Centrality in Strategic Transportation Network Design: An application to less-than-truckload networks [1st ed.] 978-3-658-24240-4, 978-3-658-24241-1

Table of contents :
Front Matter ....Pages i-xxix
Introducing network centrality for strategic transportation network design (Anne Lange)....Pages 1-7
Transportation networks and their optimal design (Anne Lange)....Pages 9-51
Transportation network centrality (Anne Lange)....Pages 53-94
Less-than-truckload network centrality (Anne Lange)....Pages 95-118
An algorithm for less-than-truckload network design (Anne Lange)....Pages 119-151
Generated networks and their performance (Anne Lange)....Pages 153-192
Closing remarks (Anne Lange)....Pages 193-197
Back Matter ....Pages 199-237

Citation preview

Edition KWV

Anne Lange

Centrality in Strategic Transportation Network Design An application to less-than-truckload networks

Edition KWV

Die „Edition KWV“ beinhaltet hochwertige Werke aus dem Bereich der Wirtschaftswissen­ schaften. Alle Werke in der Reihe erschienen ursprünglich im Kölner Wissenschaftsverlag, dessen Programm Springer Gabler 2018 übernommen hat.

Weitere Bände in der Reihe http://www.springer.com/series/16033

Anne Lange

Centrality in Strategic Transportation Network Design An application to less-than-truckload networks

Anne Lange Wiesbaden, Germany Bis 2018 erschien der Titel im Kölner Wissenschaftsverlag, Köln Dissertation Universität zu Köln, 2010

Edition KWV ISBN 978-3-658-24240-4 ISBN 978-3-658-24241-1  (eBook) https://doi.org/10.1007/978­3­658­24241­1 Library of Congress Control Number: 2019934949 Springer Gabler © Springer Fachmedien Wiesbaden GmbH, part of Springer Nature 2011, Reprint 2019 Originally published by Kölner Wissenschaftsverlag, Köln, 2011 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, express 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 Gabler imprint is published by the registered company Springer Fachmedien Wiesbaden GmbH part of Springer Nature The registered company address is: Abraham-Lincoln-Str. 46, 65189 Wiesbaden, Germany

Foreword Industrialized societies cannot exist without powerful transport networks. The quality of such networks can be measured along many dimensions: time and cost efficiency, flexibility, and reliability, to name a few. However, in the scientific literature the design of networks is primarily discussed under the objective of cost minimization. Furthermore, it is regularly assumed that all necessary information for the optimization is available, while in real life strategic network design decisions this is hardly the case. Thus, the usual assumptions of strategic network design models do reflect real life conditions only to a limited extent. To overcome this discrepancy, strategic information about the relation between the structural and procedural shape of transport networks and their performance is essential. In this context the centrality of a network should play an important role as key element of network design. Based on these thoughts, Anne Paul formulates two objectives for her dissertation: “firstly, to conceptually devise the relationship between centrality and network performance in order to emphasize the outstanding importance of network centrality for network design, secondly, to suggest quantitative measures for transportation network centrality”. These objectives are fully achieved. Based on the precise definition of network centrality, Anne Paul links convincingly and for the first time the previously separate ways of thinking of network topology and network concentration. The centrality of transport networks is identified as overarching factor for network design. Established concepts as the exploitation of consolidation potentials or the minimization of distanceor volume-related costs can be linked by the concept of network centrality, integrating service aspects. Hereby the concept takes on an extended perspective, complementing classical, cost-oriented network design in an innovative way. Furthermore, it is shown that network centrality has two dimensions. The qualitative dimension of network centrality and its relation to performance indicators of transport networks can be used to structure the search for superior network alternatives in the design process efficiently. This concept is of a conceptual nature and of highest relevance for practical strategic network planning. The quantitative dimension of network centrality is structured by the development of indices of centrality for the general cargo market. This must be understood as a fundamental methodological contribution in the area of network analysis. All together the present thesis provides a remarkable scientific contribution and is pointing a new way for the strategic design of transport networks. By linking the concept of network centrality with classical criteria of OR for network optimization

v

Anne Paul builds a bridge between previously separate research concepts. With a wide range of applied analyses in her thesis she proves in a convincing way the fruitfulness and sustainability of her innovative approach and opens up a new, scientifically ambitious and intriguing field of research with immediate utility potential for the practical design of transport networks and far beyond. I wish this thesis a very positive reception in academia and practice and hope it will trigger an intensive discussion and further research. Werner Delfmann

vi

Preface Figuratively speaking, you are holding in your hands a travelogue. It describes the course of a doctoral voyage through the fascinating land of transportation networks. The land had steep hills and smooth plains, both traversed with equal care. I experienced hostile weather and very sunny days. And I realized there were a surprising number of dead end streets along the way. Many things are to be learned on such a trip. Traveling widens knowledge. Traveling allows one to gain new perspectives regarding one’s own territory. Traveling teaches one to take decisions. Traveling deepens the understanding of home. Inevitably, traveling is about the people that have been there before the departure and are awaiting the arrival, and about those one meets along the way. A doctoral trip must, by definition, be a solitary one. Yet, it never truly is. So many people have their share in it, and the trip would be doomed without them. Some laid the foundation for the trip. They nurtured the wish to pursue it. They taught how to walk, how to hike. They nourished the curiosity that generates the momentum necessary to depart. Others aided in the preparations for the very trip. Provisions were gathered. Maps were sketched and modified as obstacles appeared. Rest camps were prepared, facilitating the accruement of new material on the way. Helpers awaited me at those camps, sometimes unexpected, sometimes long planned. Many walked alongside for some time, some even for the entirety. I do not believe I spent a single day alone on this road. The faces changed. Some travelers took off on crossroads to pursue their own journey. Others joined at some point and continued all the way to the end. Travel companions kept up the spirit and helped to carry the baggage. Some sat by the campfire at nights, listened to the stories of the day, told their own stories, helped to modify and refine ideas. Advice was much needed and freely given along the way: Which general direction to head, which turn to take, which companion to seek. There were lively discussions that challenged and modified the endeavor. I found myself defending the entire trip itself more than once during this journey. Admittedly, I sometimes ignored advice and learned from that experience as well. In particular, I would like to thank my doctoral supervisor and mentor Professor Dr. Dr. h.c. Werner Delfmann, for being the chief-cartographer for this trip. Not sparing me the steeply rising paths, but always making sure there were a sufficient number of bridges mapped to cross the crevasses and rivers along the way. Additional thanks are due to Professor Dr. Dr. Ulrich Derigs and to Professor Ulrich W. Thonemann, Ph.D.

vii

who, as members of the doctoral committee, brought in different viewpoints, acted as final gatekeepers as well as first receivers upon arrival at my destination. Some companions are sadly missed upon the arrival, but most are here to see and to celebrate the end of this journey - far and near. Thanks to all of you. I am grateful for your preparation, your challenges, your support, and your presence. Pars pro toto: Sascha Albers, Rowena Arzt, Lisa Brekalo, Trisha Conway, Jost Daft, Björn Götsch, Ralf Günther, Caroline Heuermann, Vera Kimmeskamp, Jürgen Klenner, Kai Krause, Finn Lange, Tobias Lukowitz, Antje Möckelmann, Ralph Müßig, Christoph Paul, Edda Paul, Hans-Helmut Paul, Markus Reihlen, Ingo Reinhardt, Hilde Reuter, Jens Rühle, Heike Schwegler-Kirch, Bastian Schweiger, and Pierre Semal. A straightforward travelogue would not make a good story as such. I had to leave out many parts of the voyage and elaborate more on certain aspects. All the same, stories from the entirety of this trip are included and I am certain that each travel companion will find familiar elements of the days spent together along the route - a route that I have enjoyed every single day. Anne Paul

viii

Ganz nebenbei oder Das Derivat des Fortschritts Indes sie forschten, röntgten, filmten, funkten, entstand von selbst die köstlichste Erfindung: der Umweg als die kürzeste Verbindung zwischen zwei Punkten. Erich Kästner Kurz und bündig - Epigramme (c) Atrium Verlag, Zürich 1950 and Thomas Kästner

xi

Contents Figures

xvii

Tables

xxi

Acronyms

xxiii

Variables

xxvii

1. Introducing network centrality for strategic transportation network design

1

1.1. Emphasizing the role of network centrality in transportation network design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.2. Structure of the thesis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

3 4

2. Transportation networks and their optimal design 2.1. Transportation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.2. Transportation networks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.2.1. Network types . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.2.2. Transporting cargo or passengers . . . . . . . . . . . . . . . . . . . . . . . 2.2.3. Creating efficiencies in transportation networks . . . . . . . . . . . . 2.3. Optimal transportation network design - an Operational Research perspective . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.3.1. Graph theoretic terminology for network design . . . . . . . . . . . . 2.3.2. Strategic decisions for network design . . . . . . . . . . . . . . . . . . . . 2.3.3. Tactical decisions for network design . . . . . . . . . . . . . . . . . . . . 2.3.4. Operational decisions for network design . . . . . . . . . . . . . . . . . 2.3.5. Service-oriented aspects of OR network design . . . . . . . . . . . . . 2.3.6. Performance indicators for transportation networks . . . . . . . . . 2.4. Trade-off identification as an alternative perspective on network design 2.4.1. Empirical challenges for optimal network design . . . . . . . . . . . . 2.4.2. Trade-offs in network design . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.5. Enriching optimal network design with qualitative aspects . . . . . . . . . .

3. Transportation network centrality 3.1. Network topology and concentration as dimensions of network centrality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

9 9 13 15 20 22 27 29 30 34 39 40 42 44 44 47 50

53 53

xiii

Contents

3.2. Impact of network centrality on network performance . . . . . . . . . . . . . 3.2.1. Consolidation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.2.2. Link frequency . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.2.3. Transportation distance and time . . . . . . . . . . . . . . . . . . . . . . . 3.2.4. Schedule reliability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.2.5. Network vulnerability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.2.6. System capacity flexibility . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.2.7. Cost . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.2.8. Summarizing network centrality and its relation to network performance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.3. Centrality measures in networks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.3.1. Freeman’s centrality framework . . . . . . . . . . . . . . . . . . . . . . . . 3.3.2. Measures of inequality in social groups . . . . . . . . . . . . . . . . . . . 3.3.3. Measures of market concentration . . . . . . . . . . . . . . . . . . . . . . 3.4. Measuring network centrality of passenger airline networks . . . . . . . . . 3.4.1. Assessing the outcome of deregulation . . . . . . . . . . . . . . . . . . . 3.4.2. Metrics for the centrality of passenger airline networks . . . . . . . 3.5. The role of network centrality in strategic network design . . . . . . . . . . .

4. Less-than-truckload network centrality 4.1. Logistics service providers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.1.1. Markets for logistics service providers . . . . . . . . . . . . . . . . . . . . 4.1.2. Categorizing logistics service providers . . . . . . . . . . . . . . . . . . . 4.1.3. Research topics related to logistics service providers . . . . . . . . . 4.2. Less-than-truckload transportation . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.2.1. Less-than-truckload characteristics . . . . . . . . . . . . . . . . . . . . . 4.2.2. Less-than-truckload operations . . . . . . . . . . . . . . . . . . . . . . . . 4.2.3. European less-than-truckload market . . . . . . . . . . . . . . . . . . . . 4.2.4. Services accompanying less-than-truckload transports . . . . . . . 4.3. Measuring less-than-truckload network concentration . . . . . . . . . . . . . 4.3.1. Similarities of airline and less-than-truckload networks . . . . . . . 4.3.2. Traffic flows in less-than-truckload networks . . . . . . . . . . . . . . . 4.3.3. McShan-Windle index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.3.4. Network concentration index . . . . . . . . . . . . . . . . . . . . . . . . . . 4.3.5. Hubbing concentration index . . . . . . . . . . . . . . . . . . . . . . . . . . 4.3.6. Comparing the indices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.4. Relevance of measuring less-than-truckload network concentration for network design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

xiv

59 60 62 63 65 67 69 71 72 74 75 78 80 81 82 84 93

95 95 96 97 100 102 102 103 105 105 107 108 109 110 110 114 115 118

Contents

5. An algorithm for less-than-truckload network design 5.1. Algorithm specifications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.1.1. Stable overall transportation time in a less-than-truckload network . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.1.2. Centrality-driven network design . . . . . . . . . . . . . . . . . . . . . . . 5.1.3. High level specifications for network generation . . . . . . . . . . . . 5.2. Input and output . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.2.1. Input . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.2.2. Cost and transport time assumptions . . . . . . . . . . . . . . . . . . . . 5.2.3. Parameters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.2.4. Type of generated output . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.3. Assembling the algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.3.1. Trade-off direct vs. indirect: Separating direct and indirect traffic 5.3.2. Terminal allocation: Defining the number of hubs . . . . . . . . . . . 5.3.3. Number of transshipments: Ensuring full connectivity . . . . . . . . 5.3.4. Demand estimation: Determining the right number of vehicles . 5.3.5. Testing the generated network . . . . . . . . . . . . . . . . . . . . . . . . . 5.3.6. Possible enhancements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.4. The bigger picture: Contributions to strategic network design . . . . . . . .

6. Generated networks and their performance 6.1. Case study . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6.1.1. Input and parameters to the algorithm . . . . . . . . . . . . . . . . . . . 6.1.2. Generated results on stable transportation time . . . . . . . . . . . . 6.1.3. Generated results on cost . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6.1.4. The role of network centrality for the case study . . . . . . . . . . . . 6.2. General Insights . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6.2.1. Minimal frequency . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6.2.2. Maximal accumulating days . . . . . . . . . . . . . . . . . . . . . . . . . . . 6.2.3. Distance-related transportation cost . . . . . . . . . . . . . . . . . . . . . 6.2.4. Percentage of sold empty capacity . . . . . . . . . . . . . . . . . . . . . . 6.2.5. Daily distance by vehicle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6.2.6. The impact of the potential hubs in the network . . . . . . . . . . . . 6.2.7. Summarizing the most influential findings . . . . . . . . . . . . . . . . 6.3. Indicators of network centrality to compare networks . . . . . . . . . . . . . . 6.4. The benefit of including network centrality in the network design process

119 120 120 122 124 129 129 130 131 132 135 136 140 142 143 145 146 149

153 154 154 158 160 161 164 165 167 172 172 175 176 182 183 188

7. Closing remarks

193

A. Appendix

199

References

215

xv

Figures 1.1. Line of argument. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.1. 2.2. 2.3. 2.4. 2.5. 2.6. 2.7.

A typology of transportation networks. . Transshipment volume distribution. . . . Network types. . . . . . . . . . . . . . . . . . . . Economies of distance. . . . . . . . . . . . . . Economies of scale. . . . . . . . . . . . . . . . . Categorization of cost economies. . . . . . Situative examples of location problems.

. . . . . . .

. . . . . . .

. . . . . . .

. . . . . . .

. . . . . . .

. . . . . . .

. . . . . . .

. . . . . . .

. . . . . . .

. . . . . . .

. . . . . . .

. . . . . . .

. . . . . . .

. . . . . . .

. . . . . . .

. . . . . . .

. . . . . . .

. . . . . . .

. . . . . . .

. . . . . . .

. . . . . . .

. . . . . . .

7

. . . . . . .

16 18 20 23 24 25 31

Topological centrality scale of networks. . . . . . . . . . . . . . . . . . . . . . . . Two networks of identical topological centrality. . . . . . . . . . . . . . . . . . . Topologically central networks differing in concentration. . . . . . . . . . . Typology of network centrality. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Network centrality. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Consolidation strategies. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Average distances of networks with different degrees of topological centrality. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.8. Taxonomy of transportation network flexibility. . . . . . . . . . . . . . . . . . .

55 55 56 57 58 61

3.1. 3.2. 3.3. 3.4. 3.5. 3.6. 3.7.

64 70

4.1. 4.2. 4.3. 4.4. 4.5. 4.6. 4.7. 4.8.

Sizes of European LSP markets in bn Euro. . . . . . . . . . . . . . . . . . . . Relation of customization and functions for LSPs. . . . . . . . . . . . . . Routing schemes in LTL operations. . . . . . . . . . . . . . . . . . . . . . . . . MW index example. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . NC index example. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Example of NC index values in markets with concentrated demand. HC index example. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Examples of networks to illustrate centrality index values. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

97 99 104 111 112 113 114 116

5.1. 5.2. 5.3. 5.4. 5.5.

Impact of waiting time on overall transportation time. Algorithm: Construction process. . . . . . . . . . . . . . . . Line of identical costs in direct and indirect routing. . . Algorithm: Separating direct and indirect traffic. . . . . Algorithm: Defining the number of hubs. . . . . . . . . . .

. . . . .

. . . . .

. . . . .

121 135 138 139 141

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

xvii

Figures

5.6. Algorithm: Ensuring full connectivity in the network. . . . . . . . . . . . . . . 142 5.7. Algorithm: Determining the right number of vehicles. . . . . . . . . . . . . . 143 5.8. Overview of the developed algorithm. . . . . . . . . . . . . . . . . . . . . . . . . . 150 6.1. Shipment pay weight distribution. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6.2. Coefficient of variation of transportation time per OD-market in base case. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6.3. Coefficients of variation of transportation time per OD-market in an exemplary 1-hub scenario. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6.4. Costs in the base case scenario. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6.5. Comparison of least cost scenarios to base case. . . . . . . . . . . . . . . . . . . 6.6. Vehicle utilization against minimal frequency. . . . . . . . . . . . . . . . . . . . 6.7. Share of direct shipments against maximal accumulation time in 255 scenarios. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6.8. Coefficient of variation of indirect transportation time against maximal accumulation time in 255 scenarios. . . . . . . . . . . . . . . . . . . . . . . . . . . 6.9. Ratio of HC index / connecting share depending on the maximal accumulation time in 255 scenarios. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6.10. Handling cost at transshipment terminals against share of sold empty capacity. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6.11. Centrality indices against share of sold empty capacity. . . . . . . . . . . . . 6.12. Number of transshipment terminals connected to each EoL terminal against share of sold empty capacity. . . . . . . . . . . . . . . . . . . . . . . . . . . 6.13. Coefficient of variation of transportation times against daily distance. . . 6.14. Total cost against number of potential hubs. . . . . . . . . . . . . . . . . . . . . 6.15. Indirect transportation distance against number of potential hubs. . . . . 6.16. The impact of two distinct EoL terminals on the ratio of HC index / connecting share. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6.17. Centrality metrics against indirect vehicle utilization. . . . . . . . . . . . . . . 6.18. Centrality metrics against indirect transportation distance. . . . . . . . . . 6.19. Centrality metrics against number of arriving and departing vehicles at a terminal per indirect connection. . . . . . . . . . . . . . . . . . . . . . . . . . . . 6.20. Indirect transportation distance against cost. . . . . . . . . . . . . . . . . . . . . 6.21. Centrality metrics against total cost. . . . . . . . . . . . . . . . . . . . . . . . . . . A.1. A.2. A.3. A.4.

The Lorenz curve of a distribution. . . . . . . . . . . . . . . . . . . . . . . . . . . . Coefficient of variation of transportation times in 255 scenarios. . . . . . . Share of direct shipments against minimal frequency. . . . . . . . . . . . . . Number of connected transportation terminals per terminal against minimal frequency constraint. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . A.5. Coefficients of variation of transportation time against minimal frequency constraint. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

xviii

155 158 159 160 161 167 168 169 171 173 174 175 176 177 178 181 184 186 187 188 189 200 203 204 204 205

Figures

A.6. A.7. A.8. A.9.

Mean shipment distance against minimal frequency constraint. . . . . . . Centrality metrics against minimal frequency constraint. . . . . . . . . . . . Transportation cost in the network against share of empty capacity sold. Indirect transportation time per scenario against maximal accumulation time. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . A.10.Centrality indices against maximal accumulation time. . . . . . . . . . . . . A.11.Number of direct vehicles in the network against share of sold empty capacity. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . A.12.Centrality indices against number of hubs. . . . . . . . . . . . . . . . . . . . . . . A.13.Direct transportation against number of potential hubs. . . . . . . . . . . . . A.14.Indirect transportation times against number of potential hubs. . . . . . .

205 206 208 209 210 211 212 213 214

xix

Tables 2.1. US GDP attributed to for-hire transportation services. . . . . . . . . . . . . . 2.2. Europe of 17 revenues in transportation. . . . . . . . . . . . . . . . . . . . . . . . 2.3. Global top 25 LSPs. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

11 12 13

3.1. 3.2. 3.3. 3.4. 3.5.

72 73 86 87

Relevance of cost categories associated with network types. . . . . . . . . . Characteristic network performance depending on network centrality. . Network topology indices. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Scope of network concentration measurements. . . . . . . . . . . . . . . . . . Markets and event periods for selected studies on aviation market change. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.6. Compared networks in selected studies. . . . . . . . . . . . . . . . . . . . . . . . . 4.1. 4.2. 4.3. 4.4.

LSP literature contributions. . . . . . . . . . . . . . . . . . . . . . . . . . . . . Top 10 LSPs in European LTL 2008. . . . . . . . . . . . . . . . . . . . . . . . MW index example. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Values of concentration indices of networks depicted in figure 4.8.

. . . .

. . . .

. . . .

. . . .

91 92 101 106 110 117

5.1. Stability of transportation time: One departure on Thursday. . . . . . . . . 125 5.2. Stability of transportation time: Departures Tuesday, Thursday, Friday. . 126 5.3. Algorithm: The spread of departures over the week. . . . . . . . . . . . . . . . 145 6.1. 6.2. 6.3. 6.4. 6.5. 6.6. 6.7. 6.8.

Relevant shipment information. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Shares of initial routing schemes. . . . . . . . . . . . . . . . . . . . . . . . . . . . . Case study algorithm parameters. . . . . . . . . . . . . . . . . . . . . . . . . . . . . Transport cost at different load factors. . . . . . . . . . . . . . . . . . . . . . . . . Shipment shares by pay weight. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Centrality indices per scenario. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Connections and runs in the 5-hub example. . . . . . . . . . . . . . . . . . . . . Centrality indices of 255 generated networks in relation to maximal accumulation time. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6.9. Centrality indices of generated networks against number of potential hubs. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6.10. Indirect transportation time indicators as influenced by two explicit EoL terminals. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

155 156 156 157 162 163 166 170 179 181

xxi

Tables

A.1. A.2. A.3. A.4. A.5.

xxii

European LTL: Coverage and delivery times. . . . . . European LTL: Services. . . . . . . . . . . . . . . . . . . . Value approximation for base case scenario. . . . . KPIs against distance-related transportation cost. Regression results. . . . . . . . . . . . . . . . . . . . . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

201 202 203 207 211

Acronyms 3PL 4PL

third party logistics. fourth party logistics.

BB BC bn

break-bulk. before Christ. billion = 1,000 million.

C4 C8 CEP CLSCM CSCMP

4-firm concentration. 8-firm concentration. courier, express, parcel. closed loop supply chain management. Council of Supply Chain Management Professionals.

DSS

decision support system.

EC ECMT EoL EOQ EU

European Community. European Conference of Ministers of Transport. end-of-line. economic order quantity. European Union.

FAA FSC FTL

Federal Aviation Administration. full service carrier. full-truckload.

GDP

gross domestic product.

H&S HC HH HH&S HMH&S

hub-&-spoke. hubbing concentration. Hirschman-Herfindahl. hybrid hub-&-spoke. hybrid multi-hub-&-spoke.

xxiii

Acronyms

xxiv

ICT IT ITF

information communication technology. information technology. International Transport Forum.

JIS JIT

just-in-sequence. just-in-time.

kg KPI

kilogram. key performance indicator.

LCC LLP LSI LSP LTL

low cost carrier. lead logistics provider. logistics service intermediary. logistics service provider. less-than-truckload.

MH MH&S MW

multi-hub. multi-hub-&-spoke. McShan-Windle.

NC

network concentration.

OD OR

origin-destination. Operational Research.

P2P P&D PTL

point-to-point. pick-up and delivery. part load.

SC SCM SME

supply chain. supply chain management. small and medium size enterprise.

t tkm TSP

metric tonne. tonne-kilometer. traveling salesman problem.

UK US

United Kingdom of Great Britain and Northern Ireland. United States of America.

VRP

vehicle routing problem.

Acronyms

WLP

warehouse location problem.

xxv

Variables A α

Adjacency matrix. Detour incurred by going to a transshipment terminal.

B

Set of links to and from break-bulk terminals.

cap f ix ci ch chf ul l  χ Ci j

Capacity of a vehicle. The fixed cost associated with a node a node v i . Handling cost per kg. Handling cost per full vehicle. Eigenvector. Share of connecting passengers on the market for transport from node v i to node v j . A value associated to an edge in a network from node v i to v j , typically a cost. Cost per full vehicle km. Coefficient of variation.

ci j ckm CV di dg dij in d hu bi d i nd d i s ti j ou t d hu bi d ov e r

The weight of a node v i , typically a demand at this node. Demand for traffic class g . Demand for transport from node v i to node v j . Incoming demand for transport from hub i . Demand for indirect transports, market may be specified by index. The distance from node v i to node v j . Outgoing demand for transport to hub i . Overflow demand from direct transports, market my be specified by index.

E

Set of edges or links in a graph.

Fm i n Fπ

Minimal frequency. Frequency of service for product π in a network.

xxvii

Variables

xxviii

G g gij

Denominates a graph. Traffic class. Number of geodesics (shortest paths) between nodes v i and v j .

L l

Set of direct links. Number of available vehicles.

m

A number of nodes in a graph or of terminals in a network.

n

Number of all nodes in a graph or network, may be used for the number of airports or terminals in a network as well.

ω

Incidence relation mapping edges in a graph to two nodes i and j .

p φ π pi j p i∗j

A path in a graph. Entries in the adjacency matrix. A transportation product (or service). A path in a graph from node v i to node v j . Shortest path from node v i to node v j .

Q qi j

Overall traffic in a network. Traffic on a market for transport from node v i to node v j .

2

2-dimensional real space.

S si σ s ii n c pop si

Any arbitrary subset of V . Share of traffic at node i . Sample standard deviation. Income share of node v i . Population share of node v i .

T θ

Set of transshipment terminals. Maximum eigenvalue.

u

Average vehicle utilization.

Variables υ

Number of vehicles.

V vi v ix y vi

Set of vertices or nodes in a graph. A node in a graph or network. The x-coordinate of node v i . The y-coordinate of node v i .

x¯

Sample mean.

xxix

1. Introducing network centrality for strategic transportation network design Nothing is in the air. What an idyll. People sunbathe right in front of major airports. Barbecues without the sound of aircraft flying above. Still, some do not seem to enjoy it. Crowded trains. A never seen price jump for rental cars. The British Navy picks up travelers on the European continent. Car manufacturers run short on parts (BBC News, 2010). Video-conferencing peaks. Perishable goods do not appear on store shelves (Economist, 2010). Politicians are stuck like everyone else (Erlanger et al., 2010). Those were some European days in April 2010. Only too well, everything came back to normal quickly. It all started with a short note: the Icelandic volcano Eyjafjallajökull erupted on March 21, 2010; the largest fear of scientists being a likely breakout of the nearby-located volcano Katla, having a violent history of eruptions (Simons, 2010). About 500 people were temporarily evacuated. Journalists quote a farmer who had to leave all her animals behind (Spiegel Online, 2010a). Tourists are attracted to the site (Spiegel Online, 2010b). It was spectacular, but local. And it was petering out. Then came the second outbreak on April 14, 2010 (Devlin, 2010), causing the European airspace to be partially shut down for several days. The news were filled with the danger of flying through ash clouds. Companies started to reclaim their losses caused by the closed airspace. The airline industry stated that the disruption in flights resulted in costs of $400 million a day (Economist, 2010). Eventually, people realized how much they depend on transportation networks. The Eyjafjallajökull outbreak is a perfect example of how strongly modern societies rely on effective and efficient networks: to ensure the availability of products, to allow for mobility, to permit data exchange, to provide energy to households and production, to communicate worldwide, and so on. Transportation networks are interwoven with modern societies. The design of transportation networks has been of interest to researchers for decades. Leonhard Euler’s work on the seven bridges of Königsberg in 1736 is an early

© Springer Fachmedien Wiesbaden GmbH, part of Springer Nature 2011 A. Lange, Centrality in Strategic Transportation Network Design, Edition KWV, https://doi.org/10.1007/978-3-658-24241-1_1

1

1. Introducing network centrality for strategic transportation network design

and prominent example of research on transportation networks (West, 2001, p. 26). Transportation network design is a relevant research field in Operational Research (OR), where seminal work on optimal network design from different viewpoints is available. Consequently, it is legitimate to ask, what an optimal network is. What does the term optimal refer to? In cost-oriented OR models, an optimal network is one with minimal costs; for example resulting in minimal transportation cost. Having the above example of Eyjafjallajökull in mind, an optimal transportation network would have probably been a network without airplanes. When thinking of the Internet, the optimal network is available at all times. An optimal spider-web is a network that is structured to catch enough flies, yet not too difficult to construct, and strong enough to withstand threats. Defining optimality depends on the context; no general definition can be stated. Nature has developed networks that comply with many restrictions and are optimized to meet multiple goals. Modern artificially constructed networks must, equally, fulfill different aspects of desired performance, but we have not achieved this objective. Thus far, we can only marvel upon the efficiency that nature is able to construct suitable networks with, virtually without any central coordination (e.g. Tero et al. (2010)). There is a long way to go for modern science until this sophistication can be met. Overcoming the strong cost-orientation in traditional network design and widening the perspective to more encompassing objectives is a valuable first step. Service and risk-management have become vital aspects of the performance of modern transportation networks from a market perspective. Manifold dimensions are used to describe network service. Unfortunately, there is no common understanding regarding which indicators best describe the service expectations for a transportation network. Even so, it is understood that transportation time, the reliability of schedules, and the frequency of departures are elementary aspects of it (Pfohl, 2004b, pp. 94-100). These are among the relevant indicators for network performance in general and can be translated to network key performance indicators (KPIs). KPIs include cost and service elements but also keep track of further indicators such as risk. The wide spectrum of potential KPIs, and thereby their sheer number, creates a challenge for decision-making as it is infeasible to consider them all simultaneously. Many of the KPIs are strongly influenced by the underlying network’s appearance as expressed by spatial network centrality.1 Network centrality describes to what degree certain locations in the transportation network assume an outstanding role for the network’s operations. A star-shaped hub-&-spoke (H&S) network with an important

1

2

Centrality may have an organizational connotation (in the sense of organizational centralization) apart from its spatial dimension. This thesis does not address challenges in finding suitable organizational structures to manage a transportation network. Whenever the term centrality is used in the following, it is understood in its spatial sense.

1.1. Emphasizing the role of network centrality in transportation network design

hub in its middle is the prototype of a central network, whereas a point-to-point (P2P) network is a decentral transportation network.2 Network centrality should be seen as a core KPI for transportation networks, but is currently not perceived as such, neither in academic transportation network design, nor in decision-making on transportation networks in the logistics sector. Transportation network design will benefit from exploring the relation between network KPIs and network centrality. Network centrality enriches cost-oriented decision-making by allowing for a sound inclusion of service-orientation and further aspects of network performance into network design.

1.1. Emphasizing the role of network centrality in transportation network design The exceptional relation between centrality in transportation networks and network performance indicators is at the focus of this thesis. Network centrality influences network KPIs applied in network design and plays an exceptional, yet currently neglected role in transportation network design. Network centrality is the concept interlinking network appearance and network performance. Therefore, the objective of the thesis is twofold. Firstly, to conceptually devise the relation between network centrality and network performance in order to emphasize the outstanding importance of network centrality for network design. This aspect refers to the qualitative understanding of network centrality. Secondly, to suggest quantitative measures for transportation network centrality. These measures are necessary for analyzing and comparing in depth the centrality of transportation networks. The research field of network design is well-developed. However, its strategic application to real life problems creates challenges for planners, including e.g. the requirement for detailed data on a long-term planning horizon. To bridge this gap between research and application, the identification of trade-offs that impact network design strategically is beneficial. The knowledge of strategic influence factors will not, most likely, help to find optimal transportation networks, but rather guide the design process. Transportation network centrality may serve as a very expressive influence factor for strategic network design. Thereby, it supports the identification of good design alternatives. The general objective to emphasize the outstanding role of network centrality for transportation network design allows to formulate three subordinate goals set by this thesis.

2

Comparing the transportation distance of different networks (as one possible KPI) is a tangible example of the impact of centrality on network KPIs: a H&S network is typically characterized by longer transportation distances than a P2P network.

3

1. Introducing network centrality for strategic transportation network design

Enriching OR by generating strategic insights for ill-defined decision situations. The thesis strives to highlight that optimal network design in the way it has been conducted by researchers in OR is not sufficient to support good decision-making for strategic, long-term network design. Strategic decision situations are often characterized by high uncertainties, general assumptions, and expectations about future demand for the network being planned. At this stage, strategic insights into trade-offs involved in planning are more valuable to decision-makers than theoretically sound, optimal solutions on unreliable data. This aspect is not new in the literature, but appears to have been of lesser importance over the last few years. It is, therefore, necessary to reiterate and further enrich this line of thinking in the literature. A qualitative understanding of transportation network centrality and its relation to network performance is a significant step in this direction.

Implementing an algorithm to simulate network scenarios. The theoretically devised interdependencies between transportation network centrality and transportation network behavior will support the development of an algorithm for less-thantruckload (LTL) network design. The insights from a qualitative understanding of transportation network centrality allow to pre-structure the algorithm. Hence, the algorithm serves as an example for the practical implementation of the knowledge of the relation between network centrality and network performance.

Generating conclusions on non-measured network performance. It is challenging to assess all aspects of network performance simultaneously in a decision situation, even though simulations may be used to gather large amounts of data. It will be shown how the quantitatively measured network centrality of a network may be interpreted in order to draw conclusions on aspects of network performance that were not directly recorded. The interpretation is based on qualitative knowledge of network centrality and its relation to network performance.

1.2. Structure of the thesis The line of argument of the thesis is illustrated in figure 1.1. This first chapter outlined the background and the objective of the thesis. The second chapter, Transportation networks and their optimal design, will broadly introduce the concept of transportation network design with the goal of presenting the width of the research field, but also of explaining the key deficiencies that exist for strategic decision-making. To do so, it will give fundamental information on transportation network design before presenting an overview of the manifold problems that were discussed in OR in order to solve various aspects of network design. This overview is structured around the time horizon for decision-making and the decisions to be made at each stage. Literature reviews in the field of OR would oftentimes be structured according to the solution

4

1.2. Structure of the thesis techniques for existing problems.3 Orienting the presentation around the planning time horizon appears more appropriate for this thesis, as it directly highlights one main deficiency, namely the sequential planning approach. Most importantly, the section will point out the strong cost-orientation of OR network design approaches. This thesis can by no means claim to be the first work to identify the additional importance of general, strategic insights in OR network design. Selected contributions from this literature stream are, therefore, presented after the weaknesses of standard transportation network design and its application in practice were highlighted. The chapter closes with a claim for more general insights into strategic network design. Network centrality will prove to be a helpful concept for this purpose. The third chapter, Transportation network centrality, is devoted to presenting the outstanding role of transportation network centrality in transportation network design. The first section defines terms and will provide the reader with a fundamental understanding of network centrality. The second section, then, thoroughly discusses the impact of network centrality on six network performance indicators of transportation networks. This qualitative understanding of transportation network centrality will be exploited to structure network design later on. It is well known that several prototypes of transportation networks exist which differ in their advantages and disadvantages and are, therefore, applied in different environmental settings. Their performance is typically measured along different scales, including transport distance and time, service frequency, or schedule reliability. By relating these KPIs to network centrality, it is argued that it is the very concept of network centrality that determines the other KPIs and, therefore, is at the root of transportation network performance. In more concise terms, network centrality is the concept that discriminates networks from each other. In the forefront of these considerations, the remainder of the section is oriented to addressing the question of how the multifarious term of network centrality is understood and measured in different contexts in the literature. Research in social sciences is rich in network centrality indices and is consequently reviewed. The last remaining issue to be examined in the chapter is how indices from the social sciences were adapted to transportation networks. Airline networks provide the widest field of application and will serve as an example for transportation networks. They are even more interesting, as researchers have started to develop customized indices for the context of airline network analysis. Knowledge of existing measures of (transportation) network centrality sets the stage for developing a quantitative understanding of network centrality for LTL metrics in the subsequent chapter. The remainder of the thesis will demonstrate how transportation network centrality can be exploited for network design. To this point, the thesis will have been oriented upon transportation networks in general. From here forward, the discussion will be focused on one field of application, namely that of road-based LTL networks. Notwithstanding, the general concepts can be easily applied to other transportation 3

Examples of this type are Laporte (2009), Wieberneit (2008), or Grünert and Irnich (2005).

5

1. Introducing network centrality for strategic transportation network design

networks that possess similar characteristics. The similarities between LTL networks and passenger airline networks will explicitly be dealt with, but applications to e.g. air cargo networks, structured intermodal transportation networks, and express mail networks may also be considered with ease. Chapter 4, Less-than-truckload network centrality, encompasses two important aspects. At first, logistics service providers (LSPs) in general are presented as these are the main providers of LTL networks on the European market. The chapter then continues to illustrate the LTL business: the market, the offered services, the customer demand, and the operations. It thirdly outlines suggestions on how network centrality - more precisely network concentration - can be measured expressively for LTL networks. This is a necessary step to operationalize the conceptual claim from chapter 3 for further application and discussion in chapter 6. Chapter 5, An algorithm for less-than-truckload network design, presents an algorithm in detail that was developed to generate LTL networks in order to compare network scenarios. Qualitative transportation network centrality was a fundamental element for the development of the algorithm. It allows to identify the network appearance that best suits the given context and leads to the desired network performance. The algorithm creates and tests transportation networks of this appearance and accounts for various KPIs that may serve to compare the networks in detail. It is a practical application of the concept of network centrality and it is, therefore, valuable to present the logic behind the algorithm. The reasoning behind the algorithm, including the role of centrality, is discussed. This frames the more technical algorithm details on input and output, as well as on the detailed procedure. The main goal here is to present the way of thinking that structures the algorithm. The algorithm presentation itself merely supports this aim and will clarify the results obtained using the algorithm. The sixth chapter, Generated networks and their performance, once again focuses on two important aspects. It will firstly present, in the form of a case study, the results generated for a project that the algorithm was applied in. The project was conducted with a large LSP and aimed at identifying options for LTL network redesign on a European scale. Insights are obtained, allowing the generation of managerial implications that are relevant for the field and, in their core, related to network centrality. These are presented subsequently. Quantitatively measuring network centrality by the indices suggested in chapter 4 supports conclusions regarding typical network performance that was not in the set of measured KPIs. Lastly, the information acquired this far serves to review the benefits generated by the underlying knowledge of transportation network centrality. The conclusions stated here regarding network centrality bring us full circle to the conceptual contribution of the thesis. Chapter 7 summarizes and therein closes the thesis.

6

1.2. Structure of the thesis

Demonstrate the relevance of efficient and effective transportation networks.

Operational Research: Optimal network design.

2.2

Empirical challenges for the application. 2.3

2.4

Chapter 2

Network centrality enriches strategic decision-making concerning network design.

Definition of transportation network centrality.

Outline relation between network performance f andd network t k centrality t lit as a medium to structure network design. 3.2

Introduction to the European LTL market market.

44.11 4.2

3.1

Measure transportation p network centrality.

Suggest measures for LTL network centrality centrality.

Develop an algorithm to generate and test networks based on qualitative knowledge of network centrality. centrality

3.3 3.4

Chapter 3

4.3

Chapter 4

Chapter 5

Compare scenario KPIs. Conclusions for non-simulated KPIs provide additional perspectives for decision-making.

Chapter 6 Figure 1.1.: Line of argument.

7

2. Transportation networks and their optimal design This chapter sets the stage for the discussion of transportation network centrality in the chapters to follow. To do so, it firstly points out how transportation blends in with logistics and supply chain management (SCM), and continues by introducing transportation networks. Where do they play a role in practice? Why are they important for modern societies? It will become clear that transportation networks must be designed carefully, which leads to reviewing network design as a research field in Operational Research (OR). As network design is an immensely wide field in OR, the presentation is intended to spotlight the breadth of research questions and developed models. This suffices to illustrate shortcomings of OR approaches for real life network design. Most importantly, an overly strong cost-orientation is identified. One possibility to overcome these is to strive for general trade-offs in transportation network design. The chapter ends with a statement to the need for strategic insights for the network design process, which will be resumed by chapter 3.

2.1. Transportation This first section introduces transportation and its relevance for economies and societies. It furthermore exhibits the role of transportation in logistics and SCM. There is no uncontested terminology for logistics and SCM in academia or practice. The understanding of the two terms is of minor importance here as transportation takes an outstanding role in both concepts. In order to identify the assumed perspective, it should be pointed out that the following is based on a wide understanding of logistics: customer and service orientation, the thinking in flows, and a systemic perspective are fundamental building blocks.4 Transportation has always been con4

This thinking follows the unionist view by Halldórsson et al. (2008), or the European perspective as identified by Delfmann and Albers (2000, p. 7). The United States of America (US)-based Council of Supply Chain Management Professionals (CSCMP) defines logistics as: “the process of planning, implementing, and controlling procedures for the efficient and effective transportation and storage of goods including services, and related information from the point of origin to the point of consumption for the purpose of conforming to customer requirements. This definition includes inbound, outbound, internal, and external movements” (Council of Supply Chain Management Professionals, 2010). The European perspective on logistics differs from this one. It has evolved from a narrow viewpoint to a wide one nowadays. In the early 1980’s, logistics was limited to activities related to transporting goods or passengers and to storing goods (Berens and Delfmann, 1984, p. 32). Today, a systemic perspective

© Springer Fachmedien Wiesbaden GmbH, part of Springer Nature 2011 A. Lange, Centrality in Strategic Transportation Network Design, Edition KWV, https://doi.org/10.1007/978-3-658-24241-1_2

9

2. Transportation networks and their optimal design

sidered to be one of the key sub-systems of logistics, in one line with transshipment and warehousing / inventory management (Pfohl, 2004a, p. 8).5 Transportation is highly important for economic development. Efficient transportation systems provide economical benefits, whereas deficient systems will hinder development. Transportation directly impacts development by increasing the size of markets as more places are interconnected and, furthermore, to that it decreases travel time and product cost. Transport indirectly influences economic development by opening up competition in markets (Rodrigue et al., 2009, p. 83). It is even seen as a factor of production (Rodrigue et al., 2009, p. 88). The US generate 6% of their national gross domestic product (GDP) by transportation (Stock and Lambert, 2001, p. 312).6 For-hire transportation contributes about 3% to the US GDP (table 2.1).7 The European perspective on the importance of transportation over the years is shown in table 2.2.8 Both tables highlight that transporta-

5

6 7

8

10

is common for European logistics, emphasizing the interdependencies between different sub-systems, players and entities in networks. It further includes an orientation to customers as well as to flows (Delfmann and Albers (2000, pp. 6-7), Delfmann (2000)). Furthermore, logistics is an applied science (Delfmann et al., 2010, p. 61). This idea closely relates to what is currently understood as SCM by the CSCMP: “Supply Chain Management encompasses the planning and management of all activities involved in sourcing and procurement, conversion, and all logistics management activities. Importantly, it also includes coordination and collaboration with channel partners, which can be suppliers, intermediaries, third-party service providers, and customers. In essence, supply chain management integrates supply and demand management within and across companies. Supply Chain Management is an integrating function with primary responsibility for linking major business functions and business processes within and across companies into a cohesive and high-performing business model. It includes all of the logistics management activities noted above, as well as manufacturing operations, and it drives coordination of processes and activities with and across marketing, sales, product design, finance and information technology.” (Council of Supply Chain Management Professionals, 2010). This definition is coherent to stating that SCM has an important philosophical aspect. Philosophical SCM characteristics are the above mentioned systems approach that values the wholeness of the supply chain (SC), the strategic orientation towards cooperation in the supply chain and, finally, a strong customer focus (Mentzer et al., 2001, p. 7). The definition identifies logistics as a part of SCM. This is not uncontested: other research groups see SCM as a sub-area of logistics (e.g. Delfmann et al. (2010, p. 60)). Thus, the understandings of the connection between the terms logistics and SCM are fuzzy in research and practice (Halldórsson et al. (2008), Delfmann and Albers (2000, pp. 6-10)). However, apart from the discussion on terminology, a systemic perspective on SCs and a strong customer focus are agreed upon as constitutive elements amongst practitioners and researchers. Logistics is said to create time and place utility. Time utility mainly arises out of warehousing and storage. The value created by transportation lies in major shares in place utility. As long as a product is not available to a customer, it has a limited value. Transportation bridges spatial distances between production and consumption and, thereby, provides place utility. The speed of transport and the reliability to scheduled arrival of goods may also create time utility for customers. If goods arrive with delay at their destination, their value may be significantly decreased (Stock and Lambert, 2001, p. 313). This includes derived production in non-logistics sectors such as the automotive production. For-hire transportation is provided by common carriers, by contract carriers, and by exempt carriers. All three carrier types are legally defined for the US. They offer their transport services publically on the market for transportation, as opposed to private carriers (Stock and Lambert, 2001, pp. 321-322). Table 2.2 provides data for the so-called Europe of 17: Austria, Belgium, Denmark, Germany, Finland, France, Greece, United Kingdom, Ireland, Italy, Luxembourg, The Netherlands, Norway, Portugal, Sweden, Switzerland, Spain. Data based on internal and outbound cargo moved.

2.1. Transportation

Transportation service Air transportation Rail transportation Water transportation Truck transportation Transit and ground passenger transportation Pipeline transportation Other transportation and support activities Warehousing and storage Sum GDP % GDP

2002 48.3 26.2 7.0 95.7

2005 48.3 33.5 10.0 118.4

2007 55.2 40.5 10.7 127.6

15.7

17.9

19.3

11.5

9.5

12.0

73.4

91.6

101.5

26.8 304.6 10,469.6 2.9%

35.6 364.7 12,421.9 2.9%

40.3 407.2 13,807.5 2.9%

Table 2.1.: US GDP attributed to for-hire transportation services in current bn US $ (U.S. Department of Transportation, 2010).

tion not only supports economic development but is, furthermore, an important industry by itself. Transportation activities always involve several partners. Different classifications exist in the literature, in the given context here, the most important transport participants are the shipper, the consignee, and logistics service providers (LSPs). The shipper (or consignor) is the party that is in possession of the goods before the transport and, thereby, is the origin of transport. The destination of transports, or the receiver of goods is referred to as the consignee. These two parties generate demand for transportation (Bowersox et al., 2010, pp. 194-195). The third group of transport participants are LSPs, who carry the freight, organize the transport, or provide additional activities related to the transportation. These three parties, shipper, consignee, and LSP, form the so called logistics triad (Selviaridis and Spring, 2007, p. 137).9 The global top 25 largest LSPs 2008 (by revenues) are listed in table 2.3. The table is restricted to providers of goods transportation, omitting passenger transport. Not surprisingly, one finds many well known LSPs amongst the top 25, including diversified service providers such as Deutsche Post DHL, Deutsche Bahn, and Nippon Express, as well as courier, express, parcel (CEP) providers such as UPS, FedEx, and 9

The logistics triad is usually discussed for cargo transportation, but the transition to passenger transport can be done easily. Passengers assume the role of shippers and consignee in one party. Yet, this distinction contributes only marginally to the understanding of the situation. The challenges in cargo transport arise out of the triangular relation formed by the three parties and do not apply as such for passenger transport.

11

2. Transportation networks and their optimal design

Mode Road Rail Inland waterways Sea cargo Pipeline Air cargo Sum GDP % GDP

2002 167.5 11.9 3.8 49.0 3.0 12.1 247.4 8,959.5 2.8

2005 222.7 11.8 3.5 67.4 3.6 8.46 317.5 10,820.0 2.9

2008 277.0 13.1 3.4 72.0 4.1 10.42 380.0 12,175.0 3.1

Table 2.2.: Europe of 17 revenues in transportation in bn Euro in 2002 (Klaus, 2004, pp. 51,54), in 2005 (Klaus and Kille, 2007, pp. 50,54), and in 2008 (Klaus et al., 2009, pp. 50,54).

TNT. The important share of bulk transport service providers, rail, and ocean freight is highly influenced by the large transportation volume that is moved by these providers. Large airlines are not included in the given dataset. Even though e.g. Lufthansa reports an annual revenue of 24.8 bn Euro in 2009, respectively 22.3 bn Euro in 2008, most of these revenues are passenger-related and are therewith not relevant for table 2.3. Lufthansa Cargo with 2.0 bn Euro revenues in 2009 is too small to be included in the list by itself (Lufthansa, 2010, pp. 2-3).10 Other players participate in transportation apart from those forming the logistics triad. The government has a high interest in efficient transportation, as economies rely on it. On the other side, the government is an important infrastructure provider for many transportation modes, e.g. road. Governments, moreover, influence the transport industry through regulation and law-setting (Bowersox et al., 2010, pp. 195196). Some operative impacts on transportation exist, such as slot coordination at airports (Doganis, 1992, p. 103). Last but not least, the public impacts transportation. Firstly, people create demand for transportation indirectly by purchasing goods or directly by using public transportation. Furthermore, the public is touched on by the external effects of transportation (e.g. air pollution, noise) and will therefore ask for regulation on environmental and safety aspects of transportation (Bowersox et al., 2010, p. 196). This section provided information about the role of transportation in various contexts and the involved players. It is now necessary to focus on transportation networks. Efficient and effective transportation typically needs transportation networks; the reasons for this will be illustrated in the next section. 10

12

LSPs are introduced in more detail in chapter 4.

2.2. Transportation networks Company Deutsche Post United Parcel Service Maersk A/S FedEx Deutsche Bahn NYK Line CMA-CGM Kühne & Nagel Mitsui OSK Chinese Railway BNSF Union Pacific RZB Nippon Express MSC SNCF (incl. Géodis) CSX Norfolk Southern Indian Railway TNT Sagawa Express Neptun Orient Lines Ltd. Yamato Transport Panalpina Hanjin Shipping

Service (-s) diversified CEP, contract logistics sea freight CEP railway, diversified sea freight sea freight diversified sea freight railway railway railway railway diversified sea freight diversified railway railway railway CEP CEP, diversified sea freight diversified forwarder, diversified sea freight

Revenues 44.1 36.5 25.3 23.7 19.3 18.5 15.1 13.6 13.1 13.0 12.8 12.7 12.4 12.0 8.5 8.0 8.0 7.6 7.0 6.9 6.9 6.6 6.3 5.6 5.0

Table 2.3.: Global top 25 LSPs (Klaus et al., 2009, p. 193) by worldwide logistics revenues 2008 in bn Euro.

2.2. Transportation networks Transportation networks are a special case of networks. All networks share common elements. By its most neutral definition, a network is an abstract model of reality consisting of nodes and links relating them. Usually the nodes in the network are actors or locations such as people in social networks, airports in airline networks, or bus stops in urban transportation networks. The links constitute a relation between the actors or locations. Examples can be personal relationships in social networks, flights in airline networks, or bus routes in the urban transportation case. In transportation networks, the links are often permanent tracks e.g. rail tracks or channels, or refer to scheduled services conducted between the nodes (Rodrigue et al., 2009, p. 18).

13

2. Transportation networks and their optimal design

Various networks are discussed in the literature, including e.g. transportation networks, telecommunication networks, distributed computer networks, centralized teleprocessing networks, and energy networks (Minoux, 1989, p. 314). These networks have many similarities but differ in details. The following briefly highlights the peculiarities of different networks before common features are presented. Telecommunication and distributed computer networks are used to transport digital information packages. There is virtually no variable cost of transportation on these networks. Furthermore, the information transmitted on the network is usually replicable and it is possible to resend it if necessary. Current research questions include inter alia reliable service in case of node or link faults (e.g. Abd-El-Barr (2009), Kim and O’Kelly (2009)), quality-of-service routing (e.g. van den Berg et al. (2007), Pliakas et al. (2008), Yao et al. (2008), Skianis (2008)), and the design of ad-hoc and wireless sensor networks (e.g. the special issue in Telecommunication Systems (Woungang, 2010)). Energy networks connect the points of harvesting the energy or the points of transformation with the points of final utilization. Energy networks nowadays depend on large power plants as the center of energy generation and transmission to many distributed points of energy consumption. New energy sources such as wind turbines may lead to more decentral energy networks in the future (Bouffard and Kirschen, 2008). Transportation networks exist for various modes of transport, especially the classical ones: road, rail, air, water (inland waterways and ocean), and pipelines (Ballou, 2004, pp. 171-176). The networks are used to transport cargo, passenger, or both. Transportation networks describe the form of connecting terminals (e.g. railway stations or airports) by links operated with the transport vehicles.11 It could be drilled down much deeper on differences in these networks, but the short descriptions already show that networks differ in details but that structural similarities dominate. The design of all of these networks - including transportation networks - can initially be approached by the same toolkit mostly rooted in OR.12 It should be mentioned that the planning and design of transportation networks is not a trend of industrialized societies but has been of importance all throughout history. Their existence has been the backbone of societies, from ancient to modern times (Rodrigue et al., 2009, p. 5).13 Modern economies would not exist without efficient transportation networks. The convergence of information and telecommu11

12 13

14

Westlund distinguishes single mode transportation networks and multi-mode transportation systems. A transportation system is an elaborated transportation network in the sense that it consists “of different types of transport networks which, taken together, form a system in which the principal task of the nodes is to act as points of transshipment between the constituent primary networks” (Westlund, 1999, p. 100). This distinction is mainly definitional in nature as most challenges arising in transportation network design are not mode-specific. OR approaches to network design are presented in section 2.3. The ancient Greeks built agglomerations around the Mediterranean and the Black Sea between the eighth and sixth century BC. They connected their agglomerations mainly by a shipping network which hardly required any infrastructural investments (Westlund, 1999, p. 96). The Romans later realized that governing an enormous empire requires mobility of goods and information, and started to invest heavily

2.2. Transportation networks

nication networks and their potential for transportation networks is seen as the new milestone in this evolution (Rodrigue et al., 2009, p. 86).

2.2.1. Network types Daganzo (2005) defines four types of networks: one-to-one networks, one-to-many networks, one-to-many networks with transshipments, and lastly many-to-many networks. One-to-one network types apply e.g. for full-truckload (FTL) transportation or courier services. One-to-many networks with or without transshipment are common for production networks dominated by one production site.14 Finally, many-to-many networks are at the focus in the following; the less-than-truckload (LTL) market is a typical example for this kind of network, other examples include e.g. CEP, airline, and railway networks. Rodrigue et al. (2009, pp. 19-21) approach their overview of transportation networks by laying out a typology of transportation networks (depicted in figure 2.1) that illustrates twelve of the manifold ways to classify transportation networks; these are briefly described in the following.

(a) Transportation networks can be formulated concretely (closely related to reality) or abstractly (with a graph theoretic description).

(b) They can be classified according to their real life location as right by a coast or a river.

14

into a widespread road network, whose impact is still visible in modern times. The road network permitted trade, transport of goods from the provinces to Rome, and was needed for defending the empire (Westlund, 1999, p. 96). Another well-known infrastructural network built in Roman times are the aqueducts constructed all over the empire. With the fall of the Roman Empire, the quality of the infrastructure slowly degraded. It is known that the Mongols had an early communication network in the late 13th century. Messengers carried information through the empire, either walking or riding on horse-back depending on the message priority. Stations existed where horses were exchanged or new messengers awaited those on foot prompt at their arrival. These exchange stations and the smart interchanges conduced there allowed for a significant decrease in travel time (Kidd and Stumm, 2005, pp. 1250-1251). During the European medieval times, independent fortified castles, cities, and cloisters developed with rather little interchange by transportation amongst them at first. With the increasing importance of trade, transportation and its infrastructure also rose in importance. Italian commercial cities, the Hanseatic League as trading enterprises (Westlund, 1999, p. 97), and the Princely House of Thurn and Taxis as pioneers in postal services later on are only few examples for the rise in relevance of transportation networks. The starting colonialism by the Portuguese and the Spanish in the late 15th century and the related importance of high-sea sailing opened the door for early globalization (Westlund, 1999, p. 97). From this time on, economic development was strongly coupled with innovations in transportation. The industrial revolution in the 18th century drove, but also depended on, the simultaneous development of transportation modes with high transport capacity: inland waterways, steamships, and railways. Road transportation gained its share with industrial mass production in the early 20th century. The idea of coupling different transportation modes - including air transport - emerged around the mid-twentieth century (Rodrigue et al., 2009, p. 86). Many-to-one networks depict the mirrored situation for one-to-many networks.

15

2. Transportation networks and their optimal design

Concrete

Abstract

(a) Abstraction level

1

4 7

6 2

Railyard 3

5

4

5

(b) Relative location

Highway Secondary road Maritime

1

2

3

Port

(c) Orientation and extent

6

(d) Number of edges and nodes

Road

R il Rail

100

50

125 km

Depot

(e) Modes and terminals

90 km

(f) Distance, road type, and control of the vehicle

0.3

Contiuous

0.9

07 0.7 0.5

Divided

(g) Type of traffic

(h) Volume and direction

Hierarchical

0.7

(i) Load and capacity

Linear

Non-hierarchical

R d Random Mesh

(j) Type of correspondence

(k) Pattern

(l) Change (dynamics)

Figure 2.1.: A typology of transportation networks following Rodrigue et al. (2009, p. 21).

(c) The networks’ extent and orientation reflect their geographical coverage and the market they serve.

(d) The number of edges and nodes of the network is important for the formal model: this information is necessary for the graph theoretic translation.

(e) Networks can further be described by the modes and terminals involved in them: the terminals are represented by nodes and the modal routes by links in the model.

(f) The information about modes and terminals can be refined for a distinct mode such as road transportation, showing for example roads, speed limits, and dis-

16

2.2. Transportation networks

tances.

(g),(h) The type of traffic as well as volume and direction of traffic flows allow to further cluster transportation networks.

(i) The load of a network is defined as the ratio of transported volume to available capacity and is another characteristic to describe a transportation network.

(j),(k) Networks differ in node hierarchy as well as in the pattern (or topology) they form.

(l) Finally, Rodrigue et al. (2009) suggest the network dynamics related to new circumstances as the final criterion for their network typology. This typology is beneficial in its descriptive way to show different network types. The literature on network design addresses abstract networks and expresses the infrastructural and environmental contingencies therein. Two main types of network patterns for transportation research on many-to-many networks are typically identified: hub-&-spoke (H&S) networks and point-to-point (P2P) networks. The following presents and compares these two network patterns. Some terminology must be introduced to do so. The term (transshipment) terminal describes any node in a transportation network that is used for logistical activities. These activities include in particular transshipment operations, consolidation, or bulk-breaking. The term hub is oftentimes used to describe a terminal that acts as a “major sorting or switching center in a many-tomany distribution system” (O’Kelly and Miller, 1994, p. 32).15 The concrete definition of the term is subject to debate and naturally depends on a specific context. Most authors would agree that the outstanding share of sorting and switching activities, or transshipment operations, is what distinguishes a hub from a simple transshipment terminal.16 Accordingly, when the term hub is applied in the following, this is done

15 16

Crainic (2003, p. 456) similarly defines the term hub as a major consolidation center. Taaffe et al. (1996, p. 17) restrict their usage of the term hub to airline networks and put gateway forth as their all-embracing term. Their interpretation is not applied in the following. Alternatively, a hub may also be a terminal that is highly connected to others (e.g. Reggiani et al. (2009, p. 260)). This characteristic by itself appears less expressive for many-to-many networks than defining the term as related to sorting or switching activities. Notwithstanding, high connectivity and major switching activities will not be independent from each other. Yet, the mere existence of highly connected nodes does not imply high transshipment volumes. Button (2002) presents many perspectives of interpreting the term hub only in the context of air transportation. The McShan-Windle (MW) index will be presented in section 3.4.2 as a measure of network concentration. It measures the traffic share of the 3% largest transshipment terminals in the network. When interpreting this as a metric of hubness, it can be stated that McShan and Windle (1989) perceive as hub the 3% most important transshipment terminals in the network.

17

2. Transportation networks and their optimal design 100%

% transshipment volume

50%

0% 1%

33%

64%

95%

Transshipment terminals

Figure 2.2.: Transshipment volume distribution.

to point out that the very transshipment terminal conducts significantly more transshipment operations than most of the other terminals in the network.17 It is common for transportation networks to have some transshipment terminals that conduct significantly more transshipment operations than others. Figure 2.2 depicts the distribution of transshipped volume in an exemplary LTL network. The example shows that few terminals contribute strongly to the overall transshipment volumes in the network, e.g. in this case only 4% of the terminals transship more than 30% of the entire transshipment volume in the network. Connections in transportation networks refer to links in the network that connect terminals. In airline networks, the term (flight) leg is common (e.g. Barnhart et al. (2003a, p. 370) or Derigs et al. (2009, p. 370)), in LTL networks one would rather use the term (trade or traffic) lane for the connections defined by direct services (e.g. Keaton (1993, p. 345)). Connection is suitable for all network types. A path is a sequence of connections through the network that relates an origin to a destination. Again, the airline expression for a path is a route flown by an aircraft or traveled by a passenger (Ball et al., 2007, p. 42). Path is the embracing term used for all modes. The Oxford English Dictionary defines vehicle as “any means of carriage, conveyance, or transport; a receptacle in which anything is placed in order to be moved” (Oxford English Dictionary, 2010). It is an umbrella term for all transportation equipment. Commonly used for road and rail transportation, it will also denote any other equipment such as aircraft and vessels that serve to hold and transport goods or passengers in discrete loads. The two prototypes of networks that are contrasted in most literature contributions on transportation networks are P2P networks (also referred to as fully-meshed or grid17

18

The term hub appears in hub-&-spoke (H&S) network. Originally, this stems from the idea, that all H&S networks have one or very few such hubs. Transportation networks with many transshipment terminals are commonly referred to as multi-hub-&-spoke (MH&S) networks as they are seen as an extension of H&S networks, but here, the outstanding role of some transshipment terminals may be contested. Even so, the terms H&S network and MH&S network are used to describe these networks in the literature and in practice, whether the transshipment terminals actually have an outstanding amount of traffic or not.

2.2. Transportation networks

networks) and H&S networks (e.g. Hall (1987b), Bryan and O’Kelly (1999), Lumsden and Dallari (1999), Mayer (2001), Jara-Díaz and Basso (2003), Alderighi et al. (2007), Derudder and Witlox (2009)). Figures 2.3(a) and 2.3(b) respectively depict these two. In a P2P network, the nodes are interconnected with direct links. No transshipment is necessary as direct transports between the nodes are possible. In a single-H&S network, one of the nodes assumes the central role of a major transshipment terminal, a hub.18 When all nodes are connected only to the hub (the startype network), it is sometimes also referred to as a 1-hub-system (Mayer, 2001, p. 11). No additional links exist in this most pure H&S network. All traffic flows between terminals (apart from flows originating or terminating at the hub) are channeled through the hub. One distinction must be pointed out here as it will become important in the following. From a topological point of view, the network depicted in figure 2.3(b) is a star-shaped network. The dotted line around the central node intends to indicate it as a hub. A star-shaped network with a hub at its center is a H&S network. However, not all star-shaped networks are H&S networks: if no transshipment operations take place at the central node, the network is not a H&S network. This difference is sometimes overlooked in the literature.19 Burghouwt (2007, p. 257) lists definitions of H&S networks and by far most accept the notion of transshipment operations at the central node a constitutive element in the definition. In practice, most star-shaped networks are H&S networks.20 Even though this is a lookahead on chapter 3 and will be lined out in detail there, it is worth noting that the definition of a H&S network has two components: the topological star-shaped network and outstandingly strong traffic flows at the central node. The cost elements of H&S networks and P2P networks differ as the operations are not the same. Jara-Díaz and Basso highlight that the costs assessed in a study of transportation network costs should always include a flow-oriented term21 to assess the handling operations at the terminal as well as a flow-distance-oriented term22 to account for costs incurred in the transport (Jara-Díaz and Basso, 2003, p. 286). Based on this, the key cost differences between the two networks can be identified. The networks differ in the importance of flow- or flow-distance-oriented costs. H&S networks require transshipment operations, whereas P2P networks do not. The transported 18

19

20

21 22

It is also possible to introduce an additional hub node into the network. The difference between the two cases is marginal. For the context here, however, it is more convenient to assume that the hub is one of the nodes in the network and not explicitly added for the transshipment purpose. For the case of telecommunication networks, Kim and O’Kelly (2009, p. 285) point out that hubs are often reduced to an important node out of a number of nodes, neglecting the importance of switching operations. Then again, it is known that some airlines, especially regional airlines or low cost carriers (LCCs) (e.g. Hawaiian Airlines (Hawaiian Airlines, 2010), Wizz Air (Wizz Air, 2010), or SmartWings (SmartWings, 2010)) operate star shaped networks with only very few connecting passengers at the central node. These networks are no H&S networks. The flow-oriented term is measured e.g. in weight such as metric tonnes (t). The flow-distance-oriented term is measured commonly in tonne-kilometers (tkm).

19

2. Transportation networks and their optimal design

(a) Point-to-point-network.

(b) Hub-&-spoke-network.

(c) Hybrid multi-hub-&-spokenetwork.

Figure 2.3.: Network types.

distances will be longer in H&S networks than in P2P networks as channeling the traffic through the hub implies that a detour can be expected for passing there, which increases distance-related costs. Whether flow or flow-distance costs are higher in any network type depends on the given situation. Many different network types are derived out of the two network prototypes. Well known are multi-hub (MH) networks and hybrid hub-&-spoke (HH&S) networks. MH networks are an extension of H&S networks in that they possess several hubs. The hubs are interconnected in a fully-meshed network, but the other terminals will be allocated to one of the defined hubs. In other words, MH networks are several distinct H&S networks that are connected by a fully-meshed network only containing the hubs (Mayer, 2001, pp. 12-13). Another mixed form of networks are HH&S networks. They are based on a H&S network but include “short cuts” (Lumsden and Dallari, 1999, p. 54) between terminals so that it becomes possible to bypass the hub if necessary. It seems rather uncommon to set up a HH&S network with a 1-hub-system at its base, so hybrid multi-hub-&spoke (HMH&S) networks are commonly found in practice, e.g. airline networks and LTL networks are typically structured accordingly. Figure 2.3(c) sketches a HMH&S network.

2.2.2. Transporting cargo or passengers It is important to be aware of what is transported in a network when it is designed: cargo or passengers. There are differences in expectations and feasibility for transportation operations. Transported objects is the umbrella term that covers all goods and passengers that are transported in transportation networks. In the context of transportation, goods are also referred to as cargo or freight which is used synonymously here.23 A shipment 23

20

The literature as well as practice seem to use the term cargo more frequently for air and rail transportation whereas freight is more common for road based transports. As this distinction is not generally agreed upon, it is not uncommon to use the two terms synonymously.

2.2. Transportation networks

describes a defined bundle of goods, possibly several pieces that belong together, and that is transported on behalf of one shipper. By definition, transportation deals with the two categories of transported objects, cargo and passengers. By far most modes of transport are able to handle both, either by actually transporting them on the same vehicle (such as transporting air cargo in the belly of passenger aircraft (Doganis, 2010, p. 21)) or with distinct vehicles (such as cargo trains and passenger trains). The general concepts are the same for both categories of transported goods. The demand by the two transport objects is, however, different and will influence the actual operations. It might be seen as a trend that for certain subsets of cargo and passenger, the demand converges almost to the same. For example, express letter delivery shows certain characteristics that are usually related to passenger transport. Similarly, the demand fulfilled by low cost airlines resembles that of cargo transportation in some aspects. Nonetheless, typical demand patterns for passenger and cargo transportation may be distinguished (Rodrigue et al., 2009, pp. 151-152).

Speed of transport. In general, passengers value the speed of travel highly, as traveling by itself does not fulfill a purpose; the service offered is to arrive at the destination.24 Apart from express shipments and mail, cargo is less sensitive to the speed of transport.25

Frequency of service. Passengers have a higher demand concerning the frequency of service than cargo does. Especially in urban transport, frequency by itself is a goal for network design. In certain contexts, frequency is of relevance for cargo as well, yet, the time intervals between two consecutive services will typically not be as short for cargo as for passenger transport.26

Schedule reliability. If schedules are set up, passengers expect transports to meet those schedules. Delays are unacceptable. Cargo is less sensitive to that. It must, nevertheless, be mentioned that in certain supply chain contexts, schedule reliability will be crucial as well. For instance, just-in-time (JIT) and justin-sequence (JIS) is infeasible without a high reliability of the defined arrival of goods.

Peaks in demand. Passenger transport demand has high peaks during the day, especially at commuter times in the morning and the late afternoon. Seasonal peaks around holidays are perceived as well. Cargo demand is more evenly 24 25

26

Exceptions do certainly apply, especially when thinking about cruises or scenic drives in tourism. Various time components are commonly included within the planning objectives for urban passenger transportation (e.g. Wirasinghe and Vandebona (2010)). Continuous transportation modes exist for passengers and for cargo for specific applications: e.g. escalators for people, conveyor belts in production processes for goods, or pipeline transports for cargo in terms of liquids and gas. Thus, for specific contexts, the desired frequency of service for passenger and cargo converges.

21

2. Transportation networks and their optimal design

split throughout the day, mainly depending on production cycles and opening hours of businesses. Nonetheless, exceptional high demand is experienced on a seasonal level (e.g. highs before Christmas and lows over the summer), on a repetitive basis (e.g. at the end of the month, related to so called MRP-jitters (Lee et al., 1997, p. 554)), and on a weekly schedule (peaks towards the weekend (Humphrey et al., 2007)).27

Traffic balances. Passenger traffic flows are usually assumed to be balanced, as most passengers return home after their trip. This is a major difference when compared to cargo, as such flows are characterized by high imbalances of traffic.28

Routing options. Passengers are free to choose any routing through a transportation network and will have many personal reasons to prefer one routing over another. Cargo does not have these individual preferences and will usually only demand transportation from origin to destination without specifying the routing (Derigs et al., 2009, p. 371). Hence, it can be routed following a planning by the transport provider.

2.2.3. Creating efficiencies in transportation networks Efficient many-to-many transport operations rely on transportation networks. Several triggers allow to create cost efficiencies in transportation networks. It may not be possible to exploit all of them simultaneously in a given situation. Still, it is beneficial to be aware of the typical efficiencies that may be generated in transportation networks. Economic research has long since discussed the cost impacts of production process characteristics. The focus lies on identifying the relation of input and output to a production process. Efficiencies exist if the output increases more than the required input or if the same output can be achieved with lower input. The same can be done for the special production process of transporting goods through a transportation network. The input to produce transportation on a network is the network itself (Jara-Díaz and Basso, 2003, p. 274). This includes inter alia the number and location of nodes in the network, the routes between the nodes, the route frequency, the route structure, and the transportation distance. This input induces costs for the network operator. The output of the process are the transports that are conducted. A product in this context is an origin-destination (OD)-connection of a certain service class. A service class is e.g. a type of transported object. Since distinct OD-transports are different 27

28

22

For the case of time-sensitive cargo transports, daily peaks may occur. For instance, (express) mail transport experiences high traffic demand at night-time when sorting operations must be carried out during a short interval of time to allow for fast transport to the destinations afterward (Crew et al., 1990). In that sense, specific cargo segments may experience similar constraints as for passenger transport. Doganis (2010, p. 298) stresses this aspect for the comparison of air passenger and air cargo traffic.

cost

2.2. Transportation networks

distance

Figure 2.4.: Economies of distance.

products, any many-to-many network by itself offers multiple products. Disaggregating all products in a transportation network is unfortunately rarely feasible in practice (Pels and Rietveld, 2000, p. 322).29 It is nevertheless beneficial to disaggregate outputs to highlight different efficiencies that can be created in transportation in the following. An often stated research question in economic research is whether or not certain economies exist in certain production or transportation situations. Furthermore, cost function analysis has been looked at broadly.30 The following sections are intended to highlight the main categories of economies that are at the root of efficiencies in transportation networks and give some high-level reasoning in what situations they are likely to prevail.

Transport-related economies Certain sources for cost economies are particular to transportation processes. First of all, transportation cost is highly impacted by characteristics of the transported object such as stowability and easy handling (Bowersox et al., 2010, pp. 219-221). Economies of distance and of vehicle size are, furthermore, specific for the context of transportation.

Economies of distance Economies of distance exist, if the transportation cost per unit of distance decreases as the transport distance increases. The reasoning for their existence is straightforward: the fixed cost associated with a transport is distributed over a longer distance and the cost per unit of distance decreases. Figure 2.4 shows the generalized relation between distance and cost. The term tapering principle is usually associated with economies of distance in transportation (Bowersox et al., 2010, p. 219). 29

30

The most popular aggregate measurement of the output is tkm, even though its significance is contested in the literature (Jara-Díaz and Basso, 2003, p. 273). It is, however, most commonly used in the industry and will be easy to understand for practitioners. See Pels and Rietveld (2000) for an overview of cost functions in transportation.

23

cost

2. Transportation networks and their optimal design

shipment weight

Figure 2.5.: Economies of scale.

Economies of vehicle size If the cost for increasing vehicle size is proportionally less than the gain in transport capacity, the situation exhibits economies of vehicle size. Morrison and Winston (1985, p. 59) note that most transport industries may exploit economies of vehicle size. The special case of economies of aircraft size certainly is the most prominent one in the literature,31 even though the same arguments apply for any other transport vehicle. Economies of vehicle size are “crucial for the existence of economies of traffic density” (Basso and Jara-Díaz, 2006b, p. 158). Only because an increase in capacity of a vehicle will not induce the same increase neither in purchasing nor in operating costs will transport providers have the possibility to exploit economies of traffic density.

Economies of scale Economies of scale exist in many contexts and are well known in production research. They refer to the long run development of average total cost. Economies of scale in production exist, if “long run average total cost falls as the quantity of output increases” (Mankiw, 2001, p. 284). The very concept has been translated to transportation networks and is “the concept used in applied research to study issues such as ownership, regulatory reform, the scope for competition, or postderegulation market structures” (Basso and Jara-Díaz, 2006a, p. 260). Economies of scale in transportation are generally defined as the decrease of cost per unit as the size (or volume, weight) of the shipment increases (e.g. Bowersox et al. (2010, p. 194)), as depicted in figure 2.5. The input dimension is the weight of the shipments, but may as well be in terms of volume. The increase in efficiency is mostly related to the fixed cost of transportation being distributed to a larger weight of shipments. The fixed cost includes e.g. time to position a vehicle to load and unload, invoicing, and cost of equipment. Economies of scale can be looked at in more detail. The size of a transportation network, being the input to the system, is understood as the number of nodes n in the network (Oum and Waters, 1996, p. 430). The OD-dimension of the product of a trans31

24

The interested reader is referred to Wei and Hansen (2003) for an overview.

2.2. Transportation networks

Network ssize N

Route structure

fi fix variable

fix

variable

Economies of density

Economies of scale with ith fixed fi d network t k size

Economies of (spatial) scope (or economies of network size)

Figure 2.6.: Categorization of cost economies following Basso and Jara-Díaz (2006b, p. 152).

portation network is not independent of the network size. Accordingly, economies of scale in transportation networks are often seen as a mixture of the impact on cost of traffic density and network size (Oum and Waters, 1996, p. 429). Moreover, it is highly probable that with an increase in traffic, the routing through the network will tend more to direct routings bypassing central transshipment nodes. For the context of transportation networks, the concept of economies of scale usually assumes a variable network size and, thus, “studies both product and network growth” (Basso and Jara-Díaz, 2006a, p. 259). Basso and Jara-Díaz (2006b) split the concept of economies of scale in transportation networks based on the observation that route structure and network size impact the output of transportation processes. The authors distinguish between (i) economies of density, (ii) economies of scale with fixed network size, and (iii) economies of scope as sketched in figure 2.6.

Economies of density Economies of traffic density refer to “the impact on average cost of expanding all traffic, holding network size constant” (Oum and Waters, 1996, p. 429), which is the same as saying that they show “the unit cost implications of a change in the volumes transported in a given network” (Pels and Rietveld, 2000, p. 323). Especially economies of vehicle size are important for their existence in transportation networks (Basso and Jara-Díaz, 2006b, p. 153). Basso and Jara-Díaz (2006b, p. 151) point out that in order to be able to interpret returns on traffic density, the routing of the network has to be kept at a fixed level.32 From a strategic point of view, economies of traffic density can be expected to exist in most transportation networks. More traffic on existing network arcs allows inter alia to achieve higher fill rates on the transport vehicles used. Additional shipments can be transported with hardly any additional cost.

32

The changes in traffic routing must not be included into the assessment of economies of traffic density. If the possible modification in the traffic routing is to be taken into account, the authors suggest the measure of economies of scale with fixed network size (Basso and Jara-Díaz, 2006b, p. 160).

25

2. Transportation networks and their optimal design

Economies of scale with fixed network size Economies of scale with fixed network size are used to assess returns to scale when the network size is kept constant but route structure may change. They “are related with the convenience or inconvenience of expanding proportionally the flows in all OD-pairs”(Jara-Díaz and Basso, 2003, p. 283). A proportional increase of flows in a transportation network will induce a shift from a H&S network towards a P2P network since the need to consolidate flows decreases. This situation is typical for the transportation context: “The need to make a decision on a route structure is, finally, a consequence of the spatial dimension of [the] product” (Jara-Díaz and Basso, 2003, p. 274).

Economies of (spatial) scope Economies of scope are “the impact on cost of adding new outputs to the line of production” (Jara-Díaz and Basso, 2003, p. 283). They “exist when it is cheaper to produce two or more outputs jointly by a single firm than producing each of them separately by an independent firm” (Oum and Waters, 1996, p. 435). As mentioned above, in transportation networks the products typically are OD-connections and different types of transported objects. And so, when economies of scope are discussed in the literature on transportation networks, one usually refers to situations where cargo and passengers are transported on a single vehicle or to situations where new OD-connections are offered. Whereas the former may be treated as a special case of economies of density, the latter cannot be discussed without modifying network size. The notion of economies of spatial scope explicitly refers to adding new OD-connections. Making this distinction allows to analyze correctly the advantageousness of network expansion and, thereby, supports economic research (Jara-Díaz and Basso, 2003, p. 286). Moreover, being aware of the distinction between the different cost economies allows a better understanding of the different elements of potential cost reduction.

Elements of cost reduction in transportation networks The presentation of different cost economies above sheds light on some key elements that may or may not lead to cost reductions as transportation networks grow. In summary, the key elements discussed above that may influence the cost efficiencies of transportation networks are: • the demand for transportation on certain network links, • the nodes in the network, which relate to the OD-connections offered in the network, • the routing through the network, • bundling different products on one transport vehicle, • the transportation distances, and finally

26

2.3. Optimal transportation network design - an Operational Research perspective • the transportation vehicles. One way or the other, these elements will influence the costs in transportation networks and should therefore be kept in mind when designing these. The following section will present OR models that find optimal - usually costoptimal - transportation networks. The above mentioned elements are of importance for OR research as well.

2.3. Optimal transportation network design - an Operational Research perspective “OR is the discipline of applying appropriate analytical methods to help make better decisions.” (INFORMS, 2010) The term Operational Research (OR) has been known since around the Second World War. It is rooted in two areas: the military area focusing on applied research concerning, at first, radar systems as well as the private sector that was eager to apply methodology from sciences to management questions. From early on, OR was an applied science with a special interdisciplinary focus. Research continued to develop new models, solution algorithms, and heuristics to support real-life decisionmaking. It benefited strongly from advances in computer science that provide researchers nowadays with computational support beyond thought decades ago (Zimmermann, 2008, pp. 6-9).33 Many different terms are used to describe the application of analytical methods for managerial decision-making. Operational Research is the term the INFORMS Society decided to put forward in 2004. Amongst others, the prominent wording of Management Science basically covers the same aspects; Operations Research is more frequently used than Operational Research in Germany but describes the same research field. Thus, Operational Research shall be the umbrella term for the entire field (INFORMS, 2010) and, therefore, also the term used in the following. Network design creates networks explicitly, e.g. by defining the nodes and the links that should be included in the networks.34 The problems arising in network design can often be solved to optimality with the toolkit provided by OR. Even a short glance at the literature on network design will highlight the importance of optimization models in this context. Nevertheless, optimization is only one method established in OR: scenario analysis and simulation will also support the decision-making process. 33

34

When taking a look at research streams in OR, an orientation to methods and less to applications is discovered. This might be a misleading picture due to the fact that application topics are more difficult to publish in academic journals. The term network design is sometimes used in a restricted understanding in the field of urban planning: “The network design problem involves the optimal decision on the expansion of a street and highway system in response to a growing demand for travel” (Yang and Bell, 1998, p. 257). As opposed to this, a broad understanding - independent of a specific type of network - will be applied in the following.

27

2. Transportation networks and their optimal design

Furthermore, good heuristics play a major role in network design. Any method fulfills a different purpose. Scenario analysis suggests decision alternatives, simulation helps to evaluate alternatives, heuristics find good solutions for difficult problems, whereas optimization provides optimal solutions for a given model (Grünert and Irnich, 2005, pp. 8-9). Abstractly, network design will always address the same questions for all networks:

Number of nodes. Some nodes in the network will be externally given, such as customer or supplier locations. It remains an open question how many additional nodes are to be added to the network in order to achieve the expected performance.

Location of nodes. If nodes are added or should be relocated for the network, one will have to decide where to locate these. One example of this is the decision where to locate a new production site, considering location information of suppliers and customers.

Connecting nodes with links. For most networks, the planner will have to decide which links are to be added or operated in the network. Energy providers can think about actually constructing new cables while airlines will rather have to decide which airports to connect by a scheduled aircraft.

Operating the links and nodes. Finally, companies will have to set policies concerning how to operate the constructed or opened links and nodes. This includes setting schedules for public transport but also figuring out how the landing / take-off and ground operations at an airport are conducted efficiently. Networks usually consist of several layers that are linked to and embedded in each other. The self-similarity of the network model (Delfmann et al., 2010, p. 59) implies that the above illustrated questions in network design may arise on all layers. For instance, intralogistics transport flows within a production facility form one layer of a network that is linked to the external transportation network for final products. Transportation network design is an important field of application for OR methods. Crainic and Laporte state that “transportation planning is undoubtly one of the great success stories” of OR (Crainic and Laporte, 1997, p. 435). This section presents important streams of research to design optimal transportation networks. Certain overlaps in content exist with neighboring fields such as SCM, SC design or supply network design.35 Different time horizons are relevant when designing networks. Decisions are taken on a strategic, long-term planning horizon; others on tactical, mid-term planning 35

28

Chopra (2003) illustrates different options for designing distribution networks. Klibi et al. (2010), Melo et al. (2009), Meixell and Gargeya (2005), and Lapierre et al. (2004) provide overviews of distribution network design from an optimization point of view.

2.3. Optimal transportation network design - an Operational Research perspective

horizon and eventually on the operational, short-term horizon. Crainic and Laporte (1997) draw an overview of the different planning issues and have influenced the following sections.36 The next section will very briefly introduce some graph theoretic terminology that is necessary for the presentation of the models in the following. Subsequently, OR formulations for strategic and tactical decisions in transportation network design are presented. Further, operational decisions are sketched to complete the picture. The goal of the presentation is twofold. Firstly, the type of decisions to be taken at each planning horizon is characterized by exemplary model formulations. This will provide a clear picture of the manifold challenges to be solved in transportation network design. Secondly, the presentation allows to identify a strong cost-orientation in the OR models for transportation network design. Yet, the service-oriented design of networks is also present in the pertinent literature; the addressed aspects of serviceorientation are briefly outlined following the model orientation. Based on the presentation of the models, some key performance indicators (KPIs) for transportation network design are identified for further discussion in chapter 3.

2.3.1. Graph theoretic terminology for network design Networks have a graph representation.37 A brief introduction to the graph theoretic terminology that will be applied in the following will simplify the understanding of the OR models in this section.38 The mathematical field of graph theory provides researchers with a wide terminology and toolkit to support the analysis and design of networks. A graph G consists of a set of nodes (or vertices) V , a set of edges (or links) E as well as a relation ω (incidence relation) that maps each element of E onto two elements i and j in V . Thus, E contains 2-tuples of nodes. If the pairs in E are ordered, the graph is directed; if the pairs are unordered, the graph is undirected (Domschke and Drexl, 2005, p. 65). Edges in directed graphs are also referred to as arcs or arrows. Directed graphs have paths: a sequence of arcs (a 1 , ..., a t ) is a path p 0t in G if a sequence of nodes (v 0 , ..., v t ) with a h = (v h−1 , v h ) for all h = 1, ..., t exists (Domschke and Drexl, 2005, p. 67). A graph (directed or undirected) is a weighted graph if some value is associated with the edges of the graph. The value can have many interpretations, common are the 36

37 38

Barnhart et al. (2002) suggest a time line of four planning activities: strategic planning, tactical planning, operations (or market) planning, and contingency planning. The tasks that Crainic and Laporte (1997) allocate to the tactical and operational horizons are allocated to the tactical, operations, and contingency planning by Barnhart et al. (2002). Figure 2.1 touched on this aspect before. The formal definitions express the same ideas as presented above in section 2.2.1, where the terms were already introduced without making reference to the graph-theoretic interpretation of it. The brief presentation of fundamental graph theoretic terminology here, is aimed to support the understanding of the next sections and not intended to be a comprehensive introduction to graph theory. Many textbooks exist to this goal, e.g. West (2001).

29

2. Transportation networks and their optimal design

cost to use that edge, the length of the edge or the capacity of the edge. The notation of c i j is used for the value (especially the cost) associated to edge (v i , v j ). If p 0t= (v 0 , ..., v t ) is a path in G , the sum of all associated values with edges in p , t c (p ) = h=1 c v h−1 v h is defined as the length of the path p . A path p i∗j is a shortest path from i to j if no other path p i j in G exists with c (p i j ) < c (p i∗j ). In that case, c (p i∗j ) is the shortest distance from i to j (Domschke and Drexl, 2005, p. 68).

2.3.2. Strategic decisions for network design Strategic issues in transportation network design are usually related to infrastructural decisions, as these must be taken on a long-term planning horizon. A typical example is the location of terminals in the network and their capacity planning. In contexts where the transport connections depend on infrastructure networks such as tracks, decisions on their location and capacity are just as important.39 Similarly, for the planning process of freight carriers, Crainic (2003, p. 469) characterizes strategic decisions to “determine general development policies and broadly shape the operating strategies of the system over relatively long-term horizons”.

Optimal terminal location Many different location models have been developed in OR. Generally speaking, all location models find optimal geographic positions (e.g. for constructing a production plant or a transshipment terminal). Figure 2.7 depicts the different location decisions and classifies the examples given by ReVelle and Eiselt (2005, pp. 2-3). Location problems differ in the topology of potential positions: either in the plane or on a network. Planar problems are usually formulated in 2-dimensional real space, but e.g. cluster analysis maps locations in n-dimensional space by making use of planar location problems. Location models on networks limit the potential positions to a given network which is common in transport-related problems. Both types of problems may further be divided into continuous and discrete models. Continuous problems allow the location decisions to be taken anywhere in the plane or network whereas discrete models limit the potential locations to choose from. Even though one finds relevant applications for all four possible combinations of problem classes, by far most of the problems discussed are either continuous planar or discrete network models (ReVelle and Eiselt, 2005, Klose and Drexl, 2005).

Continuous location models in the plane In the context of location decisions, the models usually refer to 2 but the methodology is easily adapted to n . The following 39

30

Song et al. (2008, p. 1265) identify two different perspectives in strategic network design: the shipper and the carrier perspective. The shipper will usually include inventory costs into the models while the carrier is more interested in optimally utilizing its assets. The following presentation will provide a general overview of strategic decision in network design and not differentiate between the two perspectives.

Set of ppotential loccations

2.3. Optimal transportation network design - an Operational Research perspective

discrete

continuous

Positioning of transmitter stations at some permissible points

Locating retail facilities on lots that are zoned for them

Landing a hhelicopter li t for f trauma pickup

Placing a tow truck along l a stretch t t h off highway

planar

network

Topology of potential locations

Figure 2.7.: Situative examples of location problems.

assumptions are at the core of continuous planar location models (Domschke and Drexl, 1996, p. 162): 1. Customer locations are situated on a homogeneous plane. 2. Every point on the plane is a potential location. 3. The distance between two points on the plane is measured according to a certain metric. Domschke and Drexl (1996, p. 164) present common metrics to measure the disy y tance between two points v i (v ix ,v i ) and v j (v jx , v j ): • the rectangular distance metric, often used in intralogistics contexts      x  y y x d i s t i j := v i − v j  + v i − v j  • the euclidean distance metric , suitable for many situations  y y d i s t i j := (v ix − v jx )2 + (v i − v j )2 • the squared euclidean distance metric, especially applicable when long distances are strongly unfavorable, such as in defining locations for emergency buildings (firefighters, hospitals, etc.) y

y

d i s t i j := (v ix − v jx )2 + (v i − v j )2 The Steiner-Weber model is a formulation for a continuous location model with euclidean distance metric (Delfmann, 1987). The n known customer locations are in y the form (v ix , v i ), each having a weight of d i . The coordinates of the potential location

31

2. Transportation networks and their optimal design is given in the form (x , y ). The model is stated as follows (Domschke and Drexl, 1996, p. 167): min F (x , y ) =

n 

di



y

(x − v ix )2 + (y − v i )2

i =1

The Steiner-Weber model cannot be solved analytically, but iterative solution procedures exist, especially Miehle’s algorithm converges to accurate solutions after only a few iterations (Miehle, 1958, p. 239). It is apparent that the Steiner-Weber model targets at minimizing the distance-weight-related costs in the plane.

Discrete location models in networks The warehouse location problem (WLP) is the most prominent discrete network model. The simple WLP describes the following situation. A company delivers to n customers and wants to decrease its distribution cost. It plans to open one or several warehouses. m possible locations are presef ix lected for these. There are fixed costs c i associated with opening a warehouse at a location i . Transports from warehouse i to customer j cost c i j monetary units. The objective of the model is to minimize the overall costs of the network, by opening the best number of warehouses and define the allocation of customers and warehouses (Domschke and Drexl, 1996, pp. 51-52).

min F (x , y ) =

m  n 

ci j xi j +

i =1 j =1

m 

f ix

ci

yi

(2.1)

i =1

s .t . x i j ≤ y i for all i = 1, ..., m and j = 1, ..., n m 

x i j = 1 for j = 1, ..., n

i =1

y i ∈ 0, 1 for i = 1, ..., m x i j ≥ 0 for all i and j The decision variables in the WLP are understood as:  1 if customer j is fully served by warehouse i xi j = 0 if j does not deliver to i 0 ≤ x i j ≤ 1 if i delivers d j x i j units to j  1 open warehouse at location i yi = 0 else

32

(2.2) (2.3)

2.3. Optimal transportation network design - an Operational Research perspective

Extensions of the simple WLP include capacitated warehouses, non-linear costs, and the introductions of additional echelons (which leads to transshipment problems).40 It can already be seen from the simple WLP that the minimization in this case includes not only transport-related costs but also additional cost elements such as fixed warehousing costs. Alumur and Kara (2008) review in detail network location problems.

Optimal network links The other strategic perspective - apart from finding the optimal terminal location is to decide on the links in the network. Formulations start with a given demand for transport services from origins to destinations and will then find the best set of links to produce the related transportation service (Crainic and Laporte, 1997, p. 413).41 Finding optimal links is of varying importance for different modes of transport. In rail transport, finding optimal links often relates to the construction of tracks which is a strategic decision for companies. In road transportation, finding optimal links supports the mid-term purchase of truck capacity,42 since roads as the infrastructure is mostly provided by the states and part of the planning of companies. Many model formulations for the identification of network links stem from graph theory. Two basic versions of the problem are minimum spanning trees and shortest path algorithms for a graph. Both allow to identify links that should be included in a network. They may also be used for tactical decisions. On a strategic level, both formulations may support the strategic infrastructure planning concerning network links. A minimum spanning tree is a group of arcs connecting all nodes in a network with minimal cost (Winston (2004, p. 456), West (2001, p. 95)).43 Shortest paths problems identify the shortest path from an origin to all or a predefined destination. Algorithms to find shortest paths are applied in network design (Bazaraa et al., 1990, p. 575). Summing up, strategic decisions in transportation network design identify optimal network nodes and - in certain contexts - network links. The solutions depend on cost and demand assumptions. The strategic decisions are usually taken as input for tactical decision-making.

40

41

42 43

For a capacitated model, each customer has a demand of d j units. An additional constraint to force the model to meet the entire demand is then included. The WLP already includes the variable x i j that hints at which links are to be used in the network design. From this perspective, the WLP finds optimal nodes and links simultaneously. Opening a link in a road transport networks usually means to schedule a truck on that link. Ahuja et al. (1993) give a comprehensive overview of elaborated algorithms to find minimum spanning trees.

33

2. Transportation networks and their optimal design

2.3.3. Tactical decisions for network design Once strategic decisions about infrastructure in a network were taken, tactical issues arise in network design with a mid-term planning horizon. Strategic decisions often act as input to tactical decision-making. The tactical planning is concerned with “services on an existing network configuration” (Song et al., 2008, p. 1266). The arising challenges differ by network purpose. For networks that require consolidation, such as LTL networks, finding the optimal service network is of crucial importance. The routing of freight on this network is referred to as the load plan and its generation constitutes an important challenge in this context. As opposed to this, in cases where several pick-up and delivery operations are performed, such as in the courier business, the vehicle routing problem (VRP) is highly relevant. Moreover, crew scheduling is important for all transportation networks based on tight schedules and relying on a skilled workforce. The following sections will shed light on these challenges and related models. Further elements are of importance in the tactical planning but will not be discussed as typical and prominent examples for tactical network design. Amongst them are terminal policies. These include the definition of tasks to be performed within the terminals, e.g. how shipments are to be consolidated. O’Kelly (2010) draws attention to the planning of routing in the transshipment terminals. He points out that this questions has hitherto been neglected in the literature on transportation network design. The efficient task distribution between terminals is also seen as part of the terminal policies to be set (Crainic and Laporte, 1997, p. 421). In MH networks this touches the task of allocating terminals to hubs. The literature provides generic trade-offs in this decision.44 Another element of tactical planning is related to the balancing of empties. The question to be solved includes how to reposition empties (vehicles, containers, boxes, etc.) for future (such as next-day) operations (Crainic and Laporte, 1997, p. 421). The literature on repositioning empty vehicles has many contributions directed to the rail transport industry as empty rail car distribution is an important planning issue in this industry (e.g. Holmberg et al., 1998) but is not limited to this field (Turnquist, 1985, pp. 361-362). The management of returnable packaging such as containers of any kind (e.g. Kroon and Vrijens, 1995, Mevissen, 1996) is another element but less directly connected to transportation network design.45

44 45

34

Section 2.4.2 explicitly addresses these. Closed loop supply chain management (CLSCM) has developed more recently. Reverse flows and (re) distribution circuits for returnable containers are explicitly treated together with forward flow SCs with the goal to create models and therewith solutions that can describe and provide answers for the flows in the entire networks. Dekker et al. (2004) and Dyckhoff et al. (2004) provide valuable overviews of the variety of aspects discussed in reverse logistics and CLSCM.

2.3. Optimal transportation network design - an Operational Research perspective

Optimal service network The goal of service network design models is to “plan services and operations to answer demand and ensure the profitability of the firm” (Crainic, 2000, p. 280). The term service in this context expresses the frequencies of transport. Coming from the terminals or facilities located during the strategic planning phase, the service network design now aims at selecting the connections between the network’s nodes and especially their frequencies (Crainic and Laporte, 1997, p. 421). Crainic (2000) gives an overview of different models for service network design. These can be distinguished into two categories: models treating frequencies as decision variables as well as models deriving service frequencies as their output. The former use decision variables that express how often the service is offered in the planning period. The latter have binary “operate or not” variables (Crainic, 2000, p. 281) and define the service frequency based on these and a minimum service constraint. Both formulations lead to a load plan, i.e. the plan on “how a shipment will be routed through the network, in the form of the sequence of [break-bulk (BB) terminals] through which the shipment must pass before reaching the destination [end-of-line (EoL) terminal]” (Powell and Sheffi, 1983, p. 472). A short example of a model with frequencies as decision variables gives an idea about the types of formulations found in the literature (Crainic and Laporte, 1997, p. 423). min Ψ(X pg , Fπ ) s.t. 

X pg = d g for all g

(2.4)

(2.5)

k

X pg ≥ 0 for all p, g Fπ ≥ 0 and integer for all π Fπ is the decision variable representing the frequency of service for each ODconnection π in the network. Hence, Fπ stands for the fixed cost of the transportation g system. The variable cost of the system is expressed in X p being the routing of freight. It is the amount of flow of traffic class g on the itinerary p .46 An itinerary may cover several services. The transportation demand per traffic class is d g . The objective g function Ψ(X p , Fπ ) accumulates the total cost of the transportation system by incorporating fixed and variable elements. Crainic and Laporte (1997, p. 423) provide an example of the explicit form such an objective function may take. Their formulation further highlights that delay costs or penalty terms may be included directly into 46

A traffic class represents the transport demand for a certain commodity, e.g. passenger or cargo, from an origin to a destination.

35

2. Transportation networks and their optimal design

the objective function. As frequencies in this model are decision variables Fπ , the optimal frequency per service can be read from the solution of the model. An often referenced example of a model that derives service frequencies as its output is the model provided by Powell and Sheffi (1983). In this case, the frequency F (x ) depends on the traffic flow x in vehicle loads on a route. A minimal service frequency Fm i n is enforced with the help of a minimal service restriction. This situation reflects nicely how LSPs often operate their networks. The service frequency then is defined as ⎧ 0 x =0 ⎨ F (x ) = Fm i n 0 < x ≤ Fm i n ⎩ x x > Fm i n The primary decision variable y isj stands for the developed load plan  y isj =

1

if freight at node i , destined for s should go next to terminal j , i j ∈ L, s ∈ T

0

otherwise

The formulation further requires the binary variable δir js  δir js

=

1

if link i j is on the path defined by y from r to s

0

otherwise

Let x be the vector of flows as F is the vector of frequencies. Several sets are of importance: L includes all direct links joining terminals, the set of all nodes V , the set of all transshipment terminals T and finally the set of all links representing movements through break-bulks B . The costs taken into account are the transportation cost per vehicle from i to j c i j , and the respective cost for handling operations of a full vehicle: c h f u l l . D r s is the total flow from origin terminal r to destination terminal s . t i j is the time spent on a link and ςr s is the maximum delivery time constraint. With these variables, the full model can be stated: min Z ( x , F ) =



c i j Fi j (x i j ) +

i j ∈L

s.t. 

y isj = 1, for all s ∈ T, i ∈ V

J

xi j =  ij

36



D r s δir js



hf ul l

ci j

xi j

(2.6)

i j ∈B

(2.7) (2.8)

r ∈T s ∈T

t i j (x i j ) · δir js ≤ ςr s , for all r s

(2.9)

2.3. Optimal transportation network design - an Operational Research perspective This formulation also relates to the frequencies but needs the binary variable δir js to match if a service is operated at all. Powell and Sheffi (1983) suggest a solution algorithm to solve the problem. Looking at both of these optimization formulations for service network design it is again worth mentioning that their objective functions are purely cost-oriented and service considerations are incorporated either as constraints or by using the workaround of penalty costs. The service network design problems stated above are relevant in the long distance transportation where terminal consolidation plays an important role. The generated load plan represents the routing of freight. For settings with shorter distances such as distribution tours in urban areas, the desired information is less oriented towards efficiencies by consolidating but rather by reducing transportation distance. OR treats these questions with the help of VRPs.

Optimal routing Basic vehicle routing problems (VRPs) are used to find cost optimal tours in a specific situation. The number and location of customers as well as their demand is known. A service provider aims to fulfill demand from its depot with a limited number of transportation vehicles. The service provider has some restrictions as to the capacity of its trucks and as to time. It strives to find the most cost efficient tour (as a sequence of customers to visit one after another) by respecting the constraints (Domschke and Drexl, 1990, p. 131). The single depot VRP reads as follows (Laporte, 2009, pp. 409-410):  ci j xi j (2.10) min (i ,j )∈E

s.t. n 

x 0j = 2l

j =1



xik +

i