorithm. The hybridization goal is achieved by splitting the organization of algorithm into two phases, where first phase is accomplished by Firefly Optimization algorithm and the second phase is accomplished by the Genetic algorithm as stated in Fig. 1. The first phase involves the mapping the tasks to a population of fireflies. The fireflies are then optimally placed according to heuristic logic fetching a base set of results. The placing of fireflies is dependent on objective or fitness function that aims to reduce the execution cost of tasks. The second phase picks the final results from Firefly and those results are initialized as the base population of chromosomes for the Genetic algorithm. Since base results were already fine-tuned by firefly algorithm, the genetic algorithm looks for only superlative optimal solutions left from the point after the execution of the firefly algorithm. Thus, hybrid algorithm of Firefly Optimization and Genetic is capable of producing better results than each of the algorithm taken individually. 3.1

Firefly Algorithm Methodology

The firefly algorithm proposed by Yang has following three idealized rules: 1. All fireflies are attracted by each other without considering their sex. 2. The attractiveness of one firefly to the other varies directly with the brightness of the other firefly and attractiveness reduces with the increase in the distances between the fireflies. 3. The firefly will move randomly if there are no fireflies brighter than itself at that instant.

Optimal Scheduling of Tasks in Cloud Computing

681

Fig. 1. Proposed Firefly-Genetic algorithm architecture

Encoding Solutions. The solution or search space for optimization problem has limitless candidate solutions. Encoding of solutions in candidate search space enables to visualize results and helps in the further exploration. The resource in the cloud computing scenario is the Virtual machine (VM) and each VM is identified by its id or number. A vector of size m (total number of subtasks) named as Ti denotes the search space where the value at each index i gives the resource number allocated to the ith task represents a sample candidate encoded solution. Same encoding strategy is used for both firefly and genetic algorithm and hence the length of each firefly and the length of the chromosome is the same which is the total number of subtasks (m). For example, consider 3 jobs (k = 3) and 3 resources (n = 3). Each of the 3 jobs are broken to 3, 4, 3 subtasks as m1, m2 and m3, i.e. sub-task (1) = 3 (m1 = 3), sub-tasks (2) = 4 (m2 = 4), sub-tasks (3) = 3 (m3 = 3). So total number of subtasks is the sum of m1, m2, and m3 which is 9 (m = m1 + m2 + m3 = 9). So the length (l) of the chromosome is 9. One possible allocation of 3 resources to 9 subtasks is n1 = {1,3,5,9}, n2 = {2,4}, n3 = {6,7,8}. This allocation is encoded in the vector T as [1,2,1,2,1,3,3,3,1] which represents a sample firefly or a chromosome in a population. Each entry in T has values in the range of 1 − n. Let s be the total population of fireflies in the population. Similarly, if a population of 10 fireflies (s = 10) each with a job length of 9 is considered, a random valued population matrix of size 10 * 9. To summarize in a generic way, population matrix is of the form Xij where i ranges from [1, s], j ranges from [1, l] and each entry in Xij has values from [1, n]. Objective Function. The objective or fitness function is the function that defines and formulates the essential conditions that are needed to achieve optimization goal. In this problem domain, the optimization must result in reducing the total execution time of all the tasks in the cloud environment which results in the enhanced performance.

682

A. Rajagopalan et al.

Therefore, the objective function caters to minimize the total execution time of all subtasks. Execution time of a task depends upon two variable entities namely instruction execution speed as powered by the processor core of the VM and size or length of the instructions of each task. Each resource or VM has its own execution speed given in units of Million Instructions per second (MIPS) which is denoted in vector Ri where i 2 ½1; n. Each subtask has its length is represented in units of Million Instructions encoded in Li where i 2 [1, l]. The execution time for each task in seconds is obtained by the division of length of the task with the speed of the VM allocated to it. The execution time or fitness function for each firefly is then given by: fi ¼ Fi ¼

Xl i¼1

Li Ri

ð1Þ

1 fi

ð2Þ

Xl i¼1

The aim of the algorithm is to minimize the total execution time of all tasks which is obtained in (1). The Eq. (2) gives the inverse relation of execution time. The greater the execution time of tasks fi obtained by a particular firefly i, the lesser is the value obtained for F for that firefly. Hence the best solution is to find a firefly or chromosome with maximum fitness F, i.e. Most fit firefly ¼ maxi Fi

ð3Þ

Movement of Fireflies. The firefly movement is based on its attractiveness towards other the firefly. The firefly with higher attractiveness quotient attracts other fireflies towards itself. For measuring attractiveness, the brightness of the firefly has to be calculated. Brightness (I) is a direct measure of the result of the objective function to that firefly. Attractiveness ðbÞ between fireflies i and j are calculated as: bi ¼ Fi  ecr

2

ð4Þ

bj ¼ Fj  ecr

2

ð5Þ

Here c is the coefficient of light absorption which is a constant value, e is the exponential constant and r is the distance between the fireflies i and j. The distance is the number of noncorrelated elements between two firefly entries (Hamming distance). The movement aims to reduce the distance between two fireflies by adding dominant traits present in more attractive firefly to the firefly with weaker attractiveness quotient.

Optimal Scheduling of Tasks in Cloud Computing

683

Firefly Optimization Pseudo code

Begin Define objective function: F(x), x= (x1, x2,…,xl); Generate an initial population of fireflies Xij (i=1,2…m) (j=1,2,..n); Formulate light intensity I with relation to F(x) ; Define coefficient of light absorption Set t => 0 While (t Ii) Vary attractiveness (β) with distance e^(- r); Move firefly i towards j; Evaluate new solution and update I end if end for j end for i Rank fireflies and find the current best; end while end 3.2

Genetic Algorithm Methodology

After predefined iterations from Firefly algorithm, the best set of solutions formed from firefly procedure are made as initial population for the genetic algorithm. So the number of chromosomes is identical to the number of fireflies. The same encoding used for Firefly is replicated for genetic solution encoding as well. The three major genetic operators are selection, crossover, and mutation. The three operators together evolve the population. Fitness function for the genetic algorithm is the same as the objective function for the Firefly algorithm where the notation of firefly is interchanged by chromosome. Genetic algorithm offers the advantage of faster converge of solutions because the initial set of optimal solutions were already found using firefly algorithm. Selection. Selection is the process of selecting two parents that mate to produce an offspring. To produce a better and fitter offspring, it is crucial to select fitter parents. This paper chooses the roulette wheel selection operator Crossover. Of the two parents selected one will be stronger and the other will be weaker with respect to their fitness value. According to Darwin’s theory, only fittest individual survives. So, a weaker chromosome is decimated by a stronger chromosome by the crossover of genes. This paper uses a two-point crossover strategy.

684

A. Rajagopalan et al.

Mutation. The aim of the mutation is to produce random twitch of genes in the chromosome in order to introduce the diversity of the chromosomes. This paper uses the scramble mutation operator. From the final population of chromosomes after the required number of iterations, the best chromosome or most fit chromosome is chosen as the result and scheduling order is followed according to order encoded in the best chromosome. Genetic Algorithm Pseudo code Begin Same objective function :( same as firefly algorithm); Generate population of chromosomes (firefly result); Initialize crossover and mutation rate; Set iteration => 0 do Select 2 solutions Mutate the candidate solutions Perform crossover iteration => iteration+1 until a maximum iteration limit is reached return the solution with the best fitness end

4 Result Analysis The experiment to determine the efficiency of the Firefly-Genetic hybrid algorithm for task scheduling was simulated in CloudSim. The experiments can be categorized into 3 major divisions as proving that efficiency of the Firefly-Genetic hybrid algorithm increased over the increase in number of iterations, proving the efficiency of the FireflyGenetic hybrid algorithm over the popularly used First In First Out (FIFO) algorithm and finally proving the efficiency of the combined Firefly-Genetic hybrid algorithm over the Genetic algorithm individually. 4.1

Parameter Setting

To establish the effectiveness of the hybrid Firefly-Genetic metaheuristic algorithm was experimented under different cases. The Firefly and Genetic algorithms were initialized with parameters as summarized in Table 1.

Optimal Scheduling of Tasks in Cloud Computing

685

Table 1. Parameters initialization Algorithm Parameter Genetic Total population Crossover rate Mutation rate Firefly Total fireflies Alpha

4.2

Value 30 0.6 0.2 30 1

Experimental Result Analysis

In the first case, the experiment is carried in order to determine the relationship between iterations and execution time. The number of tasks was fixed at 20 for all iterations and population size of both fireflies and chromosomes were fixed at 30. The number of resources or VM’s was fixed as 5 for all cases. The rest of the global parameters remained the same as in Table 1. Figure 2 shows that as the number of epochs increases, the execution time decreases. The iterations were increased in factors of 10 starting from 10 iterations. For each case, the iteration value mentioned in fig corresponds to the same iteration value denominated for both Firefly and Genetic algorithm.

Fig. 2. Performance analysis of variation between execution time and iteration

In the second case, the experiment was aimed at establishing the supremacy of Firefly-genetic algorithm over the FIFO algorithm. The number of tasks was incremented in the factors of 20 starting from 20 and the total number of resources or VM’s was fixed at 5 for all the cases. Throughout the experiment, the iteration value was fixed at 30 for both the Firefly as well as the Genetic algorithms. Figure 3 summarises the result of the experiment. Clearly, as the number of tasks increases the FireflyGenetic algorithm had lower execution time than the FIFO algorithm.

686

A. Rajagopalan et al.

Fig. 3. Performance analysis of FIFO vs hybrid Firefly-Genetic algorithm

In the third case, the experiment was aimed to prove the power of Firefly-Genetic algorithm combined together, the experimental situation was similar to that of scenario 2. The number of tasks was incremented in the factors of 20 starting from 20 and the total number of resources or VM’s was fixed at 5 for all the cases. Iteration value was fixed at 30 for both the algorithms. Figure 4 summarizes the result. The hybrid Fireflygenetic algorithm had a lower execution time result for an increasing number of tasks proving that it is better than the Genetic algorithm.

Fig. 4. Performance analysis of Genetic vs hybrid Firefly-Genetic algorithm

5 Conclusion The ability to allocate very limited computing resources to a large number of tasks with an optimization goal has always inspired a wide variety of solutions in the cloud computing domain. This paper provides an approach of integration of two powerful heuristic search algorithms of Firefly and Genetic to form a combined metaheuristic. This paper tries to explore the novel hybrid Firefly-Genetic metaheuristic algorithm to the task

Optimal Scheduling of Tasks in Cloud Computing

687

scheduling problem in cloud computing with the objective of minimizing the execution time of all tasks to the near optimal solution. The results from simulation indicate that hybrid Firefly-Genetic algorithm has an effective and swift searchability in a dynamic cloud environment of vast search space over other popular algorithms like FIFO and genetic. Therefore, the hybrid Firefly-Genetic algorithm can be beneficial in the optimal allocation of resources and task scheduling in the cloud computing environment due to its ability to swiftly converge to a near approximate global optimum solution

References 1. Yang X-S, He X (2013) Firefly algorithm: recent advances and applications. Int J Swarm Intell 1:36–50 2. Ismail L, Barua R (2013) Implementation and performance evaluation of a distributed conjugate gradient method in a cloud computing environment. Softw Pract Experience 43 (3):281–304 3. Abadi DJ (2009) Data management in the cloud-limitations and opportunities. Bull IEEE Comput Soc Tech Committee Data Eng 32(1):3–12 4. Calheiros RN, Ranjan R, Beloglazov A (2011) Cloudsim – a toolkit for modeling and simulation of cloud computing environments and evaluation of resource provisioning algorithms. Softw Pract Experience 41(1):23–50 5. Liang C, Wang JZ, Buyya R (2014) Bandwidth-aware divisible task scheduling for cloud computing. Softw Pract Experience 44:163–174 6. Feng L, Zhang T, Jia Z, Xia X, Qin X (2013) Task schedule algorithm based on improved particle swarm under cloud computing environment. Comput Eng 39(5):183–186 7. Bitam S (2012) Bees life algorithm for job scheduling in cloud computing. In: International conference on computing and information technology (ICCIT), pp 186–191 8. Verma A, Kaushal S (2012) Deadline and budget distribution based cost-time optimization workflow scheduling algorithm for the cloud. In: IJCA proceedings on international conference on recent advances and future trends in information technology (iRAFIT 2012), pp 1–4 9. Xue S, Li M, Xu X, Chen J (2014) An ACO-LB algorithm for task scheduling in the cloud environment. J Softw 9:466–473 10. Kumar P, Verma A (2012) Scheduling using an improved genetic algorithm in cloud computing for independent tasks. In: Proceedings of the international conference on advances in computing, communications and informatics, pp 137–142 11. Dean J, Ghemawat S (2004) Mapreduce simplified data processing on large clusters. In: Sixth symposium on operating system design and implementation, San Francisco, CA, USA 12. Dean J, Ghemawant S (2008) MapReduce: simplified data processing on large clusters. Commun ACM 51(1):107–113 13. Zhang XH, Zhong ZY, Feng SZ, Tu BB, Fan JP (2011) Improving data locality of MapReduce by scheduling in homogeneous computing environments. In: IEEE 9th international symposium on parallel and distributed processing with applications, pp 120– 126. https://doi.org/10.1109/ispa.2011.14 14. Morton K, Balazinska M, Grossman D (2010) ParaTimer – a progress indicator for MapReduce DAGs. In: Proceedings of the 2010 international conference on management of data (SIGMOD 2010). ACM, New York, NY, USA, pp 507–518 15. Zhu Z, Du Z (2013) Improved GA-based task scheduling algorithm in cloud computing. Comput Eng Appl 49(5):77–80

A Novel Approach for Rice Yield Prediction in Andhra Pradesh Nagesh Vadaparthi(&), G. Surya Tejaswini, and N. B. S. Pallavi MVGR College of Engineering, Vizianagaram, India [email protected], [email protected], [email protected]

Abstract. The importance of yield prediction in Andhra Pradesh is increasing day to day life. As most of the people in Andhra Pradesh depend on agricultural income, it is essential for the farmer to know the demand for each crop in gain Minimum Support Price (MSP). It is understood from past experience that most of times either the yield goes high or falls down based on the price obtained in the previous season. Hence, in this paper an attempt is made to predict the crop yield especially Rice which is the major crop in Andhra Pradesh. In this work, we proposed Multiple Linear Regression Model. The experimental results clearly indicate that the methodology applied effective. Keywords: Minimum Support Price (MSP)
Average Rainfall (AR) Area Under Cultivation (AUC)
Multiple Linear Regression (MLR)




1 Introduction Agriculture is the prime source of income in India. About 54% of the income in India raises from agricultural sector and about 50%–60% of the population depends on agriculture. The production of crop decides its price. Most of the agricultural products are even more exported to foreign countries. So the productivity of the crop is essential. Andhra Pradesh is one such state in India where about 60%–70% of the population depends on irrigation sector. The state is known for food production especially rice. The yield of crop is based on the climatic conditions and the extent of rainfall in the area. Hence, forecasting of crop yield depending on the seasonal forecasts helps in minimizing the loss due to effect of climate variability and extremes under present-day climatic conditions [1]. The climatic conditions differ from place to place and state to state in India resulting in variation of productivity. This indicates the importance of grain yield monitors. Especially, in the state of Andhra Pradesh also called as “Annapurna” means the goddess of food grains has some regions where the Rice is produced for 3 times a year, in some regions for 2 times a year and few regions in which for only 1 time a year. As the departed state of Andhra Pradesh consists of 13 states in which about 9 states are in the coastal belt to the Bay of Bengal which is prone to 3 cyclones during October,

© Springer Nature Switzerland AG 2020 S. C. Satapathy et al. (Eds.): ICETE 2019, LAIS 4, pp. 688–692, 2020. https://doi.org/10.1007/978-3-030-24318-0_78

A Novel Approach for Rice Yield Prediction in Andhra Pradesh

689

November and December months every year apart from the seasonal rainfalls occurring during June to September every year. Hence, the need for dynamic yield prediction based on rainfall is essential in Andhra Pradesh where the sometimes there might be severe rainfalls due to which the crop might get washed away and sometimes there shall be severe droughts where the crop might get dried away due to lack of water wherein the Rice crop needs huge water during initials days of cultivation. Thus, the need for yield prediction is very much essential to plan for next crop season. Exploring the literature, the research work has been done widely based on various computational intelligence techniques to predict the yield [2–8]. Literature also shows that the early work has used even Self organising maps(SOM) [9] to predict the yield which are suitable for multivariate statistic problems [10], K-nearest neighbour [11], Machine Learning Techniques and Deep Learning Techniques [12, 13] suggested that the Deep Learning Techniques are more advantageous for the reason that we can overcome overfitting problem. Hence in this paper, we would like to propose Multiple Linear Regression Model to predict the crop yield. Also, the parameters considered for the yield prediction are Soil moisture, Soil fertility, Other soil parameters, Images acquired on extent of region under cultivation, Normalized Water index (NWI) [14], Nitrogen extent in soil [15–18]. The major essential parameters for yield prediction are Area Under Cultivation (AUC), Annual Rainfall (AR), and Food Price India (FPI). Therefore in this paper, the parameter AUC, AR and FPI are used to compute the productivity i.e. yield predication. The rest of the paper is organized as follows: next section explains the methodology, Sect. 3 demonstrates the results and discussion and the Sect. 4 concludes the paper.

2 Methodology Machine learning techniques are most reliable for solving the problems where relationship between the input variables and the output variables is hard to obtain or is not known [19]. In the proposed methodology we have used the dataset consisting of Average Rainfall (AR), Area Under Cultivation (AUC) and Year (Yr) fields. The dataset consists of data corresponding to the last 15 years. The below Fig. 1 demonstrates the working of the proposed system. However, the dataset provided about Andhra Pradesh is the statistical data of United Andhra Pradesh (Including present Telangana State). As the dataset is obtained from a standard source, the dataset has no missing data in it. The data cleaning techniques in this aspect is only blank spaces in the excel sheet. To overcome this problem, in this proposed model we have removed the blank unnecessary spaces by using TRIM function the obtained dataset.

690

N. Vadaparthi et al.

Fig. 1. Proposed model.

3 Experimental Results The experimentation has been carried out on Core i5 processor, 8 GB RAM and 1 TB HDD. The dataset is extracted from the “districtsofindia” website and local agricultural office in Bhogapuram, Vizianagaram (Dt.). The implementation of the model has been done in Python 3.6.5 version using Machine Learning libraries. The experimentation took about 8 min to complete the execution process. The below Fig. 2 shows the accuracy levels of the experimentation of proposed model. The Predicted values are close to the actual values. The accuracy of the model is observed to be about 90%. The predicted values are much more closure to the actual values.

A Novel Approach for Rice Yield Prediction in Andhra Pradesh

691

Fig. 2. Output showing predicted values & actual values

4 Conclusion From the experimental results it is clearly evident that the predicted values are very much closure to the actual values. Hence, the Multiple Linear Regression model yield better results compared to other models. However, the prediction shall be much apt if 13 more other parameters such as yield per unit area from time to time is fed, extent of loss occurred due to heavy rainfall or drought be given etc.

References 1. Toshichika et al (2018) Global crop yield forecasting using seasonal climate information from a multi-model ensemble. Climate Serv 11:13–23 2. Pantazi XE et al (2016) Wheat yield prediction using machine learning and advanced sensing techniques. Comput Electron Agric 121:57–65 3. Sellam V, Poovammal E (2016) Prediction of crop yield using regression analysis. Indian J Sci Technol 9(38). https://doi.org/10.17485/ijst/2016/v9i38/91714 4. Sahoo PK, Mani I (2018) Prospects of precision agriculture in india. In: Advances in agricultural engineering. Annual Technical Volume-II. The Institution of Engineers India, vol 2 5. Singh KK et al (2017) Crop yield forecasting under FASAL. FASAL Technical Report2017, IMD-Delhi 6. Miao Y, Mulla DJ, Robert PC (2006) Identifying important factors influencing corn yield and grain quality variability using artificial neural networks. Precis Agric 7(117):135 7. Effendi Z, Ramli R, Ghani JA (2010) A back propagation neural networks for grading Jatropha curcas fruits maturity. Am J Appl Sci 7(3):390–394

692

N. Vadaparthi et al.

8. Fortin JG, Anctil F, Parent LÉ, Bolinder MA (2010) A neural network experiment on the site-specific simulation of potato tuber growth in Eastern Canada. Comput Electr Agric 73 (2):126–132 9. Irmak A et al (2006) Artificial neural network model as a data analysis tool in precision farming. Trans ASABE 49(6):2027–2037 10. Drummond ST et al (2003) Statistical and neural methods for site-specific yield prediction. Trans ASAE 46(1):5–14 11. Liu J, Goering CE, Tian L (2001) A neural network for setting target corn yields. Trans ASAE 44(3):705–713 12. Ayoubi S, Sahrawat KL (2011) Comparing multivariate regression and artificial neural network to predict barley production from soil characteristics in northern Iran. Arch Agron Soil Sci 57(5):549–565 13. Norouzi M et al (2010) Predicting rainfed wheat quality and quantity by artificial neural network using terrain and soil characteristics. Acta Agric Scand Sect B - Soil Plant Sci 60 (4):341–352 14. Zolfaghari Z, Mosaddeghi MR, Ayoubi S (2015) ANN-based pedotransfer and soil spatial prediction functions for predicting Atterberg consistency limits and indices from easily available properties at the watershed scale in western Iran. Soil Use Manag 31(1):142–154 15. Shearer SA et al (1999) Yield prediction using a neural network classifier trained using soil landscape features and soil fertility data. ASAE Paper No. 993042. St. Joseph, Michigan, USA 16. Kohonen T (1988) Self-organization and associative memory. Springer, Berlin 17. Marini F (2009) Artificial neural networks in food analysis: trends and perspectives. Anal Chim Acta 635:121–131 18. Chlingaryan A, Sukkarieh S, Whelan B (2018) Machine learning approaches for Crop Yield prediction and nitrogen status estimation in precision agriculture: a review. Comput Electron Agric 151:61–69 19. Gonzalez A et al (2014) Predictive ability of machine learning methods for massive crop yield prediction. Spanish J Agric Res 12(2):313–328

Representation Techniques that Best Followed for Semantic Web - Web Mining K. Vaishali1,2 and Sriramula Nagaprasad1,2(&) 1

Jyothismathi Institute of Technology and Sciences, Karimnagar, India [email protected] 2 Tara Government Degree College, Sangareddy, India

Abstract. Web Mining is a Data Mining Technique used widely in mining billions of information from the World Wide Web (WWW) as faster as possible with the exact match of data. The huge information available in WWW with various formats, like: text format files, images, documents and other forms of data like structured, semi structured and unstructured forms. The amount of this information is increasing day by day. Data mining is the technique used to extract the data available in the internet. Web mining technique is used to determine and mine information from data sources related to web which are documents in web, contents in web, server logs and hyperlinks. The Semantic Web is used to provide information in a defined meaning that enhanced the interoperability between human and machines, which created the space for the machines to handle most of the decisions and tasks. This paper gives a brief idea regarding representation techniques that are best used in semantic web. Keywords: Web mining
Web mining techniques Data mining technique
Representation techniques


Semantic web


1 Introduction One of the major applications of data mining techniques which is used in discovering models or patterns for the content, structure and usage within the web pages from the WWW - World Wide Web is popularly known to be web mining. Web mining as the word describes, the data or information will be gathered by mining the web. This can be applied for both structured and unstructured information in the form of browser activities, page content, website, server logs, link structure and different sources. Semantic web and web mining are the quick rising technologies in the study areas. Web content, web usage and web structure mining are three important web mining types used to satisfy the whole process of data mining in web mining [1]. Web content mining methods are used for Semantic Annotation creation from web page content; on the other part it also profits the content that is structured already in RDF, XML or Ontology. To understand and serve better requirements of Web-based applications and to discover usage patterns, Web usage mining methods are used effectively [1].

© Springer Nature Switzerland AG 2020 S. C. Satapathy et al. (Eds.): ICETE 2019, LAIS 4, pp. 693–699, 2020. https://doi.org/10.1007/978-3-030-24318-0_79

694

K. Vaishali and S. Nagaprasad

2 Web Mining Discovering patterns from the large databases with knowledge discovery has been processed with the help of a field in Computer Science known as Data Mining. Moreover the information are extracted and transformed to an understandable format that can be used further for prediction in the future or for any other purpose. To extract knowledge data from web such as logs used for websites, hyperlinks used between documents, documents of web etc, Data Mining technique use one of its applications – web mining [9]. One of the major differences between web mining and data mining is, web mining find patterns which are very useful from the web data such as logs, hyperlinks and documents, whereas data mining works with data from the database and find out the patterns that are useful. The raw data from data mining will be always in structured form which can be further used in mere prediction, but in web mining the raw data that is available will be either semi-structured or unstructured which then converted to structured format for knowledge extraction. Web mining techniques are divided in three types: (1) web content mining, (2) web structure mining and (3) web usage mining. (1) Web Content Mining Through this technique, information is extracted from the data available from the web in the form of web documents. The data available will be in various forms like images, audio, text, video, table etc. Most research in web content mining has been processed using knowledge extraction from text data. NLP - Natural Language Processing and IR - Information Retrieval technologies are also widely utilized in web content mining. In recent years image processing is also getting influenced for extracting data from images. (2) Web Structure Mining Web structure mining technique is mainly designed to focus on web structured data. For instance when the data is considered to be a graph, then the web pages are set of nodes belongs to that graph and edges are hyperlinks that connects different nodes in the web pages. This always deals with information that is structural from the web. Document structure and hyperlinks are the two different classifications of web structure mining. Hyperlinks are used mainly for structural units which connect web page location to different web page location, either in the same page or in different pages in the web. Then the content is organized in format known to tree-structured which is based on XML and HTML tags within the specific page [9] (Fig. 1).

Representation Techniques that Best Followed for Semantic Web

695

Fig. 1. Techniques used in web mining

(3) Web Usage Mining Mining technique that process with user’s web usage patterns from the web logs in web sites is said to be web usage mining. To realize the user’s patterns in browsing and for better service of web-based applications at present and modification in future web usage mining technique always provides its best. The data that are used for this mining technique are identity, location, browsing patterns of the user, etc which are obtained from website usage log. Web server data, application server data and application level data are the three types of web usage mining. User log information like IP address, time of access and reference, etc are collected from the web server and logged in web server data [9]. Various business events are tracked and logged using application server data. Events defined within the same logged application and sourced by creator are tracked and logged by application level data.

3 Semantic Web The transformation of information oriented web to knowledge oriented web is carried by semantic web a joined progress lead with the standards followed by international body - W3C. Semantic web is a powerful extension of (WWW) World Wide Web. Semantic web provide a standard for expressing web page relationships by allowing the machines to understand and accelerate the exact meaning of information that are hyperlinked [10]. Tim Berners-Lee who is the inventor of WWW and Director of W3C has coined the term “Semantic Web” – for the data from web that can be processed by the machines [10]. Computer basically does not understand the textual data that are unstructured, hence semantic web help computers to interpret the read data by adding meta-data to the pages in the web. Moreover this would never add any AI to machines, nor will construct self-awareness to the system, but will definitely provide machine tools for finding information, exchanging data and interpreting for a few levels [11]. The ultimate aim of semantic web is converting the web which includes semistructured or unstructured documents to a web of data by adding semantic content in the web pages.

696

K. Vaishali and S. Nagaprasad

4 Semantic Web Representation Techniques To communicate and to express semantic of information several models and accessing strategies are utilized. W3C has suggested standard systems such as XML (Extensible Mark-up Language), RDF (Resource Description Framework) and OWL (Web Ontology Language) [3]. 1 Extensible Mark-Up Language (XML) XML (Extensible Mark-up Language) strategies have the power of recouping data from the web. In engaging customers, to make their own specific marks, it licenses them to portray the content adequately. Along these lines, the set of information and the semantic connection streams of that information are able to be addressed [5, 6]. 2 Resource Description Framework (RDF) By utilizing their own domain vocabularies, RDF (Resource Description Framework) has been entitled with the capability of storing data which are retrieved and used by resources on accessing the WWW [3, 4]. The three categories of content elements available with RDF are, (a) Resources (entities are recognized by using URIs) (b) Literals (atomic series such as numbers, strings, etc) (c) Properties (binary associations recognized using URIs - Uniform Resource Identifiers) [2]. An extremely efficient method for representing several type of information which is defined in web is RDF [3]. 3 Web Ontology Language (OWL) When compared with RDF, OWL is said to be more complicated language with enhanced ability for interpreting. Nature of the resource and their relationships are accurately identified by OWL. For representing the information of semantic web, the OWL utilizes ontology which is a demonstration of proper clear clarification of common procedure and basic input [4, 6]. OWL accurately identifies the sources’ character and association. Developers of Ontology have expressed the attention on domains that is class based and properties such as representing rules and atomic distinct concepts in some further semantic languages too. Sir Berners-Lee has examined the architecture of Semantic Web in seven layers [7] (Fig. 2), (1) (2) (3) (4) (5) (6) (7)

URI XML, NS, & XML schema RDF & RDF schema The Ontology Vocabulary Logic Proof and Trust

Representation Techniques that Best Followed for Semantic Web

697

Fig. 2. Architecture of semantic web - seven layers

• URI – Responsible for encoding process of resource and their identification. • XML, NS, and XML schema layer – Responsible for (1) Division made towards content information, structural information and design performance by following linguistic (2) Providing Standard Language Format. • RDF and RDF schema – By using Semantic model, this layer defines the information on WWW and its types. • Ontology Vocabulary layer – This layer is mainly focused on disclosed semantics between data in the way of characterizing the shared knowledge and the relations of semantic inside various types of data. • Logic layer – The foundation of intelligence services like logical reasoning by providing inference principles and axioms are taken care by this layer. • Proof and Trust layers - Mechanisms based on digital signature and encryption are used for recognizing alteration made with the papers for the purpose of enhancing the web security.

5 Ontology and Web Ontology Language (Owl) – A Best Representation Technique for Semantic Web The backbone of semantic web is Ontology - a representation technique. Ontology has been defined by different literatures in different ways; some of them have been mentioned here, (1) It is a official demonstration which contains the group of ideas and associations [8]. (2) It is an explicit specification of conceptualization [12]. (3) It is a term in philosophy and its meaning is “theory of existence” [13]. (4) It is a body of knowledge describing some domain, typically common sense knowledge domain [13]. The best technique followed in semantic web which is understandable by both humans and machines is strongly said to be Ontology. Semantic web - meaning

698

K. Vaishali and S. Nagaprasad

assigned to the web, is followed by Ontology. The creation of Ontology is a semiautomatic procedure. All the data of information that is extracted from a semistructured or unstructured data forms a structured format and then inserted into the knowledge base is known as Ontology creation. To improve the results of user’s query, the information which is available from knowledge base is utilized in process of web mining. For authorizing Ontologies or knowledge bases a unit of knowledge representation languages or languages of ontology like web ontology language (OWL) is used. This language is categorized as formal semantics, and RDF/XML oriented serialization for semantic web [14].

6 Conclusion After analyzing various categories of web representation techniques in the process of extracting knowledge source from WWW information for semantic web, it is very well may be reasoned that the information which are unstructured, present in the web pages can also be verified and checked to make ontologies for colonizing knowledge base in the search of web. Data embedded with knowledge base are given in organized way so that the machine will recognize perfectly. Data that are retrieved from knowledge base are then utilized with computer system to give better enhanced results for requested web user queries. In this manner semantics can be appended to the present web through knowledge extraction method for making ontologies towards the formation of semantic web.

References 1. Sitha Ramulu V, Santhosh Kumar ChN, Sudheer Reddy K (2012) A study of semantic web mining: integrating domain knowledge into web mining. Int J Soft Comput Eng (IJSCE) 2 (3). ISSN 2231-2307 2. Stumme G, Hotho A, Berendt B (2006) Semantic web mining: state of the art and future directions. Web Semant: Sci Serv Agents World Wide Web 4(2):124–143 Semantic Grid – The Convergence of Technologies 3. Jeon D, Kim W (2011) Development of semantic decision tree. In: Proceedings of the 3rd international conference on data mining and intelligent information technology applications, Macau, 24–26 October 2011, pp 28–34 4. Sugumaran V, Gulla JA (2012) Applied semantic web technologies. Taylor & Francis Group, Boca Raton 5. Domingue J, Fensel D, Hendler JA (2011) Handbook of semantic web technologies. Springer, Heidelberg 6. Jain A, Khan I, Verma B (2011) Secure and intelligent decision making in semantic web mining. Int J Comput Appl 15(7):14–18. https://doi.org/10.5120/1962-2625 7. Yong-Gui W, Zhen J (2010) Research on semantic web mining. In: Proceedings of the international conference on computer design and applications, Qinhuangdao, 25–27 June 2010, pp 67–70. https://doi.org/10.1109/iccda.2010.5541057u

Representation Techniques that Best Followed for Semantic Web

699

8. Jayatilaka ADS, Wimalarathne GDSP (2011) Knowledge extraction for semantic web using web mining. In: The international conference on advances in ICT for emerging regions, ICTer2011 9. Srivastava J, Desikan P, Kumar V. Web mining – concepts, applications and research directions 10. Berners-Lee T, Hendler J, Lassila O. The Semantic Web. http://semanticweb.org/wiki/ Semantic_Web 11. Wilson TV. How Semantic Web Works. HowStuffWorks.com: http://www.howstuffworks. com/semantic-web.htm 12. Gruber T (1993) A translation approach to portable ontology specifications 13. Obitko M. What is Ontology. http://www.obitko.com/tutorials/ontologies–semantic–web/ what-is-ontology.html 14. Web Ontology Language. http://en.Wikipedia.org/wiki/Web_Ontology_Language

Isolated Health Surveillance System Through IoT Using Raspberry Pi Sumayya Afreen1(&), Asma Begum2, G. Saraswathi2, and Ayesha Nuzha2 1

Department of Computer Science and Engineering, Osmania University, Hyderabad, India [email protected] 2 Department of Computer Science and Engineering, Stanley College of Engineering and Technolgy for Women, Hyderabad, India {basma,gsaraswathi}@stanley.edu.in, [email protected]

Abstract. Among the array of applications enabled by the Internet of Things (IoT), smart and connected health care is a particularly important one. Patient surveillance system using Raspberry Pi is one among applications in health care to observe the patient health condition, Internet of Things makes medical equipments a lot of economical by permitting real time observation of patient health, within which detector acquire knowledge of patient’s and reduces the human error. In Internet of Things patient’s parameters get transmitted through medical devices via an entryway. The numerous challenges within the implementation of health care applications are to observe all the patients from various places. Thus this idea brings out the answer for effective patient observation at reduced value and altogether reduces the trade-off between patient outcome and wellness management. In this surveillance system we are able to store the detector parameters within the natural information. These values are retrieved to mobile application. In mobile app we are able to monitor these parameters and whenever there’s some case of emergencies the care takers are notified with messages. Keywords: Internet of Things


Raspberry Pi

1 Introduction Health is one among the world challenges for humanity. In the last decade the healthcare has drawn large quantity of attention. The prime goal was to develop a reliable good health observation system in order that the health care professionals will monitor the patient’s conditions based on which care is taken [1]. The Raspberry Pi which is a low cost, flexible, absolutely customizable and programmable small pc board brings the benefits of a laptop to the domain of sensing element network. In our system we tend to present sensing of patient’s parameters (Temperature, heartbeat, pulse etc.) with different sensors. Now successive new mega trend of web is IoT. Visualizing a world where many objects will sense, communicate © Springer Nature Switzerland AG 2020 S. C. Satapathy et al. (Eds.): ICETE 2019, LAIS 4, pp. 700–706, 2020. https://doi.org/10.1007/978-3-030-24318-0_80

Isolated Health Surveillance System Through IoT Using Raspberry Pi

701

and share data over a personal Internet Protocol (IP) or Public Networks. This can be the planet of the IoT, it’s typically thought-about as connecting objects to the net and using that connection for management of these objects or remote observation [1, 2]. 1.1

Objectives

Here the key objective is to present a Patient Health Surveillance System to diagnose the health condition of the patients. Giving care and health help to the poor health patients at vital stages with advanced medical facilities became one amongst the foremost issues within the smart agitated world [2, 3]. In hospitals wherever several patients whose physical conditions should be monitored regularly, the necessity for an economical and quick responding alert mechanism is required. Correct implementation of such systems will give timely warnings to the medical staffs and doctors and their service will be activated just in case of medical emergencies. Current systems use sensors that are hardwired to a computer next to the bed [3].

2 Problem Specifications Remote health surveillance will give helpful physiological information for patients staying at home. This surveillance is helpful for aged or sick patients. Wireless devices collect and transmit signals of interest and a processor is programmed to receive and mechanically analyze the sensor signals [5]. Employing a single parameter surveillance system subordinate approach to a remote health observing system was designed that extends attention from the standard clinic or hospital setting to the patient’s home. The system was designed to gather heartbeat detection system information, fall detection system information, temperature information and few different parameters. The information from one of the parameter observation systems was then given for remote detection. 2.1

Characteristics

During design the subsequent characteristics of the longer term medical applications are being adhered: • Integration with current trends in medical practices and technology. • Period of time, long-term, remote observation, miniature, wearable sensors and long battery lifetime of a designed device. • Help to the aged and chronic patients. The device ought to be straightforward to use with mobile application. 2.2

Methodology

In this system we have temperature, pulse, ECG sensors with Raspberry Pi. These sensors signals are send to the Raspberry Pi via Wifi module. Raspberry pi is a Linux based Operating System works as a small computer processor system. Here patient’s temperature, pulse is measured with help of several sensors and these are often

702

S. Afreen et al.

monitored within the monitor screen of pc installed Raspberry Pi likewise as watching through anyplace within the world with web connectivity. The projected technique of patient surveillance system is to monitor patient’s temperature, ECG, pulse with Raspberry Pi being present. When connecting web to the Raspberry Pi it act as a server [4]. Then the server spontaneously sends information to the cloud. It will monitor the patient’s health condition anyplace within the world with laptops, tablets and sensible phones. If these parameters area unit goes to abnormal it’ll automatically sends alert mail to the doctors and relatives. 2.3

Contributions

• The Sensors with a tool enabled by wireless communications to observe live physical parameters. • Restricted information storage at patient’s database that interfaces between sensors and different centralized information repository and/or aid suppliers. • Centralized repository to store information sent from sensors, native information storage, diagnostic applications and/or aid suppliers 2.4

Applications

The remote health surveillance system is often applied within the following scenarios • A patient is understood to have a medical condition with unstable restrictive body system. This can be in cases wherever a extra medication is being introduced to a patient. • A patient is vulnerable to heart attacks or might have suffered one before. The organ is monitored to predict and alert ahead any indication of any such condition. • Crucial body organ situation. • Matters resulting in the event of a risky severe condition. This can be for people at a complicated age and perhaps having failing health conditions. • Athletes throughout coaching. 2.5

Limitation of the System

The scope of the project was being restricted to visualization of temperature detection and remote visualization of the collected information for one patient. Here, the foremost necessary specification thought was that they must be safe to use and correct. This can be as a result of the live data being detected which determines the severity of a vital serious state of concerns.

3 Related Works In the existing system, we tend to use active network technology to network varied sensors to one Patient monitoring system (PMS). Patients varied crucial parameters are endlessly monitored via single PMS and reported to the Doctors or Nurses present for timely response just in case of crucial things. They are connected to the body of the patients while not inflicting any discomfort to them.

Isolated Health Surveillance System Through IoT Using Raspberry Pi

703

During this Patient monitoring system we tend to monitor the necessary physical parameters like energetic signal, heart beat rate and blood level force per unit. Thus, the analog values which are detected by the various sensors are then given to microcontroller connected. The microcontroller processes these analog signal values of health parameters individually and converts it to digital values mistreatment ADC device [5, 6]. Now, the digitalized values from quite one microcontroller are sent to the Central PMS. Each of the sensors connected microcontroller with a transceiver can act as a module that has its own distinctive ID. Every module transmits the information wirelessly to the entryway connected to the computer of the Central PMS. The entryway is connected to the computer i.e. Central PMS that is settled within the centre, is capable for choosing totally different patient IDs and permitting the entryway to receive different physical parameter values the patient mere by the ID [6]. The code designed Graphical user interface will maintain totally different physical parameters of every patient, consecutively with a mere quantity for every patient. At any time, any of the doctors or nurses will start browsing the Central PMS and check the history of the exposed crucial parameters of any of the patient connected to the network. 3.1

Isolated Health Surveillance System Through IoT Using Rasberry Pi

The main objective is to design a Patient Health Observation System with two-way communication i.e. not solely. The patient’s information are going to be sent to the doctor through SMS and email on emergencies, however altogether the doctor will send needed suggestions to the patient or guardians through SMS or Emails. Now we track patient’s location at associate purpose in time through Google Maps which might help to send medical services just in case of an emergency for non-bed ridden patients. The main plan of the designed system is to unendingly monitor the patient over Internet. The system is projected to supply LIVE updating of information and emergency alerts. It provides communication between doctors and patients through sensors.

4 Proposed Architecture (See Fig. 1).

Fig. 1. Proposed architecture.

704

4.1

S. Afreen et al.

Working of the Proposed Architecture

Here there is a communication between Raspberry pi (rpi), sensor device and base of operations information. The following is the description of the flow during which the communication is completed. First step is to initialize the Rpi, sensors and mobile app and connect each device to Bluetooth. – Once the association is completed, the user will initialize the device once the sensors connected to the body which start out sensing information. The observed the information is send to base of operations and therefore the user retrieves the detected information through base of operations. – Here they’re principally two varieties of user [7]. (1) Patient - The user who undergoes the observation through sensors. (2) Relative/others - The user ‘checking patient health conditions - For relative user able to access patient health condition, the relative user contact should be entered in patient user’s relative’s field. • The cloud additionally analyses the information and checks whether or not the edge values area unit met. • If the control is not met then it sends notification to relatives. • The application additionally options setting reminder for doctor’s appointment.

5 Implementation 5.1

Temperature Sensing Element

Temperature sensing element is a device that is intended specifically to record the hotness or coldness of an object. LM35 is a correctness IC temperature sensing element with its output proportional to the temperature (in °C). With LM35, the temperature is measured additional accurately than with a thermal resistor. It additionally possesses low self-heating and doesn’t cause heating over zero. 1 °C temperature usually rises in still air. The in-operation temperature varies from −55 °C to 150°C. The LM35’s low output ohmic resistance, linear output, and precise essential activity create interfacing to readout or manage electronic equipment is particularly simple [4]. 5.2

ECG Sensor

Electrocardiogram sensor records the electrical activity generated by memory muscle depolarization’s that propagate in rhythmic electrical waves towards the skin. Though the electricity quantity is in fact terribly small, it may be picked up dependably with ECG electrodes connected to the skin (Fig. 2).

Isolated Health Surveillance System Through IoT Using Raspberry Pi

Fig. 2. Sensors and devices used

6 Results (See Figs. 3, 4 and 5).

Fig. 3. Pulse record.

Fig. 4. Temperature record.

705

706

S. Afreen et al.

Fig. 5. Electrocardiogram record.

7 Conclusion Here we tend to conclude that sensors with the microcontroller/System on chip with raspberry pi can facilitate to build an environment where it would be possible to watch patient’s LIVE body parameter. It will guarantee safety of patient and facilitate keeping check of patient from remote location. This work can be enhanced by collecting the data i.e., temperature, pulse rate and heart beat recorded from the sensors and applying suitable prediction algorithms to predict the future health stability of the patient. The prediction algorithms may also help in suggesting the precautions the patients’ needs to take in future to be safe from any health issues.

References 1. Raskovic D, Revuri V, Giessel D, Milenkovic A (2015) Embedded internet server for wireless device networks. National syndicate for MASINT Research beneath Defense Intelligence Agency/independent agency Grant No. 11S-043415 International Journal on Engineering Technology and Sciences – IJETS™ ISSN (P) 2349-3968, ISSN (O) 2349-3976, Volume II, Issue IX 2. Gill K, Yang SH, Yao F, Lu X (2009) A ZigBee-based home automation system. IEEE Trans Client Nat Philos 55. https://doi.org/10.1109/TCE.2009.5174403 3. Yan H, Tsang KF, Ching H, Chui KT (2013) The style of twin radio ZigBee homecare gateway for remote patient monitoring. IEEE Trans Client Electron 59(4) 4. Wieselthier J, Nguyen G, Ephremides A (2000) On the construction of energy-efficient broadcast and multicast trees in wireless networks. In: Proceedings of IEEE INFOCOM, TelAviv, Israel 5. Kwon TJ, Gerla M (1999) Clustering with power control. In: Proceedings of IEEE MILCOM 6. Le Boudec JY, Vojnovic M (2005) Perfect simulation and stationary of class

Design and Implementation of RPL in Internet of Things M. V. R. Jyothisree1(&) and S. Sreekanth2 1

Computer Science, Rayalaseema University, Kurnool, A.P., India [email protected] 2 Department CSE, Sitams, JNTUA, Anantapur, A.P., India [email protected]

Abstract. The Internet of Things (IoT) is a fast growing technology. In IoT, the devices are connected through the Internet and controlled from any remote areas. Before the advent of IoT, the interaction between the users was only through the internet. By 2020 there will be 75.4 billion devices interconnected through the internet. Machine-to-machine (M2M) interaction is achieved by sending and receiving the information, such as room temperature, humidity etc. IoT can be viewed as heterogeneous networks that bring some security challenges like network privacy problems, confidentiality, integrity and availability. In IoT, we have Routing Protocol for Low-Power and Lossy networks (RPL). RPL is a light weight protocol which has good routing functionality, context aware and supports dynamic topology but has only basic security functionality. This paper elaborates on Routing Protocol for Low Power and Lossy networks (RPL) and its implementation. Along with that, we have surveyed on different RPL attacks in network layer based on confidentiality, integrity and availability. We further conclude with research challenges and future work needed to be done in order to have secure RPL for Internet of Things (IoT). Keywords: Internet of Things
RPL
M2M
RPL security
Low power and Lossy networks
Confidentiality
Integrity
Availability Attacks
Contiki/Cooja




1 Introduction The Internet of Things (IoT) is a technology that is moving with rapid pace. The main aim of IoT is to create an environment where the devices communicate among themselves without human interference. IoT is a world-wide heterogeneous network which consists of interconnected objects and unique address based on standard communication protocols. In IoT, ‘Internet’ is a world-wide network of interconnected computer networks based on the (TCP/IP) communication protocols and ‘Thing’ is an object or any device. IoT allows human to be connected at any time to the remote devices. A device can be connected to other device using any path/ network or by any service. Various types of communication can be utilized in IoT if the communication process transmits the information between the heterogeneous devices via heterogeneous networks. © Springer Nature Switzerland AG 2020 S. C. Satapathy et al. (Eds.): ICETE 2019, LAIS 4, pp. 707–718, 2020. https://doi.org/10.1007/978-3-030-24318-0_81

708

M. V. R. Jyothisree and S. Sreekanth

Different routing protocols are needed in IoT for device-to-device communication but we require scalable routing protocols in different scenarios to find optional routes. The routing protocol for Low Power and Lossy Networks (RPL) are standardized for IoT device networks. In this paper, we explore RPL protocol by studying it’s security with respect to different attacks in IoT. The rest of this paper is organized as follows. Section 2 talks briefly about survey on RPL protocol, Sect. 3 provides description of RPL protocol, Sect. 4 presents implementation results of RPL and attacks in RPL, Sect. 5 gives conclusion and highlights the research challenges in IoT.

2 Literature Survey In IoT, security is a highly challenging issue [4]. Survey has been done on security modes [3] available in IoT and different RPL attacks in network layer. A. RPL Security Security is associated with low power and lossy networks. RPL nodes [5] operate mainly on three security modes. They are: (1) Unsecured mode (2) Pre-installed mode (3) Authentication mode A brief description of these modes is now given as: (1) Unsecured Mode: In this mode, RPL control messages are forwarded without any extra security measures. It infers RPL network by using other security primitives to meet the specific requirements and application needs. (2) Pre-installed mode: In this mode, RPL instances have pre-installed keys to join them so as to process, safeguard and generate a secure RPL messages. (3) Authenticated mode: In this, nodes can be entered from left node. It is similar to pre-installed mode, with the pre-installed keys that enter as forwarding nodes by getting the key from an authentication authority. The RPL security is based on three factors: Confidentiality, Integrity and Availability (CIA) [2]. Delay protection and replay protection are an added option in RPL security. B. Attacks in RPL In RPL, classification of attacks [8] is based on CIA. There are different types of attacks [7] in RPL that affect the network performance. The types of attacks are as follows: (1) Rank attack: This attack mainly focuses on confidentiality and integrity which affects the network performance by generating loops, non-optional path, low packet delivery ratio, packet drops and packet delays. (2) Selective forwarding: This attack is also based on confidentially and integrity that affect the network performance by disrupting the traffic flow. (3) Sinkhole attack: This attack affects by transmitting huge traffic passing via attacker node. This comes under confidentiality and integrity attack.

Design and Implementation of RPL in Internet of Things

709

(4) Hello flooding attack: This attack affects the network performance by using excess battery power and sensor energy. This attacks comes under the availability attack in RPL. (5) Sybil and Clone ID Attack: It mainly focuses on confidentiality and integrity attack that affects the network performance by way of compromised route or Broken network and traffic is unreachable to victim’s node. (6) Denial of service (DOS) attack: This attacks occurs due the unavailability of network resources so, it comes under the classification of attacks. (7) Wormhole attack: This attack mainly focuses on the confidentiality and integrity attack because of which it affects on network performance by destabilization of route topology and traffic flow. (8) Black hole attack: This attack is classified under CIA, that affects increasing the network performance by dropping packets, and route traffic and high control overhead. (9) Version number attack: This attack affects the increased control overhead, high traffic latency, low packet delivery ratio and high end to end delay. This belongs to confidentiality and integrity attack. (10) Neighbor attack and DIS attacks: It mainly affects the performance by giving false routes or no routs, resource consumption and route disruption. Network resource depletion by neighbor attack and network resource consumption takes place by DIS attacks. These attacks falls under confidentiality, Integrity and Availability attacks. (11) Local repair control overhead attack: This attack affects the disruption of routing traffic and control which is classified under confidentiality and integrity attack.

3 RPL Description RPL is a IPV6 based routing protocol designed by IETF group known as Routing over Low Power and Lossy Networks [1]. RPL is also known as distance vector routing protocol for LLNs. It has a tree-based topology known as Destination Oriented Directed Acyclic Graph (DODAG) [6]. Path-selection is an important factor in RPL. In RPL, every node selects the preferred parent node based on some metrics. Nodes are organized as DoDAGs. The rank of the nodes depends on the arrangement of nodes in the tree. The node rank decreases in the upward direction towards the DODAG root and increases from DODAG root to the leaf nodes (i.e., sender node). In RPL, there are three control message types. They are (i) DODAG Information Object (DIO), (ii) DODAG Advertisement Object (DAO) and (iii) DODAG Information Solicitations (DIS). These messages are described as under: (i) DODAG Information Object (DIO):- In RPL network, nodes exchange DODAG information via the DIO. It is used mainly for the creation and maintenance of DODAG topology. Every node in RPL selects its preferred parents with the help of DIO. (ii) DODAG Advertisement Object (DAO):- DAO messages are used to transmit the prefix of a node to its ancestor node in the network for downward routing purposes. In RPL we have DAO – Ack which is DODAG advertisement object

710

M. V. R. Jyothisree and S. Sreekanth

acknowledgment. It is used to send an acknowledgement to its prefix node from an ancestor node that the DAO has been received by that node. (iii) DODAG Information Solicitation (DIS):- It is used by any unattached node in the network. DIS messages are used when a new node seeks topology information and is waiting to join the network. Out of the above three messages, DAO and DIS are involved only when there is a topology change where as DIO message is used only for the purpose of starting a change in topology change.

Fig. 1. RPL control messages

Figure 1 shows a typical RPL tree structure consisting of different nodes along with a root node connected together based on DODAG topology. Each node is identified by the IPV6 address and rank. In RPL network we have one-to-one, one-to-many and many-to-many communication channels between the nodes. It is unidirectional towards DODAG root and bidirectional between the constrained devices (nodes) and the DODAG root.

4 Simulation Results of RPL In the IoT, we use Instant Contiki 3.0 version platform to perform the simulation. Contiki is an open source operating system. It is designed mainly for the tiny devices and thus the memory footprint is less than that of other systems. It supports TCP with IPV6 addressing format that is mostly used in IoT applications. One of the most important features of Contiki OS is the use of Cooja simulator to emulate if any of the hardware devices is not available. Ubuntu is used to compile the programs for motes. Contiki OS is very robust and can be used as the universal operating system for smart objects (devices). Wireshark is a tool which is in built and used for analyzing the traffic between the motes (devices). The various simulation parameters are listed in Table 1.

Design and Implementation of RPL in Internet of Things

711

Table 1. Simulation parameters for RPL protocol Simulation parameters Simulation tool Mote type Simulation run time Simulation coverage area Interference range Total number of motes Router mote Deployment environment Wireless transmission range Network protocol Routing protocol

Contiki/Cooja3.0 Sky mote 3600 s 70 m * 70 m 10 m 5 1 Smart building 50 m IP based RPL

In Cooja simulator, during the simulation, system takes into account the interference range of the surroundings of the other devices or other technologies that may be in use. The steps for RPL DODAG implementation using Cooja simulator are as follows: Step 1: Open the instant Contiki 3.0 and give the password as user Step 2: Go to the terminal and use the following command (Figs. 2 and 3): $ cd/home/user/contiki 3.0/tools/cooja $ ant run

Fig. 2. Terminal Window in ContikiOS

712

M. V. R. Jyothisree and S. Sreekanth

Fig. 3. RPLDOdag simulation

Setp 3: Go to file menu, select new simulator and save it as simple-udp-rpl (Fig. 4).

Fig. 4. RPLDOdag Contiki process source

Design and Implementation of RPL in Internet of Things

713

Setp 4: Create new simulation when a window will be opened. Give simulation name:- RPL DODAG and click on create button. Step 5: Go to Motes menu and select add motes, then choose create new mote type, and finally press on Sky Mote button (Fig. 5).

Fig. 5. Adding motes in RPLDOdag

Step 6: Select create Mote type – Compile Contiki for description: Sky Mote type # Sky1 and Click on browse button. The following window will appear (Fig. 6): Step 7: Select IPv6 folder and select Simple-udp.rpl and click on Open button. Step 8: Select unicast-receiver.c as in Step 6. Step 9: Return back to window that appeared in Step 5. Click on compile button and wait until it executes and then on create button (Fig. 7). Step 10: Add one mote(sky mote type #skt1) when a window opens. Add motes and create new mote type as sky mote. Click on ‘Browse’ button. Go to Step6 and select unicast-sender.c (Figs. 8 and 9). Step 11: Add as many motes as we want and differentiate between motes by giving mote ids, addressIP or Rime, radio traffic, 10 m background grid, LEDs etc. (Fig. 10). Step 12: See the random arrangement of motes in Fig. 11. The green color area shows radio range of RPLDODAG. Step 13: Press start button in the simulation control window to see the traffic between the motes.

714

M. V. R. Jyothisree and S. Sreekanth

Fig. 6. Unicast-receiver.c

Fig. 7. Compilation of unicast-receiver.c

Design and Implementation of RPL in Internet of Things

Fig. 8. Adding motes for #Sky 1

Fig. 9. Unicast-sender.c

715

716

M. V. R. Jyothisree and S. Sreekanth

Fig. 10. Motes with IP addresses

Fig. 11. RPLDOdag with motes

Design and Implementation of RPL in Internet of Things

717

Step 14: Start serial Socket (SERVER) (sky1) to see the simulation of number of bytes moving between Socket and Motes. Step 15: Find the final output in Mote output window as in the Fig. 12.

Fig. 12. Final output of RPLDOdag

5 Conclusion There are different routing protocols available for communication between the heterogeneous devices. Routing protocol for low power and lossy networks (RPL) is a lightweight protocol having good routing functionality. This paper focuses on the implementation of RPL using Cooja simulator in ContikiOS and also discusses RPL security and different types of attacks in RPL. Our future work is to implement any two attacks in RPL and provide the prevention techniques for these attacks and there by securing RPL.

References 1. Thubert P et al (2012) RPL: IPv6 routing protocols for low power and lossy networks. RFC6550 2. Wallgren L, Raza S, Voigt T (2013) Routing attacks and countermeasures in the RPL-based Internet of Things. Int J Distrib Sens Netw 2013:794326 3. Bhabad MA, Bagade ST (2015) Internet of Things: architecture, security issues and countermeasures, 125(14) 4. Hota J, Sinha PK (2015) Scope and challenges of internet of things: an emerging technological innovation 5. Tsao T, Alexander R, Dohler M, Daza V, Lozano A, Richardson M (2015) A security threat analysis for routing protocol for low-power and lossy networks (RPL)

718

M. V. R. Jyothisree and S. Sreekanth

6. Mayzaud A, Sehgal A, Badonnel R, Chrisment I, Schönwälder J (2014) A study of RPLDODAG version attacks. In: 8th IFIP WG 6.6 international conference on autonomous infrastructure, management, and security, AIMS 2014, Brno, Czech Republic, June 2014, pp 92–104 7. Mayzaud A, Badonnel R, Chrisment I (2016) A taxonomy of attacks in RPL-based Internet of Things. Int J Netw Secur 18(3):459–473 8. Pongle P, Chavan G (2015) A survey: attacks on RPL and 6LoWPAN in IoT. In: 2015 international conference on pervasive computing (ICPC). IEEE

Detection and Tracking of Text from Video Using MSER and SIFT M. Manasa Devi1,2,3(&), M. Seetha1,2,3, S. Viswanada Raju1,2,3, and D. Srinivasa Rao4 1

CSE, VNRVJIET, Hyderabad, India [email protected], [email protected], [email protected] 2 CSE, GNITS, Hyderabad, India 3 CSE, JNTUH CE, Manthani, India 4 IT, VNRVJIET, Hyderabad, India [email protected]

Abstract. Text that looks in a scene or is explicitly added to video can offer an imperative additional basis of directory evidence as well as evidences for interpreting the video’s arrangement and for classification. Computerized text mining from a number of stationary resources quickness up the progression in workplaces, libraries, banks and an assortment of further places. Text extraction can be completed expending a quantity of various methods provisional upon the necessity of system and exactness level. In this paper, we present and implemented two popular algorithms Maximally Stable Extremal Regions (MSER) and Scale Invariant Feature Transform (SIFT) for spotting and tracking text in digital video. We analyzed results with respect to accuracy of text detection and tracking from videos. Experimental results shows that SIFT are 80% more accurate than MSER in the process of detection and tracking for extraction of text from video. Drawbacks of these two algorithms are also identified. This research paper appearance the diverse alterations that can be made to present text mining procedures by means of applying deep learning based recurrent convolution neural networks (CNN) to rectify drawbacks of two popular proposed techniques. CNN have advantages like local spatial consistency in the input (often images), which permit them to have smaller amount weights as some parameters are shared. This process, taking the form of convolutions, makes them especially well-suited to extract relevant information at a low computational cost. Keywords: Extraction
MSER Extraction
Deep learning


SIFT
CNN
Detection
Tracking


1 Introduction Extracting textual evidence from expected images is a interesting tricky with many real-world solicitations. Automated text extraction from paper forms is very frequently required these days and the associated procedure is called Optical Character Recognition (OCR). OCR comes in the field pooled by computer vision, pattern matching and © Springer Nature Switzerland AG 2020 S. C. Satapathy et al. (Eds.): ICETE 2019, LAIS 4, pp. 719–727, 2020. https://doi.org/10.1007/978-3-030-24318-0_82

720

M. Manasa Devi et al.

artificial intelligence. Optical character recognition is a procedure in which an automated device inspects the printed characters on paper and decides their outlines by spotting patterns of dark and light. Maximally Stable Extremal Regions (MSER) detector is selected to excerpt binary sections since it has confirmed to be robust to lighting situations. An improvement method for MSER images is calculated to attain pure letter margins. Images are then served into a Stroke Width Detector and numerous heuristics are utilized to eliminate non-text pixels. Later, discovered text sections are served into an Optical Character Recognition section and then strained affording to their assurance metric. The recognition of characters is not portion of the procedure and the outcomes are only about the finding of text. Proposed procedure shown to be operative on indistinct images and piercing images as well, centered on both particular and impartial assessments [1]. Text data existing in images and video comprise valuable evidence for automatic explanation, indexing, and organizing of images. Mining of this evidence includes detection, localization, stalking, mining, augmentation, and identification of the text from a given image. Conversely, differences of text due to alterations in size, style, location, and position, as well as low image divergence and complex contextual mark the tricky of automatic text extraction tremendously interesting. Though complete reviews of associated difficulties such as face recognition, document investigation, and image & video indexing can be institute; the difficult of text evidence withdrawal is not sound measured. A huge amount of practices have been projected to discourse this difficult, and the determination of this work is to categorize and evaluate these procedures, discourse standard data and concert assessment, and to point out encouraging instructions for forthcoming exploration [2]. A technique to excerpt text evidence from video structures. Primary the frequency of great horizontal liveliness in a video frame is inspected to excerpt text lumps. Structural processes are then accomplished to eradicate the background so that the text can be mined for future detection and identification. Experimentations confirm that the technique is competent and effective for mining text from numerous video forms [3]. Text discovery in a color images is an actual interesting tricky. A procedure for spotting text in images. Investigational outcomes on indoor, outdoor, captcha and Video frames imageries display that this technique can detect text characters precisely. The projected procedure expresses the bond result of the rewards of various preceding methods to find out the text, and effort on discovery the text. Investigational outcome on four dissimilar sort images demonstrate that the method based on line edge detection is sensibly healthier than the prevailing methodology [4]. An operative procedure for text mining images and video structures expending Gabor filter is projected. The projected method is finished by Gabor Filter, morphological and Heuristic filtering practice approaches is utilized to localize the text region healthier. The projected method is accomplished by text mining exploiting Gabor filter technique which is exploited for text identification within difficult images and video frames. Various experimentations were directed to evaluate the implementation of the projected intention and procedure and associate with additional approaches. Investigational outcomes experienced from a huge dataset and established that the projected technique is operative and applied [5].

Detection and Tracking of Text from Video Using MSER and SIFT

721

2 Literature Survey A procedure is proposed for each stage of mining text from a video expending java libraries and classes. The input video into stream of images utilizing the Java Media Framework (JMF) with the involvement being a real time or a video from the database. The presentation of proposed method is established by offering investigational outcomes for a set of still images [6]. Camera Based Document Analysis (CBDA) is a developing ground in computer vision and pattern recognition. In current existences, cameras are formed with numerous articles of added equipment. Proposed to extract text from camera-seized images by co-ordinate transform prototype. Later, the mined text from the changed and other manuscript images will be standard by preserving an opposite record of all letters and numbers and transformed into an editable form for instance Notepad or as a MS Word manuscript. The investigational outcomes are assessed using an innovative technique to confirm the method intended. Wide-ranging investigates have been permitted different forms and outcomes are organized. Investigational outcomes demonstrate that the efficiency of the projected technique [7]. A new text detection algorithm for biomedical images founded on iterative projection histograms. The efficiency of procedure by assessing the enactment on a set of physically categorized random biomedical images, and associate the concert compared to other up-to-the-minute text detection procedures. The proposed histogram-based text detection approach is well suited for text detection in biomedical images, and that the iterative application of the algorithm boosts performance [8]. Currently, information archives that initially enclosed pure text are fetching progressively augmented by multimedia constituents for example images, videos and audio clips. They all essential an automatic denotes to professionally index and regain multimedia constituents. If the text existences in images could be perceived, segmented, and documented inevitably, they would be a valuable source of great level semantics [9]. Now a days content-based image explanation, building and indexing of images is of prodigious significance and attention. Text appearing in images can be classified into: Artificial text (also referred to as caption text or superimposed text) and scene text (also referred to as graphics text). Artificial text is artificially overlaid on the image at a later stage (e.g. news headlines appearing on television, etc.), whereas, scene text exists naturally in the image (e.g. the name on the jersey of a player during a cricket match, etc.) [10–12]. Scene text is more difficult to extract due to skewed or varying alignment of the text, illumination, complex background and distortion. Assessment of fusion techniques are discussed and analyzed for improving the clarity of the images to increase detection accuracy [13].

722

M. Manasa Devi et al.

3 Popular Techniques for Text Extraction 3.1

Maximally Stable Extremal Regions (MSER)

It is used as a technique of blob detection in images. This method was projected to find correspondences between image essentials from two images with different perspectives. It extracts the regions which stay nearly stable through a wide range of thresholds. • Read Image frame from video • Convert the color image to grayscale • Apply different thresholds to source image to generate several binary images. For, threshold T1, all pixels with intensity less than the T1 belongs to foreground and above T1 belongs to foreground. • For each binary image, connect the white pixels of same group. These are called blobs. At T1, a pixel (p) belongs to a component (C − T1 (p)). After five gray levels, T1 belongs to (C − (T1 + 5)) (p). • Merge the binary blobs with minimum Distance which are stable. For every region, variation (v_t) is measured for every possible threshold T1. If the variation for a pixel (p) is local minimum the region will be maximally stable i.e. v_t < v{T1 + 1} • Return the center and radius of the new merged blobs. • Calculate the area of the blob by area of tightest convex shape which encloses the character [14]. Drawbacks: • Sensitive to natural lighting effects as change of day light or moving shadows. • MSER not invariant with any motion blur 3.2

Scale Invariant Feature Transform (SIFT)

It is a feature detection algorithm in computer vision to detect and describe local features in images. These features are invariant to Scale, Noise, rotation and illumination [15]. It generates scale spaces of original image frame to ensure scale invariance. Then, Laplacian of Guassian (LoG) is applied to find the interesting points in a video. Key points are obtained from interesting points by using difference of guassian(DoG). Once we get the points, all the low contrast points are removed to calculate the orientation for every point. For a given video input, • Given a center, the gradients and their magnitudes are computed in an area around the center of the descriptor given by Lxx ðrÞ þ Lyy ðrÞ • The magnitudes are weighted by a Gaussian window of size 3 * 3 neighborhood in a scale and observe for scale above and scale below to find 26 points so that a point is calculated for list of (x, y, r).

Detection and Tracking of Text from Video Using MSER and SIFT

723

• The area around the descriptor is split into 4 * 4 spatial bins. The size of each spatial bin (in pixels) is the scale of the descriptor. • For each spatial bin, a histogram of 8 orientation bins, weighted by the magnitudes is calculated. The area that is taken into account is defined by the scale of the descriptor and a magnification factor. This results in a 4 * 4 * 8 = 128-dimensional descriptor. • After the above steps, when shift is calculated, we see many circles with different sizes, it indicates sizes of vari ous scales i.e. various r values. • Different interest points are detected with different scales in order to identify the scale invariant features for which the difference of Gaussian is computed by using the below equation. @G=@r ¼ rk2 G • The above equation is popular for heat equation for guassian function. By applying the same concept we have calculated the rate of change of guassioan by using the below equation rA2 G ¼ @G=@r ¼ ðGðx; y; krÞ  Gðx; y; rÞÞ=ðkr  rÞ where; Gðx; y; krÞ  Gðx; y; rÞ ¼ ðk  1Þrk2 G here, k is scale Drawbacks: • We need to select the appropriate threshold and standard deviation values specifically to detect the text-time consuming • Computationally hard • Performance is slow • Illumination variant because of blur

4 Results and Discussions We measured the text detection accuracy as well as the two algorithms performance on ICDAR 2013-15 dataset. This dataset contains 25 videos from which we have tested on 15 videos by using MSER and SIFT. We have finalized three test cases and discussed the performance of the two algorithms in the displayed Figs. 1 and 2.

724

M. Manasa Devi et al.

(a)

(b)

(c)

Fig. 1. Results obtained from MSER for THREE test cases

(a)

(b)

(c) Fig. 2. Results obtained from SIFT for THREE test cases

Detection and Tracking of Text from Video Using MSER and SIFT

725

From the above test results we have concluded that SIFT based text detection and tracking performance is better over MSER based detection and tracking. In MSER convex hull is utilized to detect the blob for the character with the maximum enclosed area. However the convex hull couldn’t detect the blob area precisely while tracking the video. One more observation found from MSER based experimentation is that character recognition is complex when the frame is blurred and also it is facing the problem to detect the text when the text area is occluded with other objects. From the results obtained from the MSER based text detection and tracking is that accuracy is more when the text area is under stable and clear. Due to natural strengths of SIFT it is able to overcome the problems identified in MSER like invariant to scale, noise, rotation and illumination. But the problem noticed from SIFT based text detection is that we have to determine the threshold and standard deviation values precisely depends upon the type of input image. Although SIFT detects text area and text exactly it is computationally hard and performance is also slow. So, we are concluding the results here by introducing the deep learning based recurrent CNN could able to solve problems are identified in MSER and SIFT as well. Figure 3 gives the performance of the proposed algorithms in term of execution time. We have taken 10 samples in which for each sample, proposed methods are applied and evaluated for time for running the algorithm and detection of text from video. Comparatively SIFT execution time is more feasible than MSER. The performances of the two algorithms are shown in the following graph.

Fig. 3. Text detection performance comparing MSER and SIFT

Figure 4 shows the overall performance of the two algorithms in terms of accuracy. That is for 10 samples which we have tested on, MSER accuracy is 60% and SIFT is 80%. The below graph shows that SIFT is more accurate than MSER.

726

M. Manasa Devi et al.

Fig. 4. Overall accuracy of video using MSER and SIFT

5 Conclusions The text exists everywhere in our environments that can be apprehended with a tick of the mobile phone cameras. After mining and identifying, this seized text can be exploited for numerous virtuous reasons. So, the field of text extraction from scenic images seized by cameras is attaining a projecting place in the thoughts of scholars and publishers. Two popular methods; MSER and SIFT are discussed and implemented for text detection and extraction from videos. Drawbacks are also identified and can be rectified by implementing deep learning based recurrent CNN is capable from its basic features.

References 1. Chidiac N-M, Yaacoub C (2016) A robust algorithm for text extraction from images. In: IEEE international conference on telecommunications and signal processing (TSP), pp 493– 497 2. Jung K, Kim KI, Jain AK (2004) Text information extraction in images and video: a survey. Pattern Recogn 37(5):977–997 3. Yen S-H, Wang C-W, Yeh J-P, Lin M-J, Lin H-J (2018) Text extraction in video images. In: IEEE international conference on secure system integration and reliability improvement, pp 189–190 4. Kumar A, Kaushik AK, Yadav RL, Anuradhaa (2011) A robust and fast text extraction in images and video frames. In: International conference on advances in computing, communication and control 5. Kumar A (2014) An efficient approach for text extraction in images and video frames using gabor filter. Int J Comput Electr Eng 6(4):316 6. Ghorpade J, Palvankar R, Patankar A, Rathi S (2011) Extracting text from video. Signal Image Process: Int J (SIPIJ) 2(2) 7. Chethan HK, Hemantha Kumar G (2010) Image dewarping and text extraction from mobile captured distinct documents. Procedia Comput Sci 2:330–337

Detection and Tracking of Text from Video Using MSER and SIFT

727

8. Xu S, Krauthammer M (2010) New pivoting and iterative text detection algorithm for biomedical images. J Biomed Inform 43:924–931 9. Lienhart R, Wernicke A (2002) Localizing and segmenting text in images and videos. Trans Circ Syst Video Technol 12(4) 10. Wolf C, Jolion JM. Extraction and recognition of artificial text in multimedia documents. Lyon Research Center for Images and Intelligent Information Systems 11. Zhang S, Zhu C, Sin JKO, Mok PKT (1999) A novel ultrathin elevated channel lowtemperature poly-Si TFT. IEEE Electron Device Lett 20:569–571 12. Kumar S, Kumar S, Gopinath S (2012) Text extraction from images. Int J Adv Res Comput Eng Technol 1 13. Srinivasa Reddy K, Ramesh Babu Ch, Srinivasa Rao D, Gopi G (2018) Performance assessment of fuzzy and neuro fuzzy based iterative image fusion of medical images. J Theor Appl Inf Technol 9:3061–3074 14. Rabiul Islam Md, Mondal C, Kawsar Azam Md, Jannatul Islam ASMd (2016) Text detection and recognition using enhanced MSER detection and a novel OCR technique. In: 5th international conference on informatics, electronics and vision (ICIEV), pp 15–20 15. Rosenberg A (2012) Using SIFT descriptors for OCR of printed Arabic. Ph.D thesis

Blockchain Enabled Smart Learning Environment Framework G. R. Anil(B) and Salman Abdul Moiz School of Computer and Information Sciences, University of Hyderabad, Hyderabad 500046, India [email protected], [email protected]

Abstract. Smart learning provides a holistic learning to students with a major paradigm shift by using modern day technologies. The Learning Management systems are mostly built using traditional standard frameworks like that of IEEE Learning Technology Systems Architecture (LTSA). These frameworks are not suitable to smart learning environment as they lack several desired features to qualify for smartness levels in learning. In this paper a modified block chain enabled LTSA framework is proposed to realize the sense level of smartness. The proposed framework is evaluated for various security threats suitable in virtual learning. The risk impact factor for several threats is compared for several threat patterns and it is observed that risk impact factor sufficiently reduced in the proposed framework as compared to that of LTSA. Keywords: Smart learning · Blockchain · Learning Management System · Smartness levels Smart learning environment · Threats

1

·

Introduction

A web based Learning Management System (LMS) is an application that can store, deliver, track and report the teaching learning activities from anywhere, any time and from any device. A LMS can be realized in several ways. There are certain benchmark standards used by the stakeholders for the successful deployment of Learning Management Systems. Some of these standards includes IEEE Standard for Learning Technology Systems Architecture (LTSA) [4], IMS abstract framework [17], OKI (Open Knowledge Initiative), JISC (Joint Information Systems Council) [18]. IEEE Standards committee on learning technology proposed a Learning Technology Systems Architecture in 2003. However the standards were withdrawn in 2009 as there are some functional areas are not included [8], it could not be able to meet the current complex requirements. Smart learning is context aware ubiquitous learning where there is a seamless interaction between the learner and environments. ISO/IEC Joint Technical committee has proposed IT standards for Learning, Education and Training c Springer Nature Switzerland AG 2020  S. C. Satapathy et al. (Eds.): ICETE 2019, LAIS 4, pp. 728–740, 2020. https://doi.org/10.1007/978-3-030-24318-0_83

Blockchain Enabled Smart Learning Environment Framework

729

(ITLET). To achieve those requirements, new frameworks has to be developed with new dynamics [1]. As proposed by Uskov et al. [3], smart learning environments maturity can be assessed by their smartness levels. These levels are Adapt, Sense, Infer, Learn, Anticipate, Self-organize. In this paper IEEE LTSA is considered as it describes clearly each component and their functionality. However LTSA is considered as a pre-smart learning environment [1]. In order to move from Pre-SLE towards Smart SLE, the key process features at each level has to be addressed. An attempt is made to make LTSA achieve level-1 (Adapt) and level-2 (Sense) of smartness levels. This is possible by addressing three major standardization challenges viz, Security, Privacy and Data Governance. A framework which guarantees these features using blockchain technology is proposed. The threat scenarios are identified and the results are compared for LTSA-level-2 of smartness with that of IEEE Pre SLE. This paper is organized as follows: Sect. 2 deals with the Related work in the learning management systems, smart learning environments, security threats in learning environment, and blockchain structure. Section 3 discusses the proposed Blockchain enabled SLE framework and Security analysis. Section 4 deals with the Risk evaluation of LTSA and proposed framework. Conclusion and Future work is discussed in Sect. 5.

2

Related Work

There is a need for broad efficient frameworks [1] for Learning Management Systems which fulfills the standardization challenges of smart learning environment. Few of the learning management systems frameworks available in the literature includes IEEE LTSA, IMS, OKI etc. Among them, IEEE standards committee’s LTSA proposed in 2003 is well known framework. IEEE LTSA has got its attention from experts, researchers in this field due to its detailed description and simplicity. The high level LTSA is given in Fig. 1.

Fig. 1. IEEE standard for Learning Technology Systems Architecture.

Learning entity is the abstraction of human learner. Human interaction with the LMS is observed with this process. Coach is a process which communicates

730

G. R. Anil and S. A. Moiz

with the learning parameters for optimum learning experience. It communicates with all other processes with several data flows and coordinates among all other processes. Delivery is a process which collects the learning catalog and content locator from coach and learning resources and deliver the described learning content to the Learning entity. Evaluation is a process which get the learners’ response and evaluates with the data from delivery process, later it stores the assessment results in the learners’ records. However IEEE standards committee has withdrawn LTSA because of certain limitations. Few of the functional areas are not included, level of security provided for storage of records is insufficient thereby making LTSA not suitable for Smart Learning environments. IMS abstract framework was proposed by IMS Global Learning Consortium in 2003 that defines architecture for learning. OKI (Open Knowledge Initiative) is a service oriented approach for LMS architecture developed by MIT. Later, OKI was removed from the MIT archive due to its limitations. JISC (Joint Information Systems Council) framework developed by UK’s joint information systems council, which is also widely accepted by e-learning research community. These are not considered as smart learning frameworks [1,3,5]. Smart learning environments (SLEs) utilize a range of digital technologies in supporting learning, education, and training; they also provide a prominent signpost for how future learning environments might be shaped [1]. Uskov et al. [3] have defined smartness levels of Smart Learning Environments with the Standardization challenges and the Technologies involved at each stage. The SLE smartness levels are presented in Fig. 2.

Fig. 2. Driving forces for different smartness levels in SLE

A SLE can be considered as smartness level–1 (Adapt), if it can address the standardization challenges Data Governance, Security, Privacy, Systems interoperability and meeting quality criteria of smart classroom standards. SLE can

Blockchain Enabled Smart Learning Environment Framework

731

be at smartness level-2 (Sense) by addressing Data collection and storage, Data governance, Privacy, Security. Smartness level–3 (Infer) standardization challenges are Pedagogical designs, Student learner models, Student activity data and Specifying competence. Smartness level-4 (Learn) standardization challenges are Validating competence, e-assessment [16], learning design. Smartness level-5 (Anticipate) standardization challenge is to provide Predictive analysis. All the above challenges together confirms the smartness level-5 (Self-Organize). IEEE LTSA is considered as pre-smart learning environment. To achieve smartness level-2 (Sense), the standardization challenges Security, Privacy, Data governance must be addressed. The major challenge in e- learning is online examinations and assessments, as the evaluation results has a direct impact on the learning outcomes [10,12]. The Smart learning Environment should be able to provide more secure and reliable and tamper proof data generated during assessments. IEEE LTSA has a centralized database. A centralized database is always challenged against quality parameter availability. Any web-based centralized database is prone to several security breaches Data security is a global crisis [17]. Axel [15] explained that major centralized systems are hacked multiple times a year. Abrar et al. [13] has presented a taxonomy of online examination threats presented in Fig. 3.

Fig. 3. Threats Classification in remote online examinations

In order to address various threats, sophisticated data storing techniques with promising performance are needed. One of the robust data storage technique is blockchain [7]. Blockchain technology is termed as the ‘Fifth Evolution’ of computing. A blockchain is regarded as a novel approach to distributed databases. It is a disseminated database of open records which documents all transactions or advanced events that have been executed prior to that point in time when it

732

G. R. Anil and S. A. Moiz

is shared among participating members. Every transaction in the public record is confirmed by agreement by a larger number of members within that framework. A blockchain is a data structure that makes it possible to create a digital ledger of data and share it among a network of independent parties. Blockchains are secure by design and include sophisticated distributed computing systems with high Byzantine Fault Tolerance [17]. A key property of blockchain technology, which distinguishes from traditional database technology is Integrity and Transparency and Immutability, auditability, fault tolerance and trust-free operation [15]. In the case of the blockchain, modifying or deleting data is almost impossible. In this paper, blockchain enabled LTSA framework is proposed to realize the LTSA for smart learning environments.

3

Smart Learning Environment Framework for LTSA

The salient features of blockchain are integrated to basic IEEE LTSA framework. This helps in achieving the sense level (level-2) of smartness. The proposed framework is depicted in Fig. 4.

Fig. 4. Blockchain enabled Smart learning Environment Framework.

The above architecture is an extended version of IEEE LTSA that includes a component to accommodate private blockchain mechanism for storing examination transactions. The Components of the proposed frameworks includes Processes (Coach, Delivery, Learner, Evaluation) and Data Stores (Learning

Blockchain Enabled Smart Learning Environment Framework

733

Resources, Learner records, Blockchain network). The connectors includes Learning content, Learner info, history, assessment query, catalog information and blocks. The connectors specifies data couple between the various LTSA components. As online examination data is sensitive, it may be prone to several threats [11]. Further it is highly essential to record and analyze the behavioral data of the student during their interaction with the learning management systems. According to Kun Lian et al. [6], behavioral data helps to perform student profile analysis with Student Behavior Feature similarity calculation, Learning attitude analysis, Duration analysis of Online learning behavior. The proposed framework captures the behavioral analysis of a user. The difference in the behavior among the students will derive the learning patterns. It enables to predict the result of students by matching with the previous learner’s behavioral data. It leads towards more personalization [5] of online learning environment. There is a need to secure the online data so that it becomes tamper proof. Such sensitive can be stored as blocks in the modified LTSA framework. In the proposed framework, blocks store the data of a session which includes: – – – – –

User details The IP address of the system Responses recorded by the student/participant Behavioral data (Keyboard inputs, mouse movements) Pictures of the student (random per session) through web camera (optional)

For every predefined time duration (session), a hash of block is generated from each user and immediately communicated to the local network node for validation. For a time slot ti , with ‘k’ users, there would be k ∗ ti number of blocks added to the blockchain network. The duration set T is a set of timeslots of a user T = {ti |t1 , t2 , . . . , tn } For example, if there is an examination with duration 60 min and a session is considered as 5 min each, then there would be 12 such sessions with 12 (duration set) blocks being uploaded to block chain network for a particular user. The internal organization of blocks is viewed as a merkle tree [2]. A merkle tree consists of hash value of all transactions in a block. Each block has the hash value of previous block. New block is added to the tail block by placing the hash value of new block. 3.1

Flow of Online Examination in the Proposed Architecture

As private blockchain is adopted in this SLE, permissioned transactions are taken place in this system i.e. only the authenticated persons can create the blocks. Once the users are registered and approved, they are shared with the privilege to join this private blockchain. A user interacts with the learning management system with the blockchain API. Blockchain API helps in creation of blocks. The interaction of the user with the modified framework is depicted in Fig. 5.

734

G. R. Anil and S. A. Moiz

Fig. 5. User interaction mechanism with Network node and Database.

In the architecture, the Learning Entity process with the private Blockchain API records the session and put the data into a block and upload it to the network node. In the network node, each block is validated and information about the same is communicated with the other nodes in the network and placed into the database (blockchain database). 3.2

Security Analysis

The following features makes the proposed LTSA reach Sense (level-2) of Smartness levels: – Irreversibility of transactions is one of the enhanced security feature in the proposed framework. In a typical learning management system, responses recorded by student and other details are stored and committed only when the student finally exits the examination process. There may be a possibility of intruding into the transaction before the final commit. But, in blockchain enabled SLE framework, each transaction is stored at regular intervals which are irreversible. – High cost of modification: Once a block is added to the blockchain, the amount of effort required to make any modification (Proof of Work) is very high. As every block in the Merkle tree is associated with the hash values of lower level nodes, at least the entire branch of the tree need to be modified which is highly expensive in terms of resources and efforts. – Private blockchain restricts the intruders to get into the blockchain as it is permissioned. Everyone on the internet cannot know the hash functions being used and cannot generate a block. As a result the block from unknown sources cannot be validated. Thus, the security threat from outsiders is minimized in this system.

Blockchain Enabled Smart Learning Environment Framework

735

– As the IP address of each participant system including the time stamp is noted, traceability is highly assured in this system. Any attempts of inquiry or retrieval of data can be monitored and immediate actions can be taken. – Learner’s response is not directly communicated to the evaluation process. After the examination, once all responses and behavioral data is added into the blockchain, the evaluation process gets invoked, through blockchain sql queries. After evaluation, the results will be stored into the learners’ individual records and complete records will be again stored to the blockchain. This leads to efficient data and access management.

4

Quantitative Evaluation of Security in Learning Management Systems

Architecture or framework evaluation of Security can be carried out in four ways: Mathematical Modelling assessment, Simulation based evaluation, Scenario based evaluation and experience based assessment. In this paper, scenario based evaluation is used to assess the validity of the proposed framework. In Scenario based evaluation, a set of scenarios are developed that conveys the actual meaning of the requirement (Security). The proposed framework is evaluated using a well-known risk analysis model for application viz., OWASP’s Risk Rating Methodology [19,20]. Scenarios can be developed to check the security quantitatively [9–11,14]. The evaluation procedure is explained as follows. Let a set of scenarios be S = {s1 , s2 , s3 , . . . , sn }. A scenario is defined as a tuple of requirement, threats, patterns i.e si = (ri , ti , pi ) is a tuple where ri R, ti T and pi P such that R, T, and P represent security requirements, threats, and security patterns respectively. A requirement is a specific security requirement that describes the required security property of the system. A threat is a description of the threat to the system in which it explicitly or implicitly violates the requirement. Patterns, that are to be incorporated in order to mitigate the corresponding threat and safeguard the required behavior. In risk based evaluation, risk factor is computed as a function of Technical Impact (I) and Vulnerability Factors (VF). The technical impact factors and vulnerability factors and their levels of severity is presented in Tables 1 and 2 respectively.

The Technical impact (I) =

Sum of the severity of the technical factors Total number of factors considered

Isi (Technical impact of scenario ‘i’) = Vulnerability factor (VF) =

I(CIi ) + I(AIi ) + I(IIi ) + I(ACCIi ) 4

Sum of severity of the vulnerability factors Total number of factors considered

736

G. R. Anil and S. A. Moiz Table 1. Technical impact factors Confidentiality of Impact (CI)

[9] [7] [6] [1]

All data disclosed Critical data disclosed Sensitive data disclosed Non-Sensitive data disclosed

Integrity of Impact (II)

[9] [7] [6] [1]

All data totally corrupt Extensive corrupt interrupt Serious corrupt interrupt Minimal corrupt interrupt

Availability of Impact (AI)

[9] [7] [6] [1]

All services cost Extensive services interrupt Serious services interrupt Minimal service interrupt

Accountability of Impact (ACCI) [9] Completely anonymous [7] Possibly traceable [1] Fully traceable

Vulnerability factor for a scenario si is V F si =

EDi + EEi + pubi 3

The risk impact factor Rsi for the scenario si follows the standard risk equation Rsi = Isi ∗ V F si Table 2. Vulnerability factor Ease of Discovery (ED) [9] [7] [3] [1]

Automated tool available Easy Difficult Practically impossible

Ease of Exploit (EE)

[9] [5] [3] [1]

Automated tool available Easy Difficult Theoretical

Publicity (Pub)

[9] [7] [3] [1]

Public knowledge Obvious Hidden Unknown

Blockchain Enabled Smart Learning Environment Framework

737

Table 3. Risk levels of LTSA and Blockchain enabled SLE Scenario

Threat type

Pattern

LTSA

Blockchain enabled SLE framework

Technical impact factor

Vulnerability factor

Risk impact factor

Technical impact factor

Vulnerability factor

Risk impact factor

1

Interception

Tutor impersonated by an intruder (outsider)

5.4

3.6

19.44

1.25

1.67

2.075

2

Interception

Tutor impersonated by a student (insider)

5

6

30

1.25

1.66

2.075

3

Modification

Modification of responses by insider

5.25

4.33

23.7325

1.25

2.66

3.325

4

Modification

Modification of responses by an intruder

7.25

6.33

45.89

2.5

1.66

4.15

5

Fabrication

Adding a spurious transaction by insider

5.25

5.66

29.715

2.25

1.66

3.735

6

Fabrication

Adding a spurious transaction by outsider

6.25

3.66

22.875

1.25

1

1.25

7

Abetting

A third party from the same location trying to write on behalf of a student

3.25

3.66

11.895

1.25

2.33

2.915

8

Abetting

A third party from a remote location trying to write on behalf of a student

5

7

35

1.66

2.33

3.867

9

Interruption

Interruption of services by outsiders

6.5

4.33

28.145

1.25

2.33

2.915

10

Interruption

Interruption of services by insiders

6

3

18

1.25

1

1.25

Ten scenarios for Smart learning security assessment derived from literature is applied to the proposed blockchain enabled smart learning environment framework. For each threat type the risk impact factor is computed. The possible risk impacts are compared with the traditional IEEE LTSA. The results are specified in Table 3. It is clearly evident that at each scenario, the new framework proposed has a significant reduction in the risk levels. It is obtained due to the immutability nature of the sophisticated blockchain technology and the consistent challenges

738

G. R. Anil and S. A. Moiz

being faced by the centralized database of the LTSA. It is observed that threats from outsiders are very nominal as compared to threats from insiders. Impersonation in LTSA has more risk because there are several automated tools [21] available to crack the password where as it is quite complicated to get permission in private blockchain API from the network manager. Modification of transactions is widely possible in the LTSA, with the available automated tools or by other means. But in the proposed framework it is hardly possible. The entire Merkle tree needs to be modified to commit a single transaction alteration. – Even adding a spurious transaction to the database also replicates the above hardness – Interruption of services has more chances in the LTSA centralized database system, whereas blockchain is a distributed one, more availability is ensured. The pictorial representation of the quantitative results in Table 3 is as in Fig. 6.

Fig. 6. Risk factor comparison of LTSA and proposed architecture.

5

Conclusion and Future Work

Learning Technology Systems Architecture (LTSA) is a standard architecture that can be used to realize any learning management system. In this paper Blok chain enabled LTSA is proposed so that it can be used in Smart learning environments. The proposed framework guarantees secure transfer of learning records where the transactions are irreversible. Threat analysis in the proposed framework is evaluated for risk impact and compared the same with that of risk factors of traditional LTSA. It is observed that risk has considerably reduced

Blockchain Enabled Smart Learning Environment Framework

739

in the proposed LTSA thereby qualifying it for the smartness level-2. It is also observed that internal threats are more severe than that of external threats. In future attempts will be made to improve the current framework to higher level of smartness. The other frameworks can also be modified so that they are suitable for smart learning environments.

References 1. Hoel T, Mason J (2018) Standards for smart education towards a development framework. Smart Learn Environ 5:3. https://doi.org/10.1186/s40561-018-0052-3 2. Muzammal M, Qiang Q, Nasrulin B (2019) Renovating blockchain with distributed databases-an open source system. Future Genera Comput Syst 90:105–117 3. Uskov VL, Howlett RJ, Jain LC (eds) (2015) Smart Education and Smart eLearning, vol 41. Smart Innovation, Systems, and Technologies. Springer, London 4. IEEE Standard for Learning Technology-Learning Technology Systems Architecture (LTSA) (2003) IEEE Std 1484.1–2003, pp 01–97 5. Hwang G-J (2014) Definition, framework, and research issues of smart learning environments – a context-aware ubiquitous learning perspective. Smart Learn Environ 1:4 6. Liang K, Zhang Y, He Y, Zhou Y, Tan W, Li X (2017) Online behavior analysis– based student profile for intelligent e-learning. J Electr Comput Eng, Article ID 9720396 7. Cai W, Wang Z, Ernst JB, Hong Z, Feng C, Leung VCM (2018) Applications decentralized the blockchain-empowered software system. IEEE Access 6:53019– 53033. https://doi.org/10.1109/ACCESS.2018.2870644 8. Kumar P, Samadd SG, Samadd AB, Misra AK (2010) Extending IEEE LTSA elearning framework in secured SOA environment. In: 2nd international conference on education technology and computer (ICETC) 9. Adetoba BT, Awodele O, Kuyoro SO (2016) E-learning security issues and challenges: a review. J Sci Res Stud 3(5):96–100 10. Shonola SA, Joy M (2015) Security issues in E-learning and M-learning systems: a comparative analysis. In: Proceeding of 2nd WMG doctoral research and innovation conference (WMGRIC 2015) 11. Ullah A, Xiao H, Barker T (2016) A classification of threats to remote online examinations. In: IEEE 7th annual information technology, electronics and mobile communication conference (IEMCON) 12. Major Centralized Systems are Hacked Multiple Times a Year. https://medium. com/@AxelUnlimited/major-centralized-systems-are-hacked-multiple-times-ayear-9c2ad612462b 13. 10 Common Database Security Issues. https://dzone.com/articles/10-commondatabase-security-issues 14. Types of threats. https://genesisdatabase.wordpress.com/2010/12/13/types-ofthreats-interception-interruption-modification-fabrication/ 15. Makhdoom I, Abolhasan M, Abbas H, Ni W (2019) Blockchain’s adoption in IoT: the challenges and a way forward. J Netw Comput Appl 125:251–279 16. Anil GR, Moiz SA (2017) A holistic rubric for assessment of software requirements specification. In: 5th national conference on E-learning and E-learning technologies – ELELTECH

740

G. R. Anil and S. A. Moiz

17. IMS Global Learning Consortium (2003) IMS Abstract Framework: White Paper Version 1.0. https://www.imsglobal.org/af/afv1p0/imsafwhitepaperv1p0.html 18. JISC (n.d.) The E-Learning Framework. Joint Information Systems Council. http://www.elframework.org/framework.html 19. OWASP: Risk rating methodology. http://www.owasp.org. Accessed 5 Feb 2019 20. Alkussayer A, Allen WH (2010) A scenario-based framework for the security evaluation of software architecture. In: 3rd international conference on computer science and information technology 21. Beaver K (2013) Hacking for dummies, 4th edn, Paperback

Global Snapshot of a Large Wireless Sensor Network Surabhi Sharma1(&), T. P. Sharma2, and Kavitha Kadarala1 1

2

Indian Institute of Technology, Roorkee, Roorkee, India [email protected] National Institute of Technology, Hamirpur, Hamirpur, India

Abstract. A global snapshot of a distributed system reveals many facts including health of the system, integrity constraint violations and energy map etc. Specifically, in a wireless sensor network, global snapshot can be used to create an energy map of the whole network. Energy map can be defined as the aggregated information of residual energy of nodes present in the network. Such a map helps in predicting behavior of the network. Wireless sensor networks have peculiar characteristics which obstruct the use of traditional snapshot algorithm designed mainly for traditional distributed systems. Moreover, these algorithms are not scalable for large networks which have huge communication loss and message overhead. In this paper, we address this problem and formulate an algorithm for acquiring a global snapshot of large wireless network by dividing the network into concentric zones and adopting a zone wise state collection approach. Thus, with the help of concentric network topology and global snapshot, energy map is formed. We simulate this approach on Omnet++ and state network overhead and energy spent in capturing snapshot. The performance analysis of the proposed approach is done by comparing it with reliable and efficient snapshot algorithm for wireless sensor networks. Keywords: Capturing states
Energy efficiency
Energy map
Global snapshot
Wireless sensor network
Zone wise data collection

1 Introduction A distributed system comprises of spatially detached computing machines that are able to communicate with each other via messages. A state of a distributed computing machine is recognized by its event history. This consists of the history of the local actions and messages passed over communication links. The global state of a distributed system is the compilation of these properties. Global states can be used to determine protocol verification, deadlock, termination detection and discarding obsolete information. A meaningful global state has to be causally consistent [1], which can be obtained by recording states simultaneously [5]. But this situation is technically infeasible as there is no global system clock for the distributed nodes. So, a distributed snapshot is taken in a coordinated manner with the help of message passing. To generate meaningful global state (consistent global state), local states should be delivered in messages via reliable channels keeping the time frame in mind. © Springer Nature Switzerland AG 2020 S. C. Satapathy et al. (Eds.): ICETE 2019, LAIS 4, pp. 741–752, 2020. https://doi.org/10.1007/978-3-030-24318-0_84

742

S. Sharma et al.

Wireless sensor network (WSN) is a communication network with sensor nodes distributed in a region of interest. Yet, WSN cannot be considered as a traditional distributed system as shown in studies conducted [14]. Scarcity of energy resources in nodes makes it essential to have energy map of the network. Energy map [10] can be used to discover which nodes are going to die. These nodes can be avoided while choosing a route for data exchange, avoid redundant messages, and to cut on topology maintenance overhead etc. To achieve both scalability as well as energy efficiency, we propose a new snapshot algorithm which uses power transmission capability [2] of base station, dual power radio [11] of sensor nodes and features of election algorithm [4] to develop a fullyfledged energy map of the network. Proposed algorithm is named global state recording algorithm (GSRA) and we have proposed modifications to election algorithm named as zone head election algorithm (ZHEA), local state propagation (LSP) both serving different purposes. We evaluated performance of proposed algorithm in terms of messages and energy spent. Results are compared RES [18] reliable and efficient snapshot algorithms for wireless sensor networks. The paper is organised as follows. In Sect. 2, related work is surveyed. In Sect. 3, proposed algorithm (GSRA) is explained in detail. In Sect. 4, both simulation and performance results are described. In Sect. 5, conclusions and future works are mentioned.

2 Related Work Snapshot problem is not a new topic in distributed computing [17], and several algorithms have been proposed for the same [3, 6, 18]. These basically are designed for the traditional distributed algorithms where reliable channels are assumed and network is assumed to be fully connected. The main aspect of these algorithms focuses on the messages which are in transit i.e. they focus on the state of the channel more rather than the state of the node. [1, 7, 9] proposed snapshot algorithms for distributed systems assuming FIFO and NON-FIFO channels. All the mentioned algorithms focus on states of the channels i.e. the messages in the transit, but for the snapshot algorithm to work on wireless sensor network there is a need to focus on sensor states rather than the channel state. Thus, a different way is needed. [14] proposes an approach for a flat topology-based network. In the proposed approach, algorithm named DSA_WSN adaptation of Distributed Snapshot Algorithm (DSA) stated in [1] is developed. This adapted broadcast algorithm however poses a problem as the neighbours of the dead.

Global Snapshot of a Large Wireless Sensor Network

743

Thus, as an improvement over above algorithm, fault model in form of PIF (propagation of information with feedback) [13] is implemented. As far as the fault tolerance is considered, PIF algorithm still works during network fragmentation. In this case, many redundant messages are transmitted. [8] study snapshot in a totally different aspect, considering partial or historical snapshot data. This might be saving cost but is definitely not accurate. [18] propose an algorithm which has two phases: Propagation of snapshot request and collection of snapshot data. But, for a larger network the hop count increases and further increasing the number of levels. Data have to be passed through multiple paths increasing the message overhead. Compared to above mentioned algorithms, proposed algorithm is more scalable. There is no need of prior knowledge of neighbours and data propagation is done with local broadcasting. As energy is the most important aspect of a WSN, developing energy map of the nodes as they move away from the BS helps in understanding the behavioural pattern of WSN.

3 The Proposed Algorithm 3.1

Assumptions

Some assumptions included in the proposed approach: • Whole network is set up on a grid structure and BS is present in the centre of the grid. • BS has prior knowledge of number of power transmission levels or number of zones denoted by n. • Any node in the zone can start election algorithm. • Zone1 can directly communicate with base station. • Node keeps a copy of its local state saved for a small period of time. • All the nodes are stationary. 3.2

An Overview

The proposed GSRA works in three phases: The first phase is called “Initialisation”. In the Initialisation phase entire network is divided into concentric zones. BS forms an message which provides nodes with zonal identities. Along with the broadcasted , “locationBS” and “MARKER” are piggybacked. BS is considered to be equipped with a powerful antenna with powercontrolled capability [2]. It helps BS to broadcast messages by altering power transmission levels in diverse distances.

744

3.3

S. Sharma et al.

Initialisation Phase

1. BS broadcasts information < (0, 0, 0), locationBS, MARKER> at level1. 2. Nodem receiving this message set Local Statem = Energy leftm and Zonem = 1. 3. BS broadcasts information < (0, 0, 1), locationBS, MARKER> at level2. 4. Nodem receiving this information checks: • •

If Zonem ≠ null, Ignore. Else

Zonem = 2 and set Local statem = Energy leftm 5. Set Level = level3, broadcasted header (0,1,1) and go to 3. //same way whole network is divided into zones.

Once the whole network is divided into zones, ZHEA along with LSP starts in individual concentric zones for the election of zone head and propagation of local states correspondingly. BS triggers ZHEA by sending control message MARKER piggybacked alongside at various transmission levels. When nodei in zonen hears this control message, election process is started in respective zone apart from zone1 nodes. Zone1 nodes on receiving MARKER directly send their local states to BS. Rest of the zones follow the procedure mentioned below. Each node has a Snap and Accumulator flag. By default, both Snap and Accumulator flag are reset Snap flag is set whenever the node records its local state. Reset value of the same indicates the loss or non-receipt of MARKER. The nodes which have Snap flag set amalgamate the received state with its own and further broadcast to the neighbouring nodes. The nodes which propagate their local states set their Accumulator flag. The working of ZHEA is as follows: Assuming every node has a unique id in form of location coordinates. As soon as nodei receives the MARKER, it saves its local state and forms an message by adding its unique id, zone id to it. is then broadcasted locally. But message is heard by all the neighbouring nodes. To avoid these undesired messages, following heuristic approach is proposed: After forming , nodei broadcasts it to neighbouring nodes. Suppose next node in the ring structure is nodej. Thus, nodej should be the one receiving message as illustrated in Fig. 1.

Global Snapshot of a Large Wireless Sensor Network

745

Fig. 1. Filtering undesired messages

But broadcast is received by some other nodes also. To avoid unnecessary messages, each node maintains two variables called PARENT and CHILD with default value null. PARENT and CHILD variables save the location ids of left and right-side neighbours of a node in the ring structure. From Fig. 2, if nodex sends to nodeo, nodeo will update PARENT with id of nodex. When nodeo replies with an acknowledgment (ack), CHILD variable of nodex is then updated with id of nodeo. Messages received due to unnecessary broadcasting are discarded using the rules mentioned below: Rule1: When is received by a node, it checks the zone id in message with node’s zone id. If it matches, Rule 2 is applied. Otherwise, message is deleted. In the context of proposed algorithm, assume nodej and nodey belong to the same zone and hear the same election message. Assuming that nodej hears the message first. Rule2: This approach is based on the smallest distance first mechanism. Nodej will be selected as right-side neighbour of nodei as shown in Fig. 3. As soon as nodej hears , it saves id of nodei in PARENT. After this, ack is unicasted by nodej to nodei. This ack contains id of nodej. Once ack is received by nodei, it updates CHILD variable as id of nodej. When nodey sends ack, it gets no response from nodei. Nodei detects that it has already received ack from nodej. After receiving ack and updating CHILD variable, nodei unicasts its energy (saved in energy variable) along with its saved local state. Nodej after receiving energy compares it with its own energy. If value of energy in the message is greater than nodej’s energy and unique id values of both also differ, then there is no change in the and it is broadcasted further along with the aggregated local states of both nodes. But, if energy in the message is not same as that of the energy of nodej, then election message needs to be updated.

746

S. Sharma et al.

Fig. 2. PARENT CHILD variables

3.4

Fig. 3. Choosing nearest neighbor node

Algorithm ZHEA (Part1)

1. For zonei (i = 1 to m) 2. Nodej (j= 1 to n) receiving MARKER broadcasts election message. 3. Heuristic approach applied. 4. // Energy comparison starts. 5. If Energyj < Energymessage & idj ≠ idmessage. 6. Begin 7. Unchanged election message broadcasted. 8. End; 9. Elseif Energyj = Energymessage & idj ≠ idmessage. 10. Begin 11. Update idmessage = idj, rest message is broadcasted further unchanged. 12. Elseif Energyj > Energymessage & idj ≠ idmessage 13. Begin 14. Idmessage = Idj and Energymessage = Energyj. 15. End; 16. If idj == idmessage 17. Begin 18. Delete election message, move to the second part of ZHEA. 19. End;

Updating of election message is mentioned in ZHEA (part1). Zone head has aggregated local states of zone members. LSP terminates when the aggregated state message in form of an array reaches zone head, this message basically represents the energy map of zone. Same approach is applied in every zone and zone heads are elected. In addition to this, nodes also propagate a counter called NUMBER in their respective zones. NUMBER is used to count the number of nodes in each zone. Nodes’ PARENT and CHILD variables represent their left and right hand side neighbours in circular ring. After the zonal states have been calculated, third phase “Global state calculation” starts.

Global Snapshot of a Large Wireless Sensor Network

747

In 3rd phase, is unicasted by zone head to BS. This is achieved by exploiting directional antenna, dual power radio stated in works of Chan and Han [12]. BS then combines all the zonal states into one array called global state array signifying energy map of the network. 3.5

Algorithm ZHEA (Part2)

1. If idj = idmessage then nodei = zone head. 2. ZoneHeadj unicasts (idZH, zoneZH, NUMBERi, STATEi) to BS. //BS initiates reliability check.

3.6

Algorithm LSP

1. If localstatej ≠ 0 the Snap = set 2. If nodej unicasts its local state, Accumulator = set else reset 3. PARENT, CHILD variables are set to the location ids’ of left and right-side neighbours while LSP in coordination with election algorithm is executed.

3.7

Algorithm Reliability Check

1. BS check if Totalzonalstates ≠ or = NUMBER. 2. If Totalzonalstates ≠ NUMBER, then BS directly broadcast MARKER to concerned heads. 3. // LSP starts. 4. If snap = null 5. Begin = Energyj and AccummulatedLS = 6. Localstatej AccummulatedLS + Localstatej 7. End; 8. If Accumulator = null 9. Begin 10. AccummulatedLS = AccummulatedLS + LSj 11. End; 12. Zone headi unicasts AccummulatedLS to BS, and new energy map is formed.

748

3.8

S. Sharma et al.

Handling Message Losses and Ensuring Reliability

To keep a check on consistency and reliability of the network, we have introduced a mechanism to deal with loss of messages. Assuming that some nodes were not able to record their local states due to some kind of failure. When BS has a global state, it checks the number of states received against the number of nodes present in the network. If this number does not match, BS does a directed broadcasting of MARKER to zone heads of concerned zones. Now, zone heads initiate LSP in the same way, but just with a little modification. Local states of only those nodes are aggregated and propagated whose Snap or Accumulator is reset. If Snap is reset, local state is recaptured. But, if Accumulator is reset, saved states are propagated. This way states from all nodes in the network are recorded and a reliable global state is formed. Moreover, we have assumed that nodes always have enough energy to send their states, more like a residual energy to record and send their local states even if the overall energy is zero.

4 Simulation and Performance Evaluation 4.1

Simulator

An Omnet++ model consists of modules that communicate with message passing. The active modules are termed simple modules; they are written in C++, using the simulation class library. 4.2

Simulation Setup

All the sensor nodes are deployed in a grid structure with configuration parameters mentioned in Table 1. Nodes are spread randomly in concentric zones at every 100 m distance from the BS. Nodes’ density distribution need not be uniform for improving lifetime of the network as mentioned in [15]. The communication radius of a node is set to be about 15 m. Moreover, first zone formed in the network can directly communicate with BS. Apart from simulating proposed solution in the configurations mentioned in Table 1, performance analysis of proposed solution (in terms of messages exchanged) with RES [18] is done. Initially, all nodes have equal amount of energy i.e. 0.5j. Considering the case where some of the sensing operations have already taken place and nodes’ energies have decreased due to computational operations. The energy spent in transmitting messages is considered to be 50nj/bit. Size of data packet is 4000 bit. So, 20mj/packet energy is spent to transmit and receive a data message. Also, size of control message is 64 bit. Thus, 3.2mj/packet energy is consumed in transmitting and receiving control message. Variations of network sizes and corresponding number of sensor nodes employed in proposed solution are shown in table below:

Global Snapshot of a Large Wireless Sensor Network

749

Table 1. Characteristics of simulated topology S. no. # No. of sensor node Deployment area (m2) 1. 8 100 * 100 2. 20 200 * 200 3. 40 500 * 500 4. 68 1000 * 1000 5. 100 1500 * 1500 6. 150 1800 * 1800 7. 200 2400 * 2400 8. 250 3000 * 3000

4.3

Simulation Results

Here, we present the simulation results of proposed approach. Number of Messages Sent: The proposed approach uses BS to broadcast MARKER message and employs zone wise data collection approach. In RES, marker messages are forwarded till they reach leaf node and children can send their aggregated local states to their parents only. On the contrary, in the proposed approach zone head sends aggregated local states directly to BS. Figure 4 clearly shows how proposed approach stands out when compared with the other mentioned approach in terms of number of messages sent. Number of Messages Received and Processed: In the proposed approach each node is broadcasting data message only once, whereas control messages such as ack and aggregated local states are unicasted. This gives proposed approach an edge over RES, where local states of child nodes are broadcasted to parent node at each level increasing the number of messages received and processed. This difference in number of messages is shown in Fig. 6. Energy Consumed in Transmission: Energy spent in transmission is basically supposed to be proportional to messages sent. But, in addition to that one more factor affects the energy spent i.e. size of message sent. Both data and control messages consume different transmission energies. Moreover, in the proposed approach each node on average only transmits one data message and two control messages. In comparison, each node in RES transmits at least two data messages and one control message making it a more energy consuming approach. This is depicted in Fig. 5. Energy Consumed in Processing and Receiving: Reception and processing incurs energy proportional to messages received and processed. As discussed before, size factor affects energy consumption in this case also. As RES processes more data messages than GSRA, thus it consumes more energy as shown in Fig. 7. This way communication cost of RES also increases.

750

S. Sharma et al.

Figure 8 represents the energy map formed as a result of implementing GSRA. The bar code graph represents residual energy left in every node present in WSN. Proposed approach is better, formation of concentric zones reduces network overhead in a large network. As nodes under the same circular zone keep forwarding their local states in a ring formation, until an aggregated form of it reaches zone head. Rather than forwarding aggregated states at each level.

Fig. 4. Number of messages transmitted in Fig. 5. Energy spent in transmission RES and GSRA and RES GSRA

Fig. 6. Number of messages received and processed in GSRA and RES.

Fig. 7. Energy consumed in processing RES and GSRA

Same way in RES, aggregated states are forwarded by nodes at each level increasing the size of message with each level up. Thus, communications cost and energy spent also increases. Moreover, it is evident from the results obtained that proposed solution is an efficient approach in capturing global snapshots for a large wireless sensor network.

Global Snapshot of a Large Wireless Sensor Network

751

Fig. 8. Energy map

5 Conclusion and Future Work Distributed snapshot is a fundamental approach used in wireless sensor networks to form an energy map. The traditional distributed algorithms cannot be used directly in WSN. This paper uses the adapted versions of basic distributed algorithms such as distributed snapshot algorithm, election algorithm. The purpose behind using this approach is to form an efficient and cost-effective (in terms of energy spent and messages exchanged) energy map. Moreover, there’s no need of prior knowledge of neighbour nodes as messages are broadcasted locally. Energy map gives a better overview of energy left in the network. This overview can be used to determine routes for data exchange. Thus, avoids the path containing nodes with less energy. GSRA increases the lifetime and performance of the network. The proposed algorithm is better than (Greedy algorithm, Bus rapid transit system, Propagation of information with feedback, Reliable and efficient snapshot algorithm for WSN) for local state propagation in terms of network overhead and energy spent. Concentric zones help in forming a pattern indicating amount of energy left in the nodes as they move away from base station. Thus, zonal network observation can be done. If efficient data aggregation techniques are used then energy spent per data can be reduced. In future, a hierarchical relationship can be introduced. Here, one zone head acts as parent to the zone head of the outer zone and so on. Moreover, base station acts as parent for first zone. By doing this, the farthest zone spends less energy in sending messages directly to base station. Energy spent and messages exchanged can thus be further improved.

752

S. Sharma et al.

References 1. Chandy KM, Lamport L (1985) Distributed snapshots: determining global states of distributed system. In: ACM TOCS, pp 63–75 2. Curiac D (2016) Wireless sensor network security enhancement using directional antennas: state of the art and research challenges. Sensors 16(4):488 3. Garg R, Garg V, Sabharwal Y (1985) Efficient algorithms for global snapshots in large distributed systems. IEEE Trans Parallel Distrib Syst 21(5):620–630 4. Katwala H (2012) Study on election algorithm in distributed system. IOSR J Comput Eng 7 (6):34–39 5. Kshemkalyani A, Raynal M, Singhal M (1995) An introduction to snapshot algorithms in distributed computing. Distrib Syst Eng 2(4):224–233 Article id 005 6. Kshemkalyani A, Singhal M (2013) Efficient distributed snapshots in an anonymous asynchronous message-passing system. J Parallel Distrib Comput 73(5):621–629 7. Lai TH, Yang TH (1987) On distributed snapshots. Inf Process Lett 25:153–158 8. Lian J, Chen L, Naik K, Liu Y, Agnew G (2007) Gradient boundary detection for time series snapshot construction in sensor networks. IEEE Trans Parallel Distrib Syst 18(10):1462– 1475 9. Mattern F (1983) Efficient algorithms for distributed snapshots and global virtual time approximation. J Parallel Distrib Comput 18(4):423–434 10. Chan E, Han S (2009) Energy efficient residual energy monitoring in wireless sensor networks. Int J Distrib Sens Netw 5(6):748–770 11. Moon S, Han S (2010) Benefits of dual-radio wireless sensor networks. In: 7th IEEE consumer communications and networking conference (CCNC) 12. Rajamani V, Julien C (2009) Blurring sanpshots: temporal inference of missing and uncertain data. In: PerCom 10 13. Segall A (1983) Distributed network protocols. IEEE Trans Inf Theory 29:23–35 14. Silva A, Teixeira F, Lage R, Ruiz L, Loureiro A, Nogueira J (2003) Using a distributed snapshot algorithm in wireless sensor networks. In: Proceedings of the 9th IEEE workshop on future trends of distributed computing systems (FTDCS 2003), pp 31–37 15. Wang D, Cheng Y, Wang Y, Agrawal DP (2006) Lifetime enhancement of wireless sensor networks by differentiable node density deployment. In: IEEE international conference on mobile ad hoc and sensor systems (MASS), pp 546–549 16. Wong DTC, Chen Q, Chin F (2014) Directional medium access control (MAC) protocols in wireless ad hoc and sensor networks: a survey. J Sens Actuator Netw 4:67–153 17. Wu D, Cheong C, Wong M (2011) Distributed snapshots for adhocnetwork systems. Int J Parallel Emergent Distrib Syst 26(2):149–164 18. Wu W, Liu H, Wu H (2012) RES: a robust and efficient snapshot algorithm for wireless sensor networks. In: Proceedings of the 32nd international conference on distributed computing systems workshops (ICDCSW 2012), pp 231–236

Author Index

A Achayalingam, Bhuvaneshwari, 119 Afreen, Munazza, 619 Afreen, Sumayya, 700 Ahirwar, S. D., 185 Akula, Rajani, 139, 266 Alluri, Sudhakar, 281 Amjad Khan, G., 77 Anil Kumar, R., 299 Anil, G. R., 728 Anjum, Nadia, 562 Aravind, K. A., 398 Asha Rani, M., 210 Atluri, Maanasa Devi, 390 Ayyanathan, N., 545 B Badugu, Srinivasu, 562, 579, 587, 598, 619, 630 Badugu, Srivinasu, 572 Banakar, R. M., 155 Bandi, Suresh, 528 Begum, Aliya, 587 Begum, Asma, 700 Bhalke, Sangam V., 235 Bharathi, A., 226 Bhatt, Tara Dutt, 61 Bodapally, Kedarnath, 201 Boppana, Swathi Lakshmi, 29 Borkar, V. G., 10 C Chalasani, Subba Rao, 29 Chandrakanth, A., 36 Chatterjee, Samiran, 97

Chavan, Ameet, 112 Cheripelli, Ramesh, 648 D Daram, Suresh Babu, 490 Dasari, Ramakrishna, 235 Dasharatha, M., 272 Devarajula, Revathi, 390 Devi, Vimala, 424 Divya, G., 424 Doddipalli, Srinivas, 307 Doppala, Navya Namratha, 112 Durga Bhavani, K., 259 F Fathima, Amreen, 608 Fatima, Sabah, 572 Firdaus, Syeda Aliya, 658 G Ganesh, Samudrala Prashant, 520 Gugulothu, Ravi, 235 H Haran, Hari, 175 Harini, 175 Harini, V., 104 Haritha, Gorla, 97 I Inumarthy, Suma Bindu, 383 J Jacob Jayaraj, G., 528 Jahagirdar, D. R., 259

© Springer Nature Switzerland AG 2020 S. C. Satapathy et al. (Eds.): ICETE 2019, LAIS 4, pp. 753–755, 2020. https://doi.org/10.1007/978-3-030-24318-0

754

Author Index

Jaswitha, Kokku, 266 Jayalaxmi, A., 470 Jonathan, Namburi Randy, 307 Joseph Rajiv, K., 1 Jyothisree, M. V. R., 707 Jyotsna, K. A., 251

Mohiddin, Mohammad, 201 Moiz, Salman Abdul, 728 Mridula, B., 672 Mrutyunjay, Anuroop, 166 Mude, Shoban, 272 Mukkamala, Pavani, 383

K K, Lalkishore, 235 Kadarala, Kavitha, 741 Kalluru, Padma Vasavi, 383 Kamaraju, V., 398 Kamatham, Yedukondalu, 69 Karuppiah, N., 520 Kavitha, D., 320 Khasim, Sk. Md., 345 Kondrakunta, Pragnya, 166 Koppula, Sai Venkata Alekhya, 383 Korla, Devi Radha Sri Krishnaveni, 383 Kothari, Ashwin, 307 Kovvuri, Ramya Sri, 648 Kranthi Kumar, R., 36 Krishna Reddy, D., 327 Kumar, Molugu Sanjay, 266 Kumar, Tangalla Manoj, 307 Kuna, Devadas, 299 Kuruganti, Phani Rama Krishna, 29 Kusagur, Ashok, 462

N Nagaprasad, Sriramula, 693 Naik, Anant A., 235 Naqishbandi, Tawseef Ahmad, 545 Naraiah, R., 290 Narasimha, Vb., 534 Naresh, Akula, 97 Nasam, Jaya Prakash, 470 Naveen Kumar, P., 290, 320 Naveen, A., 345 Nayak, Jagannath, 194 Nirgude, Pradeep, 398 Nuzha, Ayesha, 700

L Lakshmaiah, D., 89 Latha, M. Madhavi, 218 Laxman, R., 36 Linga Swamy, R., 373 Lova Raju, K., 345 M Madhavi, B. K., 251 Madhu Krishna, K., 290 Madhu, R., 104 Madhunala, Srilatha, 130 Madhuri Ramineedi, N. V. L. H., 390 Mahalakshmi, Ch. V. S. S., 672 Mallepogu, Nagasrinivasulu, 490 Mallesham, G., 406 Mallikarjuna Rao, P., 364 Manasa Devi, M., 719 Manga, J., 89 Mangu, B., 435 Manjula, M., 335, 398 Marlene Grace Verghese, D., 528 Mittapelli, Ranjith Kumar, 451 Modale, Devesh R., 678 Mohebbanaaz, 1

P P., S. Venkataramu, 490 Paidimarry, Chandra Sekhar, 69 Paka, Priyanka, 139 Pal, Tapas Kumar, 19 Pallavi, N. B. S., 688 Pandharipande, V. M., 185 Parveen, Aqsa, 641 Patri, Sreehari Rao, 194 Peddapelli, Satish Kumar, 507 Pentapati, Hema Kumar, 147 Perumalla, Naveen Kumar, 299 Perumandla, Sadanandam, 470 Peshwe, Paritosh, 307 Pragathi, Navya, 451 Prasad, B. S. V., 19 Prasad, M. V. K. S., 194 Praveena, K., 320 Preetha, P. S., 462 Priyanka Rathod, B., 443 Puhan, Pratap Sekhar, 355 R Radhika Reddy, K., 97 Rahul, G., 44 Raja Gopal Redy, B., 520 Rajagopalan, Aravind, 678 Rajanish, N., 155 Rajendra Naik, B., 272, 281 Rajendra Prasad, M., 327 Raju, Narmala, 97 Rallapalli, Hemalatha, 119, 130, 166, 243 Rama Prasanna, Manchem, 435

Author Index Rama Rao, K., 435 Ramakrishna, D., 185 Ramavathu, Balu, 226 Rao, G. Kumaraswamy, 218 Ravali, Patturi, 97 Ravindra, K., 498 Reddy, Annapureddy Venkata, 10 Reddy, N. S. S., 272, 281 Rinku, Dhruva R., 210 Rohith Kumar, Adhimulam, 97 Rupa Devi, T., 598 S S., Jayachandra, 490 Sai, Bhagavatula Ramya, 266 Sai Kumar Reddy, Gantla, 97 Sai Pragathi, Y. V. S., 641 Sai Prasad, Desham, 89 Sai Shankar, Gunturu, 89 Sairam, M. V. S., 104 Sambasiva Rao, Kumbha, 259 Sandeep, S. D., 355 Sandhya, R., 36 Sarada Devi, M. S. N. G., 414 Saraswathi, G., 700 Sastri, P. P., 19 Satish Kumar, P., 251 Seetha, M., 719 Senthilkumar, Radha, 678 Seshapalli, Sairam, 490 Shainaz, 201 Sharma, Surabhi, 741 Sharma, T. P., 741 Shiva Krishna Prasad, A., 481 Shravani, D., 672 Shyam Kishore, G., 243 Siddaiah, P., 61 Siramdas, Sravani, 201 Sirisha, Madiraju, 1 Siva Kumar, C. H., 406 Somanatham, R., 373 Sowmya, Pabba, 89 Sravani, G., 424 Sravanthi, B., 398 Sravanti, Thota, 69

755 Sree Lakshmi, G., 424 Sreekanth, S., 707 Sreenivasa Murthy, K. E., 77 Srichandan, Kondamudi, 364 Srilatha, N., 443 Srinivasa Rao, D., 719 Sriram, Mounika, 498 Sriramula, Sai Kumar, 201 Subhani, Hajera, 630 Sujatha, 534 Sultana, Tahseen, 579 Suneetha, 175 Surya Tejaswini, G., 688 Suryanarayana, E. V., 61 Susheela, N., 481 Swamy, K. C. T., 314 Swetha, A., 52 T Tenneti, Madhu, 147 Tirumala, Satya Savithri, 119 Tharun Sai, E., 451 Thota, Surya Prakash, 507 Thotakura, Sandhya, 364 U Upadhyay, Poonam, 470 Upendranath Goud, E., 314 Usurupati, Kamal Kumar, 490 V Vadaparthi, Nagesh, 688 Vaidehi, K., 608, 658 Vaishali, K., 693 Vanidivyatha, M., 52 Venkata Ramanaiah, M., 281 Vijender Reddy, S., 335 Vinay Kumar Goud, Dubbaka, 520 Viswanada Raju, S., 719 Y Yamini, Veeramreddy, 266 Yashaswini, 175 Yaswanth Pavankalyan, K., 345 Yesuratnam, G., 414, 443