CITY LOGISTICS

NETWORK MODELLING AND

INTELLIGENT TRANSPORT SYSTEMS

EllCHl TAN1GUCHl
Kyoto University
RUSSELL G THOMPSON
University of Melbourne
TADASHl YAMADA
Kansai University
RON VAN DU1N
Delft University of Technology

I
United Kingdom - North America - Japan
India - Malaysia - China

Emerald Group Publishing Limited

Howard House, Wagon Lane, Bingley BD16 1WA, UK
First edition 2001
Copyright 2001 Emerald Group Publishing Limited
Reprints and permission service
Contact: booksandseries@emeraldinsight.com
No part of this book may be reproduced, stored in a retrieval system, transmitted in any
form or by any means electronic, mechanical, photocopying, recording or otherwise
without either the prior written permission of the publisher or a licence permitting
restricted copying issued in the UK by The Copyright Licensing Agency and in the USA
by The Copyright Clearance Center. No responsibility is accepted for the accuracy of
information contained in the text, illustrations or advertisements. The opinions expressed
in these chapters are not necessarily those of the Editor or the publisher.
British Library Cataloguing in Publication Data
A catalogue record for this book is available from the British Library
ISBN: 978-0-08-043903-7

Ccrlilk<l lCnum ber

.. ! .~~~ ...

Awarded in recognition of
Emeralds production
departments adherence to
quality systems and processes
when preparing scholarly
journals for print

r~t

~~~J?~
INVESTOR IN PEOPLE

vii

PREFACE
It is well recognised that urban freight transport plays a vital role in the sustainable
development of cities. However, urban freight transport recently faces many challenging
problems, including high levels of traffic congestion, negative environmental impacts, high
energy consumption and a shortage of labour. Within these difficult conditions freight carriers
are also requested to provide higher levels of service with lower costs.
In response to these problems a new area of transport planning has emerged called City
Logistics. City Logistics is the process of totally optimising urban logistics activities by
considering the social, environmental, economic, financial and energy impacts of urban freight
movement.
This book presents the basic concepts and approaches of City Logistics. In particular, it focuses
on modelling City Logistics. Modelling is very important, since the estimation and evaluation
of the impacts generated by City Logistics measures is required before implementing them.
The book mainly deals with network modelling including, vehicle routing and scheduling
models with time windows, location models of logistics terminals and impact models of City
Logistics measures.
Recent developments in the field of Intelligent Transportation Systems (ITS) can facilitate the
implementation of many City Logistics initiatives. Currently, advanced telecommunication
systems provide powerful tools for efficiently operating vehicle fleets. Sophisticated logistics
systems can now be developed by integrating Global Positioning Systems (GPS) and
Geographical Information Systems (GIS) in conjunction with application software. Therefore,
ITS based City Logistics has become more realistic in many industrialised countries. In this
context, we present a theoretical and practical treatment of modelling City Logistics based on
ITS.
City Logistics provides an opportunity for innovative solutions to be developed for improving
the quality of life in urban areas. The modelling approaches described in this book are
relatively new and are not yet commonly used in city planning. Although, several cities have
already implemented some City Logistics initiatives, there currently are a limited number of
evaluation tools that have been developed for predicting the consequences of such schemes.
However. the models described in this book should provide useful tools for researchers and city
planners for evaluating City Logistics policies or measures to help solve urban freight
problems.
This book was initially planned to be a textbook for a graduate course on City Logistics.

viii

However, the authors believe that it will be also beneficial for city planners and logistics
managers in industry as well as students. The first author of this book initiated a graduate
course on City Logistics at Kyoto University, Japan in 1995. He thought that it was necessary
to publish a textbook on the subject to disseminate the concepts of City Logistics not only to
students but also to city planners.
The four authors have previously conducted collaborative research in the area of City
Logistics, much of which is contained in this book. They organised the First International
Conference on City Logistics at Cairns, Australia in 1999 and published the proceedings, Ciry
Logistics I. They are also active members of the Institute for City Logistics
(http://www.citylogistics.org), which provides a scientific platform for research and
development related to City Logistics and urban freight issues. Therefore, this book is really a
collaborative product of the four authors.
This book is composed of nine chapters, with each author making a major contribution to a
number of chapters:
E. Taniguchi: Chapters 1, 5, 6 and Section 7.3
R.G. Thompson: Chapters 2 , 3,7 and Section 1.7
T. Yamada: Chapters 4 and 8
J.H.R. van Duin: Chapter 9 and Section 2.7
The authors would like to express their heartiest appreciation to the valuable advice to the
researches by Professor Y. Iida of Kyoto University, Japan and Professor R.E.C.M. van der
Heijden of Delft University of Technology, The Netherlands.

August 2000
Eiichi Taniguchi
Russell G. Thompson
Tadashi Yamada
J.H. Ron van Duin

CONTENTS
1. Introduction ......................................

1.1 Background ............................

..................
............................................................. 1
............................................................. 2
............................................................. 3

1.2 Fundamental Concepts ...........

1.3 Stakeholders ...........................
1.4 Evaluation criteria ...
................................................................................... 4
1.5 City logistics schemes
.......................................................................................... 5
1.6 Modelling .........................................................................................
1.7 Systems approach ...........................
..................

....................................................................................... 17
.......................................................................................
17
2.1 A Modelling framewo
2.2 The model development process ......................................................
2.3 Mathematical programming ................................
2.4 Optimisation ........................................................
..................
2.5 Simulation ..............................
........................................................... 29
2.6 Meta-Heuristic techniques .............................................................................................
32
36
...
2.7 Modelling actors perceptions ............................

2 . Modelling City Logistics

.................................................................................... 49
3. City Logistics with ITS ........
3.1 Fundamental concepts
..................49
3.2 Data acquisition .....................
...........................................................
50
3.3 Data processing .............................................................................................................. 54
.................
... 56
3.4 Information Dissemination .................................
3.5 Geographic information systems ................................................................................... 60
... 61
3.6 Effects of e-commerce ........................................
........................................................... 62
3.7 Current ITS applications
........................................................................................
64
3.8 Evaluation issues
4. Demand and supply models ......................................
4.1 Overviews ...........................................................
4.2 Generation .......
...........
4.3 Distribution ....................................................................................................................
4.4 Mode split .....................................................................................................................
4.5 Assignment ............................
..................
4.6 Simulation models
4.7 Travel times
...........................................................

71
73

78

vi

4.8 Generalised costs ..........................................................................................................

82

5. Impact models ......................................................................................................

.....................................................................
........................................
5.3 Environmental models ...................................
......................
5.4 Financial models .....................................................................................
5.5 Energy consumption models ..........................................................................

85

6. Vehicle routing and scheduling

.......................................
6.1 Overview ...............................................................
6.2 Travelling salesman problems ............................................................................
6.3 Vehicle routing and scheduling with time windows .....................................................

111

5.2 Social and economic models

85
86

114

6.4 Vehicle routing and scheduling with time windows and traffic simulation ..................123
7. Vehicle routing and scheduling with ITS
7.1 Overview ............................................................................................................
7.2 Probabilistic vehicle routing and scheduling with time windows ................................ 137
7.3 Probabilistic vehicle routing and scheduling with time windows and traffic simulation
...................................................................................................................................... 152
7.4 Dynamic vehicle routing and scheduling with real time information ...........
169

8. Location of logistics terminals

8.1 Overview .......................................................................................................................
8.2 Size ................................................................................................................................
8.3 Location .................................................................
.........................
8.4 Location with traffic assignment
8.5 Multi-objective optimisation .........................................................................................

175
175
176
182

......................

203

9. Future perspectives
9.1 Systems view and future developments

184
199

9.2 Intermodal transport systems

203
207

9.3 Underground freight transport systems

9.4 Virtual freight companies ..............................................................................................

218
226

References ................................................................................................................................

.231

Index ........................................................................................................................................ ,245

(Introduction) 1

INTRODUCTION
1.1 BACKGROUND
Recently urban freight transport faces many difficult problems. Freight carriers are expected to
provide higher levels of service within the framework of Just-In-Time transport systems with
lower costs. Congestion levels on urban roads have been constantly rising due to the increasing
levels of traffic demand. The environmental problems caused by traffic have become major
issues in many cities. Large trucks produce a substantial amount of air pollution in urban areas
by emitting NO,, SPM (Suspended Particle Material) and other gases. Energy conservation is
also an important issue not only because of the limited amount of natural resources available
but also for reducing COz emissions to limit global warming. Truck crashes often lead to
substantial trauma for the community.
This section describes the background of City Logistics. It highlights the problems associated
with the logistics activities by freight carriers as well as social problems including environment,
congestion and energy savings in urban areas. The emphasis is on the recent rapid development
of ITS (Intelligent Transport Systems) which has the potential to provide effective measures for
reducing freight costs as well as solving the social problems. For example, location
identification systems for trucks using GPS (Global Positioning Systems) and mobile phone
enable freight carriers to dynamically operate vehicle routing and scheduling. This can help
minimise freight costs and reduce the total time travelled by vehicles.
There is an ongoing trend towards urbanisation in the world. Cities provide more attractive
opportunities for employment, education, cultural and sport activities, etc. The concentration of
population in urban areas is observed in most industrialised and developing countries. However,
this leads to expanding urban areas and often generates freight transport problems, due to a
lack of appropriate urban logistics policies.
A high proportion of total goods movement occurs within cities. For example, in Tokyo and
surrounding areas, about 67% of total goods movement in terms of tonnage have both the
origin and destination within the 23 wards (central area) according to a survey conducted in

2 ( C i v Logistics)
1994. It indicates the importance of goods movement within urban areas.
There is a high cost in terms of money and time incurred as a result of the movement of goods
within cities. Trucks carry most goods within cities, and road transport has become expensive
due to the decreasing load factors of trucks. This reduction in load factors of trucks has been
caused by the recent trend towards smaller loads being frequently transported to meet a wider
range of consumers needs.
Cities are now facing the global competition for investment and trade with an efficient
transport system essential for sustained economic prosperity. Therefore, the efficient and
environment friendly logistics systems help cities become more competitive in terms of
economic development.
The recent development of e-commerce (electronic commerce) also makes City Logistics more
important. There are two points to discuss the impacts on City Logistics by the development of
e-commerce.
(a) E-commerce changes the logistics activities by giving a high priority to the demands of
customers or consumers.
(b) Logistics activities themselves incorporate e-commerce for matching the demand and
supply of goods movement.
The e-commerce provides a good opportunity for a quick, individual and direct commerce in
the business to business (B2B) and the business to consumer (B2C) cases at low price. As a
result, manufacturers need to change their logistics systems to faster, more reliable systems
with lower costs for meeting with the higher level of individual demand of consumers. The
platform for matching the supply and demand for goods movement through the Internet makes
it possible to rationalise the logistics systems by increasing the load factors of trucks. These
changes may or may not contribute to alleviating traffic congestion and improve the
environment. Therefore, the application of City Logistics initiatives will become more essential
for reducing freight costs and solving social problems.

1.2 FUNDAMENTAL
CONCEPTS
The concept of City Logistics (e.g. Ruske, 1994; Kohler, 1997; Taniguchi and van der
Heijden: 2000a) has the potential for solving many of these difficult and complicated problems.
Taniguchi et al. (1999a) defined City Logistics as the process for totally optimising the
logistics and transport activities by private companies in urban areas while considering the
traffic environment, the traffic congestion and energy consumption within the framework of a

(Introduction) 3
market economy. The aim of City Logistics is to globally optimise logistics systems within an
urban area by considering the costs and benefits of schemes to the public as well as the private
sector. Private shippers and freight carriers aim to reduce their freight costs while the public
sector tries to alleviate traffic congestion and environmental problems.

1.3 STAKEHOLDERS
There are four key stakeholders involved in urban freight transport; (a) shippers, (b) freight
carriers, (c) residents and (d) administrators. Each of the key stakeholders in urban freight
transport has their own specific objectives and tends to behave in a different manner. City
Logistics models need to recognise these factors.

Shippers
(manufacturers,
wholesalers,
retailers)

Freight carriers
(Transporters,
warehouse
companies)

Residents
(consumers)

Administrators
(national, state,
and city level)

Figure 1.1 Key stakeholders in City Logistics

Shippers are the customers of freight carriers who either send goods to other companies or
persons or receive goods from them. Shippers generally tend to maximise their levels of
service, which includes the cost, the time for picking up or delivering, and the reliability of
transport as well as trailing information. Recently the requirement for carriers to arrive at
customers within specified time windows for pickup/delivery has become popular. A recent
survey in Osaka and Kobe in Japan, found that freight carriers were required operate with
designated arrival times or time windows for 52% of goods delivered and for 45% of goods

4 (City Logistics)
collected in terms of weight. Such strict time windows have led to smaller loads of goods being
transported more frequently. The reliability of delivering goods has become more important for
Just-In-Time transport systems. There are two types of reliability; (a) delivery without any
damage to the goods, (b) delivery without any delay with respect to designated time at
customers.
Freight carriers typically attempt to minimise the costs associated with collecting and
delivering goods to customers to maximise their profits. There is much pressure to provide
higher levels of service to customers at a lower total cost. This is especially important when
carriers are requested to arrive at customers within a designated time period. However, freight
carriers often face difficulty in operating their vehicles on urban roads due to traffic congestion.
This has led to the inefficient use of trucks, where smaller loads are being transported and
trucks often have to wait near the location of customers when they arrive earlier than the
designated time.
Residents are the people who live, work and shop in the city. They do not welcome large trucks
coming into local streets, never the less these vehicles carry commodities that are necessary for
them. They would like to minimise traffic congestion, noise, air pollution and traffic accidents
near their residential and retail areas. Within the commercial zones of urban areas, retailers
want to receive their commodities at a convenient time for them. However, this sometimes
conflicts with residents who desire quiet and safe conditions on local roads.
City administrators attempt to enhance the economic development of the city and increase
employment opportunities. They also aim to alleviate traffic congestion, improve the
environment and increase road safety within the city. They should be neutral and should play a
major role in resolving any conflicts among the other key stakeholders who are involved in
urban freight transport. Therefore, it is the administrators who should co-ordinate and facilitate
City Logistics initiatives.

1.4 EVALUATION
CRITEIUA
There are many criteria for evaluating City Logistics initiatives since there are numerous
stakeholders involved in urban goods movement. Cost minimisation or the profit maximisation
are typical criterion for freight carriers and shippers. The minimisation of NOx, COz emissions,
noise level, vibration level and road crashes can be a criterion for residents and administrators.
Since there are a number of evaluation criteria for each stakeholder, it is difficult to determine a
single evaluation a measure for City Logistics initiatives. Here, multi-objective evaluation and
Pareto optimisation techniques and can be used to compare the performance of alternative
schemes.

(bztroductioit) 5

1.5 CITYLOGISTICS
SCHEMES
1.5.1 Overview
City Logistics usually includes one or more of the following initiatives:
(a)
(b)
(c)
(d)
(e)

Advanced information systems

Co-operative freight transport systems
Public logistics terminals
Load factor controls
Underground freight transport systems

It is common for these initiatives to be combined and varied to be compatible with local
transport planning policies.

1.5.2 Advanced Information Systems

Advanced information systems have become important in rationalising existing logistics
operations. In general, advanced information systems for pickup/delivery trucks operations
have three important functions:
(a)
(b)
(c)

To allow communication between drivers and the control centre

To provide real time information on traffic conditions
To store detailed historical pickup/delivery truck operations data

The third function has not been fully discussed in the literature, but it is very important for
rationalising logistics operations. A Japanese milk producing company experienced one
successful application of historical operations data. After introducing a satellite based
information system for one year, the company was able to reduce the number of
pickup/delivery trucks by 13.5% (from 37 to 32 vehicles) and increase their average load factor
by 10 percent (from 60% to 70%). A computer based system was used to store detailed
historical data of the pickup/delivery trucks operations, including times of starting/arriving
times at the depot and customers as well as the waiting times, travelling speeds and routes
travelled. The company was able to analyse this data and change their routes and schedules to
substantially increase the efficiency of their vehicle fleet. This type of system can reduce both
freight transport and environmental costs within a city. Further details on how information
technology can be applied as an effective City Logistic initiative is presented in Chapters 3 and
7.

6 (CityLogistics)

1.5.3 Cooperative Freight Transport Systems

Several researchers have investigated cooperative freight transport systems (Ruske, 1994;
Taniguchi et al., 1995, 2000c; Kohler; 1997) that allow a reduced number of trucks to be used
for collecting or delivering the same amount of goods. Based on survey by Kohler (1997), it is
remarkable to see competitive freight carriers cooperating in delivering goods to the inner city
of Kassel in Germany. A neutral freight carrier collects goods from 5 freight carriers and
delivers them to shops in the inner city. After introducing this system the total time travelled by
trucks was reduced as well as queues of trucks waiting to deliver goods on streets. Originally
this system started with 10 freight carriers, and now 5 companies remain in the cooperative
system. Another outstanding case is the cooperative delivering system among 11 department
stores in Osaka, Japan. In this system, basically two department stores having depots adjacent
each other, exchange their goods to be delivered in neighbourhood of the depot. This led to the
considerable reduction of travel time for trucks, person work hours and total costs. As observed
in these cases, cooperative freight systems can substantially reduce transport costs as well as
environmental impacts.

1.5.4 Public Logistics Terminals

Public logistics terminals located in areas surrounding a city can be helpful in promoting
cooperative freight transport systems (Janssen et al. 1991, Taniguchi er al. 1999b, Duin 1997).
A good example of this platform for city distribution can be seen in Monaco (Dablanc, 1998).
This platform is provided by the government and operated by a private freight carrier for
delivering goods to city areas. This company is subsidised by the government to provide a
delivery service with cheaper prices than normal. This system helps reduce the required
number of trucks used for deliveries. In Japan the first multi-functional logistics terminal is
to be built in Seki near Nagoya. This logistics terminal is referred to as a logistics town and
has various functions such as the transhipment of goods, assembling products during
distribution, warehouses and wholesale markets. This project is being planned and executed by
a group of companies from various kinds of industries with the support of the national,
prefecture and municipal governments.

1.5.5 Load Factor Controls

Controlling the loads of pickup/delivery trucks is a relatively new initiative compared with
conventional regulations such as vehicle weight limits, designated times for trucks to enter city
centres and the control of vehicle emissions. Two European cities (Copenhagen and
Amsterdam) introduced a certificate system for freight carriers who deliver or collect goods
within the central city areas in 1998. In Copenhagen, only vehicles with a certificate (green
sticker) are allowed to use public loadinghnloading terminals in the inner city. This certificate

(Introduction) 7
can only be issued to vehicles satisfying the following two conditions:
(a) Load factor over 60 percent
(b) Vehicle age less than 8 years old
Companies owning vehicles are required to produce a report on the load factors of their
vehicles every month. To maintain certification, a company must have an average load factor
during the previous month above 60 percent. In Amsterdam vehicles weighing over 7.5 tons
are not permitted to use streets other than main streets. However, vehicles weighing over 7.5
tons are able to obtain a special certificate to enter these streets, if they satisfy the following
three conditions:
(a) Load factor over 80 percent
(b) Length less than 9m
(c) Engine must satisfy Euro 11 emission standards
The police inspect the load factor of specific vehicles on the road. This initiative assumes that
higher load factors produce lower the environmental impacts.

1.5.6 Underground Freight Transport Systems

Underground freight transport systems are innovative solutions for urban freight transport
problems. Koshi et aL(1992) estimated the impacts of building an underground freight
transport system in the central area of Tokyo, Japan. The results of this study indicate that NO,
and COZ emissions would be reduced by 10 percent and 18 percent respectively and that
energy consumption would be reduced by 18 percent and the average travel speed would be
increased by 24 percent. Ooishi and Taniguchi (1999) studied the economic feasibility of the
underground freight transport system in Tokyo and concluded that this project has an internal
income rate of 10 percent when the infrastructure is constructed by the public sector.
The Dual Mode Truck (DMT) was developed and tested by Public Works Research Institute of
the Ministry of Construction in Japan. This new type of automated electric truck can travel
through an exclusive guided lane in underground tunnels with the external supply of electricity
and also travel on normal streets operated by a driver with batteries. In the Netherlands a
similar idea was proposed (Visser 1997, Duin 1998) and the feasibility of underground freight
transport system between Aalsmeer and Schiphol airport for carrying flowers was investigated,
An automated guided truck named the Combi-road system was also developed and tested by
a group of private companies (see Section 9.3).

8 (City Logistics)

1.6 MODELLING
1.6.1 Introduction
Quantification of the consequences of City Logistics initiatives is necessary for their evaluation
and planning. Predicting the impacts of City Logistics initiatives for evaluation purposes
requires modelling to be undertaken. Models should describe the behaviour of the key
stakeholders involved in urban freight transport. They should also incorporate the activities of
freight carriers including transporting and loading/unloading goods at depots or customers.
Models must also describe the traffic flow on urban roads for freight vehicles as well as
passenger cars. Models are also required to quantify the changes in costs of logistics activities,
traffic congestion, emissions of hazardous gases, and noise levels etc. after implementing City
Logistics initiatives.
Modelling City Logistics is a challenging exercise, since there are many complicated logistics
activities for each of the stakeholders as well as many different evaluation criteria available for
assessing the impacts of City Logistics initiatives. Therefore, the modeller must be very careful
about what sort of activities of the stakeholders should be considered and what sort of
evaluation criteria should be predicted. In addition, modelling transport on road networks is an
important component of City Logistics models. However, freight vehicles represent only a part
of total traffic on urban road networks. City Logistics models need to consider both freight and
passenger vehicles and focus on the impacts produced by the freight vehicles. This requires a
separate treatment of freight and passenger vehicles in the formation of origin and destination
matrices and traffic assignment.

1.6.2 Limitation of Models

At the moment, models are limited in their ability to quantitatively predict all the impacts of
City Logistics measures. This is because urban goods movement is a very complex system with
many stakeholders. Existing mathematical modelling approaches that have developed are
currently not sufficient for fully describing entire urban transport systems. In particular: the
interaction between stakeholders is not well represented in existing models. Another difficulty
encountered when modelling City Logistics comes from the fact that City Logistics involves
the economic activities of private companies and vehicle flows on road networks. It is quite
difficult to deal with both aspects of logistics activities in a single model. In addition, model
calibration and validation is not easy in real situations, because of the lack of appropriate data
describing current goods movement patterns.

(Irirroductioii) 9

1.7 SYSTEMS
APPROACH
1.7.1 Overview
Urban freight systems are complex, with numerous components and interactions. The systems
approach is a methodology for defining problems and determining solutions (Figure 1.2). It
provides an analytical framework for modelling and evaluating City Logistics schemes. This
approach defines a process consisting of a number of related activities for investigating urban
freight problems.
The role of the analyst is to assist decision makers in making informed choices by providing
technical advice (Sudgen and Williams, 1978). This involves determining an option that is
most consistent with the decision makers objectives.
The systems approach identifies the best way to use limited resources to achieve stated
objectives. It is a continuous creative problem solving process that relies heavily on modelling
and data collection to aid the task of reasoning.
Illustrations of how ITS have been developed for City Logistics will be used to demonstrate
how the systems approach can be applied to help solve urban goods distribution problems.

Problem
Definition

I
I

Models

Objectives

Criteria

I
I

4 4 I

t
Constraints

1
Alternatives

t
Implementation

Selection

Evaluation

Figure 1.2 A Systems Approach to City Logistics

10 (City Logistics)

Only an overview of the general concepts associated with the systems approach is presented in
this section. It is important however, that these concepts be applied when seeking solutions to
specific urban freight problems, since each city has its own peculiar social, economic and
environmental background that warrants careful study and analysis.
The systems approach initially focuses on defining the problem, then specifying objectives
and determining criteria. The next stage involves consideration of any constraints and
available resources that allows the range of alternatives to be generated and an appropriate
level of data collection and modelling to be determined. Models have a central role within the
systems approach. They are used to predict the performance of the alternatives. Here,
estimates of future demand, supply and impacts are produced. The sensitivity of these effects
to any assumptions should next be investigated. Alternatives are then evaluated on the basis of
their consequences. The selection procedure determines which alternative is chosen for
implementation. After the implementing the chosen alternative its performance is reviewed.
This feedback link involves checking whether the initial problem has been solved and
objectives attained. If the initial problem has not been solved it may need to be redefined or
the original objectives modified. In either case, the process continues until the initial problem
has been solved and objectives achieved. The process often continues since new problems
emerge and revised objectives are determined.
There are a number of issues within each of the activities of the systems approach that need to
be addressed to provide a rational basis for evaluating City Logistics schemes.
1.7.2 Problem Definition
Problems arise because of the difference between the actual and desired state at a particular
point in time. Defining problems therefore, involves firstly specifying the perceptions and
interpretation of the actual situation of all interested groups. This requires identifying the key
stakeholders and the issues that effect them.
Inputs from all who have an interest in the outcome of an issue should be gained. This
establishes interfaces between the various interest groups. By consulting all stakeholders
involved in urban goods distribution, the potential to shift problems between them can be
reduced. This also helps educate the stakeholders about the problems of other groups and
wider issues.
Quantification of the problem is necessary to provide an objective basis for clarifying issues.
The need to consider broader, wider problems and issues is important in City Logistics.

(Introduction) 11
Common urban freight problems include:
(a) Traffic congestion (delay)

(b) Fleet planning & management (capital and operation costs)

(c) Environmental impacts (air quality, noise and crashes)

1.7.3 Objectives
Objectives are established to provide a direction relating to the outcome of the proposed
scheme. This ensures that the successfulness of schemes can be checked after they are
implemented. Rational planning is impossible unless what is trying to be achieved is clearly
stated.
Typical City Logistics objectives are to:
(a) Reduce operation costs
(b) Increase efficiency
(c) Reduce environmental impacts

1.7.4 Criteria
Criteria measure the performance of the system. For each objective a measure of effectiveness
should be determined.

(a)
(b)
(c)
(d)

Number of trucks (vehicles)

Load factors
Average speed (km/h)
Vehicle kilometres of travel (VKT)

1.7.5 Resources
Resources are the inputs available for the project and are usually financial, physical or human.
The scale of these should be determined early within the process as they can directly influence
the amount of investigation in the form of data collection, modelling and evaluation that can
be undertaken.
The level of human expertise and skills available should be identified. A budget can also assist
in identifying the required levels of resources.

12 (Cig Logistics)

Common resources for City Logistics schemes include:

(a) Transport Infrastructure
(i) roads
(ii) terminals
(b) Telecommunication infrastructure
(i) internet
(ii) satellite (e.g. GPS)
(c) Public sector
(i) sponsorship
(ii) co-ordination
(d) Private sector
(i) Technology
(ii) Marketing
(iii) management

1.7.6 Constraints
It is essential to consider any constraints that may restrict the level of resources available or
unacceptable outputs from the project. Often financial, legal, social or political issues limit the
range of alternatives that can be considered:
(a)
(b)
(c)
(d)

Availability of specific resources required

Relevant regulations and standards
Potentially unacceptable side effects
Legislation and regulations relating to competition and privacy

1.7.7 Alternatives
Alternatives are the options having the potential to solve the problem. A wide range of
different options should be generated. This requires the analyst to be creative and innovative.
Often technology or regulations provide potential effective City Logistics initiatives.
There are a number of ITS based City Logistics schemes that have already been implemented
in cities through the world (see Chapter 3):
(a) Electronic tolling

(Iizrroductioiz) 13

(b) Matching systems for back-loads

(c) Booking systems for access to terminals
(d) Real time vehicle tracking and monitoring
(e) Performance monitoring (e.g. travel times, speeds and weights)
(f) Computerised vehicle routing and scheduling systems
(8) Route guidance

1.7.8 Data Collection

The quantification of the usage and performance of the existing system often requires
extensive number of surveys to be conducted. A range of data is often required to provide a
rational basis for decision making:
(a) Problem identification
(i) existing conditions
(ii) stakeholder issues (Larraneta, 1999)
(b) Modelling
(i) system description (land use & transport networks)
(ii) evaluation
(c) Monitoring and review

1.7.9 Models
Models are used to provide a simplified representation of urban freight systems. Computer
based procedures based on mathematical relationships are typically used to predict the
performance of City Logistics schemes. They allow the effects of various changes in the urban
freight system to be estimated without actually changing the system (i.e. implementing the
scheme).
Techniques that provide quantitative estimates of the benefits and costs of City Logistics
schemes are required to be produced. Current approaches and applications are described in
later chapters. There are 3 types of network models that produce information that can be used
for evaluation purposes:
(a) Demand
(i) goods
(ii) vehicles
(b) Supply

14 ( C i q Logistics)
(i) travel times
(ii) reliability
(c) Impacts
(i) environmental
(ii) economic
(iii) energy consumption
(iv) social
(v) financial

1.7.10 Evaluation
Evaluation involves the methodical comparison of the alternatives, based on economic, social,
financial, energy consumption and environmental grounds. The predicted consequences for
each alternative are compared.
Multi-attribute analysis techniques that highlight the trade-offs are often used. A quantitative
comparison of options involving a number of factors is often undertaken.
Evaluating ITS for City Logistics, involves a number of issues, including:
(a)
(b)
(c)
(d)

Financial viability (ability to provide a return on investment)

Political acceptability (philosophical support)
Competitive neutrality (equity to existing and future companies)
Horizontal compatibility (information flow between companies)

1.7.11 Sensitivity Analysis

Sensitivity analysis involves investigating the variability of the predicted effects of the
alternatives with respect to the assumptions made within the models. Common technological
parameters include:
(a) Penetration rates
(b) Error rates (e.g. detection and transmission)
(c) Life of assets (e.g. technology)
(d) Operation and maintenance costs

1.7.12 Selection
After evaluation, an alternative is selected by someone having the power and jurisdiction to
make decisions. Generally, the analyst only recommends the best option but there is no

(Inti-oductioiz) 15

assurance that it will be selected. The following processes often characterise how the selection
procedure is structured:
(a) Tender selection
(b) Contract negotiation
(c) Independent review
In City Logistics, the decision making process is often quite complex, with a number of actors
all influencing each other. The relationships between shipper, carrier and receiver are difficult
to generalise. Organisational structures and management styles vary considerably between
companies, often making it challenging to identify decision makers.
1.7.13 Implementation
Implementing the selected scheme often involves a range of tasks that have to be managed.
Often in the case of City Logistics schemes new operating and organisational procedures have
to be established.
There are a number of issues relating to implementing City Logistics based ITS, including:
(a)
(b)
(c)
(d)

Compatibility of modules (interfaces and information exchange)

Systems architecture
Project management (development)
Operational management (training and technical support)

1.7.14 Review
After implementing the selected scheme it is necessary to check how it is performing. It is
particularly important to determine if the initial problem has been solved and whether or not
the objectives have been attained. This involves monitoring the performance of the scheme.
Often an improved understanding of the problem or the ability of achieving the objectives can
feed back into the process or aid the planning of future City Logistics schemes.

(Modelliizg Ci? Logistics) 17

MODELLING
CITYLOGISTICS
2.1 A MODELLING
FRAMEWORK
Models play a central role within the systems approach to City Logistics (Section 1.7). They
permit estimates of the effects of various changes in the urban distribution system to be made
without actually changing the system.
There are three general types of network models necessary for predicting the effects of City
Logistics initiatives, (a) supply models, (b) demand models and (c) impact models. These three
types of models interact with each other to form an integrated modelling framework (Figure
2.1). Supply models predict the level of service of the freight system based on network
characteristics and demand. Demand models predict the demand for urban goods movement
based on industry and resident characteristics as well as the level of service. Both supply and
demand models are described in detail in Chapter 4. Impact models predict the financial,
energy, social, environmental and economic impacts of City Logistics schemes based on the
predicted demand and level of service. These models are presented in Chapter 5 .

2.1.1 Network Characteristics

The transport network must be described before modelling can be undertaken. Attributes of
the physical and spatial networks that influence costs and performance of freight movement
need to be specified as input.
Geographic information systems (GIS) provide a useful tool for managing the wide range of
data often necessary for City Logistics models. They are described in Section 3.4. A number
of characteristics are generally required:

18 (City Logistics)

C haracterstics

Supply Models

1
Industry &
Resident
Characteristics

3
Demand Models

Figure 2.1 City Logistics Modelling Framework

(a) Traffic network

Links (length, capacity, speed limits, usage restrictions, parking restrictions, geometry
and traffic volumes)
(ii) Intersections (control and prohibited movements)

(i)

(b) Rail network

(i)

Links (distance and capacity)

(ii) Terminal facilities (location and capacity)

(Modelling Cify Logistics) 19

2.1.2 Industrial & Resident Characteristics

A number of characteristics of the land use zones are often required for modelling the
movement of freight within urban areas:
(a)
(b)
(c)
(dj

Type
Location
Density
Development intensity

2.1.3 Supply Models

Supply models are used to predict the level of service provided by the transport system.
Physical characteristics of the transport network are combined with the predicted demand to
estimate the costs of using the network. Common network performance measures include
travel time and operating costs.

2.1.4 Demand Models

Demand models are required to predict the level of demand (vehicles or goods) for the
transport system. Often the temporal variation (within a day) is of interest. Also, predictions
of the demand patterns for various vehicle classes (e.g. passenger cars, light trucks and
articulated vehicles) is usually necessary when modelling City Logistics schemes.
2.1.5 Network Level of Service

The performance of the transport system needs to be determined, including:

(a) Travel times
(bj Operating costs
Here, the costs to users of the transport system are required to be predicted. Estimates of
demand are also necessary since the performance of transport systems often deteriorates as
demand levels increase.

2.1.6 Network Usage

Here, the demand is predicted for the system in terms of vehicle or commodity flows for
specific modes and routes (transport links). This information allows supply models to predict

20 (City Logistics)
the costs for users of the system. Demand estimates are also required for predicting the
impacts of alternatives schemes.

2.1.7 Impact Models

Using the estimated demand and levels of service the effects of City Logistics can then be
quantified. Impacts can be classified as being in a number of categories, including, economic,
financial, social, environmental and energy. A detailed description of impact models is
presented in Chapter 5.

2.1.8 Economic Impacts

Cost benefit analysis can be used to determine the economic viability of schemes over their
life. Typical costs and benefits for City Logistics schemes are listed below:
(a) Costs
(i) acquisition (capital equipment and facilities)
(ii) operation
(iii) maintenance
(b) Benefits
(i) reduced travel times
(ii) reduced operating costs (fuel consumption, tyres, repairs and maintenance)
(iii) reduced accident costs (including personal, vehicle repair and incident)
These costs and benefits are usually combined to estimate the net benefits of the project using
discounting techniques. The benefit cost ratio (BCR) and net present value (NPV) are usually
estimated to determine the economic viability of projects and compare the economic impacts
of City Logistics schemes (Ooishi and Taniguchi, 1999).

2.1.9 Financial Impacts

Predictions of the impacts on revenue and costs need to be estimated for companies involved
in City Logistics schemes. A number of criteria are generally used to determine the financial
viability, including:
(a) Net present value (NPV)
(b) Internal rate of return (IRR)
(c) Payback period
City Logistics schemes often lead to reduced costs for carriers, in the following areas:

(Modelling C i q Logistics) 21

(a)
(bj
(c)
(d)
(e)

Wages
Registration & insurance
Capital
Fuel &oil
Repair and maintenance

(f) Tyres

2.1.10 Social Impacts

There are often a number of impacts relating to equity for various user and non-user groups,
including:
(a)
(bj
(cj
(d)

Business competition
Crashes
Accessibility
Aesthetics

2.1.11 Environmental Impacts

There are numerous undesirable negative effects from urban goods movement that can present
a direct risk to human health. Modelling is required to estimate changes in pollutant emissions,
noise and vibration levels.
Although, trucks typically account for a small percent (eg. 10%)of total urban traffic (vehiclekilometres) they often produce a substantial percent of total emissions. Greenhouse gases
produced from exhaust gases of trucks are a major concern in many cities. Depending on the
engines and fuel used by vehicles, emission levels need to be estimated for the following
pollutants:
(a) Carbon monoxide (CO)
(bj Oxides of nitrogen (NO,)
(cj Suspended particulate material (SPMj
(d) Hydrocarbons (HCj
Residents are often impacted by the noise produced from trucks. Sleep disturbance is of major
concern in urban areas. Noise levels as well as the number of exposed persons should be
predicted when evaluating City Logistics schemes.

22 (CiqLogistics)
City Logistics schemes often involve changes in the total number and types of trucks used to
distribute goods in urban areas. This can effect the frequency as well as the level of noise
produced by trucks.
There are various criteria used to measure noise levels using the weighted A decibel scale:

o level exceeded 90% of the time

(a) L ~ noise
(b) L,, equivalent sound level: average over a specified period of time
2.1.12 Energy Impacts
Energy conservation is an important issue since there is only a limited amount of natural
resources available. City Logistics schemes have the potential to reduce the total amount of
fuel consumed by trucks. Models are required to estimate the various types of fuels consumed
when distributing goods:
(a) Petroleum - leaded & unleaded
(b) Automotive diesel oil (ADO)
(c) Liquid petroleum gas (LPG)/Liquid natural gas (LNG)

2.2 THEMODELDEVELOPMENT
PROCESS
This section discusses a number of issues that need to be considered when constructing a model
to evaluate City Logistics schemes. A general process is also described for developing models.
Models should not be overly complex. The degree of detail should reflect the objectives of the
model. Increasing the number of factors can lead to larger estimation errors and increase the
number of assumptions. A good model is able to represent the system of interest with the
minimum amount of complexity.
Network models should be able to assist decision makers by determining the range of impacts
for alternatives. The level of accuracy ought to allow the effects of different options to be
compared. Models should also be sensitive enough to allow the performance of various
options to be determined. To be useful for decision makers, models must provide information
relating to the trade-offs between options being considered.
Models are an integral component in the systems approach to City Logistics (Section 1.7). The
process for developing a model involves a number of activities (Figure 2.2).

(Modelliizg C i 8 Logistics) 23

I
I
I
I

I
I
I
I

I
I
I

I
I

I
I
I
I
I
I
I

I
I
I
I
I
I

I
I
I
I
I

I
I

I
I

Data
Collection
L

-b

I
I
I
I
I
I

I
I
I

I
I
I
I

I
I
I

Systems
Synthesis

Software
Development

-I
I
I

9:
Verification

I
I

I
I

Resources

I
I

--:

d
Validation

------a

Figure 2.2 A Model Development Process

A brief description of the major tasks that need to be undertaken within each activity is
described in this section. This process can be used for developing models for predicting the
effects of City Logistics schemes.

2.2.1 Problem Definition

The initial stage when developing a model involves defining the problem. Here, the need for
predicting the effects of a City Logistics scheme must be clearly defined. The purpose of the

24 (Civ Logistics)
model should be stated since the model should be developed to address a specific set of
questions.
A model should be responsive to the policy variables of interest. Therefore, the modeller
needs to identify the range of options that require evaluation.
The role the model will have within the decision making process should be considered. It is
also useful to identify the decision makers who will use the results produced by the model.
The system boundaries or limits of the system should be drawn wide enough to incorporate all
significant effects. Although they are usually set by intuitive assessment, system boundaries
generally have 3 dimensions (Figure 2.3):
(a) Spatial: geographic boundaries (where?)
(b) Stakeholders: population groupings (who?)
(c) Impacts: range of effects (what?)
There are numerous elements within each dimension. Note, the order of the elements in Figure
2.3 does not indicate their relative importance.

Stakeholders

Administrators

- Community
- Carriers
- Shippers
/
Terminal

Link

Area

Corridor

City

Environmental

Impacts

Figure 2.3 System boundaries

Spatial

(Modelling Ciq'Logistics) 25
At this stage issues relating to the need for a model to be constructed should be addressed. The
nature of the policy variables as well as the range of alternatives to be evaluated should be
specified.
The potential role and utility of the model needs to be examined. This requires consideration
of the decision making environment within selection stage of the systems approach (Section
1.7). The potential users of the model as well as the information produced from it should be
identified.

2.2.2 Objectives
Next, the objectives of the model ought to be determined. Here, the general results the model
is expected to produce should be stated. This involves determining the performance
requirements of the model.

2.2.3 Criteria
Criteria define the measures of performance that will be produced by the model. The model
should provide information that is useful to decision makers. It is necessary to determine the
specific types of output that the model needs to produce, including, statistics, confidence
intervals and levels of significance.

2.2.4 Systems Analysis

Systems analysis involves identifying the essential components and interactions within the
system being investigated. The major factors and direct relationships need to be identified. This
often involves collecting data. Influence diagrams and activity cycle diagrams (Pidd, 1996) can
be useful tools for describing the logic of processes within complex systems.

2.2.5 Systems Synthesis

Systems synthesis requires the factors and relationships identified in the systems analysis
stage to be represented using mathematics. Variables and equations are used to formulate the
model. Relationships between variables are represented using closed form analytical
equations.
The nature of variables also needs to be determined:
(a) Deterministic -where there is only one possible outcome
(b) Stochastic/random - where a number of outcomes are possible

26 (Cia> Logistics)
(c) Exogenous - determined outside the model
Endogenous - determined within the model

(d)

2.2.6 Software Development

The next step in the model development process involves producing computer based
procedures or programs to allow the mathematical and logical expressions defined in the
systems synthesis stage to be used to produce quantitative results. Software procedures for
determining solutions must be constructed.
There are several types of software tools available for modelling:
(a) Programming languages (e.g. Fortran, Delphi, Visual Basic or C++)
(b) Generic packages
(i) DBMS (e.g. FoxBase)
(ii) spreadsheets (e.g. excel)
(c) Specialised modelling software
(i) simulation (e.g. SIMSCRIPT and MODSIM)
(ii) systems dynamics (e.g. STELLA and I THINK)
There are a number of aids that can assist in the development of code, including, pseudo code
and flow charts.
Various types of information are generally provided as input to City Logistics models,
including:
(a) Transport network
(b) Parameters
(c) User characteristics
These are usually specified using files perhaps generated using other software procedures (eg.
GIS or DBMS).
Output from the model should be presented in an understandable and useful form to decision
makers. There are a variety of forms of output:
(a) Dynamic (e.g. animation)
(b) Graphical (e.g. maps and charts)
(c) Files (e.g. text)

(Modelling C i q Logistics) 27

2.2.7 Verification
Verification of the software is necessary to ensure that an adequate representation of the
system has been achieved. Here, the procedures are tested and checked for correct logical
structure and the output compared with anticipated behaviour.

2.2.8 Validation
Validating a model involves comparing the results produced by the computer procedures with
the actual conditions experienced in the real world. This stage assesses the models ability to
mimic reality. A range of surveys are usually required to measure the performance of the
system. Intelligent Transport Systems (ITS) often provide a rich source of data from which it
is possible to validate City Logistics models. Existing models can also be used to check the
models output.

2.2.9 Application
Application of the model involves using the model to produce estimates of the performance
measures and other effects that will be used in the evaluation of alternatives. Here, the effects
of the alternatives being considered are predicted using the model. The models parameters and
input variables can be changed to allow sensitivity analysis to be undertaken.

2.3 MATHEMATICAL
PROGRAMMING
Mathematical programming provides a concise, comprehensive and clear statement of the
problem that the model will address. It encompasses the formulation of a model to represent
the system under investigation.
Model formulation involves both the systems analysis and synthesis stages of the model
development process. It involves identifying and specifying 3 basic elements of a system:
(a) Decision variables
(b) Constraints
(c) Objective function
Decision variables are the unknowns to be determined by the model. Constraints need to be
specified to represent resource limits, unacceptable system outputs or to ensure that the
decision variables are restricted to their feasible (or permissible) values. The objective
function measures the effectiveness or performance of the system. It acts as an indicator for

28 (CivLogistics)
assessing the achievement of a solution. The general structure of a mathematical program is
shown below.

2.3.1 Program Structure

Determine the optimal values of the decision variables, xj, j = 1, ... ,n
Optimise

xo = f(x,, x2 ,.... x, )

Subject to

g,(x,, x2 ,...., x, ) I bi

XJ

2 0

i = I , ..., m

j = 1, ..., n

Solution procedures must be applied to identify the optimal solution. In logistics modelling
the objective function is usually related to minimising costs.

2.3.2 The Vehicle Routing Problem with Time Windows

The vehicle routing problem with time windows (VRPTW) is a common application of
mathematical programming in business logistics. It aims to identify the routes and schedules
of trucks to minimise costs within a framework of customer defined time windows.
The VRPTW can be defined as follows. Let G = (Vo, E ) be a graph where V,

={iso,

I,,,

....vN+,/

is the vertex set, V = V o / [ v ~ )and

, E = {(i.),v,): I,,, ~j E V,, i#j/ is the arc set. Every vertex in V
corresponds to a customer to be serviced and vo denotes where the vehicles are based. Defined
on E is the matrix T = it,,) describing the travel times for traversing between the vertices.
Associated with each customer is time window [f,,re,] where r', and t', are the earliest and
the latest arrival time respectively, within which the customer should be visited.
Chapter 6 describes a mathematical model of the VRPTW. The model minimises the total cost
of distributing goods with truck capacity and designated time window constraints. The total
cost function is composed of two components, the fixed cost and the operating cost of
vehicles.

2.4 OPTIMISATION
The concept of optimisation in modelling is generally based on identifying the besr solution
for a specified system. In classical optimisation, the best solution is the exact solution to the

(Modelling Gin. Logistics) 29

formulated problem. However, as models are formulated to incorporate more realism and
complexity, the best solution often needs to incorporate the cost of obtaining a solution and
this may not be the exact solution.
Since models are only simplifications of the real world, many systems are formulated so that
solution procedures can be applied to ensure that exact solutions can be identified. For
example, many systems are formulated as linear programs and solution procedures such as the
simplex method are then used to find the optimal solution to the formulated model.
Mathematical programs aim to identify the best value of the objective function. For some well
defined mathematical programs, classical calculus based approaches can be used to obtain
exact solutions. However, the models formulated for many practical problems, solution
procedures do not exist for ensuring that the exact solution can be identified. In many cases
the initial model is reformulated using approximations and simplifications to allow an
established solution procedure to be used that guarantees the exact solution for the revised
model can be found. Otherwise heuristic solution procedures are used to find a good solution.
There have been a variety of methods proposed for determining the exact solutions for a
number of basic versions of the vehicle routing problem (Toth and Vigo, 1998). Branch and
bound algorithms, Lagrangian relaxation, branch and cut methods as well as set partitioning
formulations have been developed for solving these difficult combinatorial optimisation
problems. However, even the most effective methods are limited to a small scale simplified
problems.

2.5 SIMULATION
Simulation is a modelling technique for conducting experiments on a digital computer, which
involves using mathematical relationships to describe the behaviour and structure of a
complex real-world system over extended periods of time. It allows a realistic representation
of the random components of a system to be constructed.
Simulation modelling involves breaking a system down into simple components and
modelling each component as well as the interactions between components and then operating
the model. Random phenomena can be represented using statistical distributions.
Microscopic models represent each unit individually whereas macroscopic represent units
collectively as flows or platoons. However, micro simulation models are widely used in the
management of terminal facilities.

30 (Civ Logistics)
Due to the complexity of urban traffic systems and the number of vehicles using the system, a
combination of both microscopic and macroscopic models provide a more practical modelling
tool for investigating the effects of City Logistics schemes (eg. modified Box Model, Section
4.5).
There are various types of simulation models:
(a)
(b)
(c)
(d)

Physical (eg. scale models)

Analogue (eg. hydraulic mechanics applied to traffic engineering)
Mathematical (ie. symbolic or algebraic relationships)
Computer (digital representation of mathematical models)

However, computer simulation models have a number of advantages, including:

(a)
(b)
(c)
(dj
(e)

Realism - can represent dynamic, interactive, random and complex systems

Transparency - easy to understand and explain
Validation - at the module level
Statistical reliability - estimates of the variability of output are provided
Understanding - analysis of sub systems and relationships

However, there are also a number of disadvantages related to simulation modelling:

(a) Understanding
(i) lack of data
(ii) knowledge
(b) Expense
(ij data collection (required for specification. testing and validation)
(ii) multidisciplinary skills (incl. statistics. computing and engineering)
(iii) computing resources (fast, interactive and graphics)
(c) Time
(i) teams
(ii) comprehensive development process

To simulate logistics systems a high degree of data and knowledge is required, since, we often
fail to realise how lirtle we know about a thing until we attempt to simulate it on a computer
(Donald Knuth).

(Modelling Ciw Logistics) 3 1

Simulation modelling involves logging the progression of units within the system. Time
progression within simulation models is based on either fixed time intervals or when discrete
events are predicted to occur.
Algorithmic recursive computer based procedures are typically used for generating pseudorandom numbers that are converted to random deviates using statistical distributions to
replicate random phenomena.
With fixed time simulation models the time is advanced in constant intervals. Changes are
analysed during the fixed time intervals. However, with event based simulation, time moves
forward in unequal amounts depending on when events occur. Events are the times when the
states of units are changed.
An executive routine is used to manage the calling of procedures and to order events. This

needs to know what state the unit is in and when this will change.
There is usually a period of time before simulation models reach equilibrium conditions. This
is called the warm-up period and results produced during this period are normally discarded.
Most simulation models either represent a specific interval of time (e.g. peak hour or 12 hour
period) or terminate when a specified event occurs (e.g. no more trucks to unload). Since each
run represents an experiment; the output from simulation models often requires statistical
techniques. Confidence intervals can be estimated.
Simulation models have typically applied to predict the performance of toll facilities, weigh
stations (Glassco, 1999), traffic signals and terminals (Young et al, 1988).
When evaluating simulation models there are a number of factors that should be considered:
(a) Validity (adequacy of system representation)
(b) User interface (ease of use)
(c) Data collection (surveys required)
(d) Calibration (parameters)
(e) Execution time
(f) Output (formats)
(g) Behavioural rules (traffic)

32 (CihLogistics)

2.6 META-HEURISTIC
TECHNIQUES
Heuristic techniques are solution procedures that seek to find good solutions at a reasonable
computation cost. They provide a practical means of obtaining solutions for complex model
formulations. However, heuristic techniques do not guarantee to find the exact solution.
Heuristic techniques do not require any simplifying assumptions relating to the formulation of
a model. They are flexible regarding the nature of the objective function and constraints.

In many modelling exercises there exists a trade-off between obtaining an exact solution to an
approximate model or an approximate solution to an exact model. The attainment of the exact
solution is much less important when modelling complex urban goods distribution systems.
There have been a number of general heuristic (meta-heuristics) methods that have been
applied to logistics problems (e.g. vehicle routing and scheduling). These include genetic
algorithms, tabu search and simulated annealing.
Meta-heuristic techniques have been applied to investigate the effects of a range of City
Logistics schemes, including:
(a) Increasing the width of time windows on routing and scheduling costs (Section 6.3)
(b) Probabilistic routing and scheduling (Section 7.2)
(c) Location of public logistics terminals (Section 8.4)
Many heuristics use the concept of neighbourhood when generating solutions. A
neighbourhood is the set of solutions that can be formed from the current solution by a simple
operation. The nature of the problem usually determines how the neighbourhood set is
constructed.

2.6.1 Genetic Algorithms

Genetic algorithms (GA) have been described as an intelligent exploitation of a random
search (Reeves, 1993). GA are a class of evolutionary computation techniques based on
models of biological evolution. They use selection and recombination in various formulations
to sample a (coded) search space.
Each individual in the population represents a possible solution to the problem. A fitness value,
derived form the problems objection function and reflecting some of the constraints is
assigned to each member of the population. The search process proceeds by maintaining a

(Modelling Citj Logisticsj 33

population of structures from which new structures are created. GA were originally developed
by Holland and his colleagues at the University of Michigan in the 1960s.
GA work by maintaining a set of solutions (population), from which a next set of solutions
(generation) is produced using a number of operators. Selecting those solutions within the
population that will be used to form the next generation must be determined. This is usually
based on the fitness of solutions which is related to the value of the objective function.
Crossover involves combining characteristics of current solutions to form new set of solutions
that form the next generation. It is also common for a proportion of solutions to be randomly
modified (or mutated) to increase the range of the solutions and decrease the chances of
premature convergence. Constraints can be incorporated by including a penalty function within
the objective function (Goldberg, 1989).
The performance of GA largely depends on the crossover and mutation operators, which
determines the solutions sampled within the solution space. Determining the best type of
operations is difficult, since diverse schemes can lead to promising regions of the solution
space, but over diversification degrades to an uncontrolled random search.
Typical GA parametres are:
(a) Population size (number of solutions in each generation)
(b) Mutation rate (proportion of elements randomly swapped)
(c) Crossover rate (proportion of solutions combined)
As well, there usually is a range of possible methods for performing crossover, mutation and
selection operations.
GA allows solutions for complex, non-linear mathematical programs to be found. They have
been applied to a range of logistics problems, including travelling salesman problem, vehicle
routing problems (Chapter 6), location of logistics terminals (Chapter 8).
Traditional coding of solutions as binary strings is not well suited to the TSP. There is also a
need for specialised crossover and mutation operators. The most natural and popular way of
coding a solution to a TSP is the path representation in which every chromosome is a string of
numbers representing the sequence of customer visitation. A number of crossover and mutation
operators have been developed based on this representation. Each can have a substantial effect
on performance for different types of problems (Potvin, 1996). GA has also been applied to a
variety of vehicle routing and scheduling problems (Osman and Kelly, 1996 and Gendreau et
al., 1997).

34 (City Logistics)

2.6.2 Tabu Search

Tabu search involves the neighbourhood of the current solution being examined for the next
solution but certain moves are considered to be tabu or forbidden (Glover and Laguna, 1993).
The tabu list or set of moves that are not allowed is updated as the search progresses and
relates to the history of the moves previously made. This attempts to avoid repetition.
At each iteration the best admissible neighbour is accepted even if this is inferior to previous
values of the objective function. Neighbourhoods in the travelling salesman problem can be
formed by removing and replacing links from an existing tour. If two links a removed there is
only one way of reconnecting the tour. Tabu search has been applied to vehicle routing and
scheduling problems (Taillard et al., 1997).

2.6.3 Simulated Annealing

SA is a neighbourhood search technique based on the analogy with the physical theory of how
material cools within a heat bath. The laws of thermodynamics describe that the probability of
an increase in energy 6 E at temperature t is given by
p (6E ) =exp (-6 E / k t )

(2.4)

where.
k

: Boltzmanns constant.

The SA technique applies this idea to combinatorial optimisation. That is, the energy in
thermodynamics can be replaced with the objective function of the optimisation problem.
Let us consider whether we should move from the current solution at point A in Figure 6.5 for a
minimisation problem. If the change in the objective function from point A to B, A E , is
negative or 0, the solution moves from point A to B, since the neighbourhood solution is
smaller than the current solution or equal to it. Let the current solution be at point C. When the
change of objective function A E from point C to D is positive, the solution can move at the
probability given by the following equation, although the objective function at point C is
smaller than that at point D.
p (A E ) =exp (-A E I T )

where,

T :temperature

(2.5)

(Modelling C i q Logistics) 35

A \
B

variable
Figure 2.4 Move in simulated annealing

SA makes it possible to move away from the local optimum (point C) and reach the global
optimum (point E) by allowing moves such as from point C to D.
For the cooling method to decrease the temperature T , either of following two equations
(equations 2.6 or 2.7) is normally used:

T(i)=log(i + I)
or

T ( i + 1)= j3 T ( i )
where,

a:j3 : cooling parameters

i : iteration number.

This allows the probability of accepting an inferior solution to reduce as the search progresses,
permitting the search to move widely in the early stages and become more restricted at is
proceeds.
Simulated annealing has been successfully applied to the travelling salesman problem (Bonomi
and Lutton, 1986) and vehicle routing problem with time windows (Chiang and Russell, 1996).

36 (Cit), Logistics)
SA was used to investigate the effects of relaxing time windows (Section 6.3). It was shown
that a considerable reduction in total delivery costs, travel and waiting times as well as the
number of trucks required could be gained by increasing the width of customer time windows.

2.7 MODELLING
ACTORSPERCEPTIONS
As the role of logistics and transport have become more important for companies (Heskett,
1977; Bowersox et al., 1986), this has caused enormous growth in the establishment of
transport and logistic services, with the relevance for society also increasmg due to their
economical value adding capabilities and their impact on spatial and environmental conditions.
Also, in recent years, many companies have opted to focus on their core activities, in line with
their comparative advantages and technological competencies. The competitive position of the
individual companies is closely related to the company's position in its network of suppliers,
authorities and customers. Current policy documents suggest that the period of network
economy has started (EZ, 1999). As a consequence of these broad (external) influences, the
number of stakeholders, either directly or indirectly involved, grows. Not only the actors
representing the logistic processes themselves are important players, but also the interaction
between stakeholders. who are not directly related to the logistics processes, but who can raise
strong restrictions to the implementation of logistic concepts. These influences, are sometimes
hard to foresee, make us believe that a different approach for logistic concept development is
needed. DeBruijn and tenHeuvelhof (Bruijn et al., 1995) identify four important characteristics
of stakeholder networks:
(a) Interdependence
Actors cannot achieve their goals without co-operation, since they are dependent on the
resources (e.g. statutory powers, information or funds) of other actors.
(b) Pluriformity
Corporate actors do not behave as individuals, but as coalitions, since their constituents may
have diverging interests.
(c) Self-containment
Corporate actors are inclined to close themselves off from their environment, developing
their own frame of reference and norms, making it harder to induce their co-operation.
(d) Instability
Positions and relations in policy networks are continually undergoing changes

(Modelling Ciq, Logistics) 37

Each of these characteristics can be seen as impeding factors in decision-making, but also as
enabling factors. Taking account of these stakeholder characteristics, implies a strong change
for the development of logistics concepts. This needed change is a consequence of what is well
expressed by Thomas (Thomas, 1966):

The enviroizmeizt by which (man) is iizflueizced and to which he adapts himself- is his world,
not the objective world of science - is nature and sociev as he sees them, nor as the scientist
sees them .
Not the way an analyst sees the (interaction between) relevant factors is directive to their
actions, but how actors see these factors themselves. To be more precisely, in their acting,
actors reason from their own subjective, possibly incomplete or even blatantly incorrect
perceptions, and they make these perceptions real as it were. This has been powerfully worded
in the Thomas-theorem: If man define situations as real, they are real in their consequences.
The Thomas-theorem implies that an analyst should not pursue the ultimate model of a
situation, but a rich set of models that reflects the diversity in actor perceptions. The soundness
of this set of models is determined not by the degree of correspondence with reality, but by the
acuteness with which it mirrors the assumptions the actors make about this reality - their
reality. This interpretation of decision modelling has led us to a modelling approach we call
dynamic actor network analysis (DANA) (Bots, 1999; Twist, 1998; Duin, 1998).
2.7.1 Perception Based Modelling
Perception based modelling focuses on how actors think and make their decisions. To achieve
their goals, actors think strategically, anticipating on the goals and moves of other actors in the
network. The more they know about dependencies between or within other actors, about their
frame of reference, and about impending changes, the more successful their strategy is likely to
be. Due to this way of thinking and reasoning, studies of decision making in networks have
caused a fundamental change in thinking about the role of decision makers, stakeholders and
analysts (Fischer, 1993). The image - illusion perhaps - of decision making as a (bounded)
rational design effort by an elite group with implementing power is replaced by an emergent
perspective on policy making in which policy is seen as the result of interaction in a network of
corporate actors (Marin, 1991). Based on this insight we have made an overview of analysis
and modelling techniques and classified them by increasing complexity:
(a)
(b)
(c)
(d)

Mono-actorhfulti-factor
Multi-actorhfulti-factor from the outsiders perspective (birds eye)
Multi-actorhfulti-factor from individual stakeholder perspectives (mouse-eye)
Multi-actorMulti-factor with hidden agendas.

38 (CiQ Logistics)
Mono-actorMulti-factor (level I)
The first way of analysing decision-making processes (level I) is the orientation from one
actors perspective facing many factors. In fact, this is the most traditional way of supporting
the decision-making. A principal asks to solve his problem, which he is confronted with.
From the principals perspective the problem orientation is mainly focussed on the
understanding of the material complexity, represented in the relations or interdependence
between the factors. Most of the research activities are focussed on defining causalities,
quantifying factors, evaluating and choosing the right options. Participation of other actors is
only advisable if the results of the study indicate that their involvement is necessary. Available
techniques are generally: multi-criteria analysis, cost-benefit analysis, and descriptive
modelling techniques for describing the causalities. As an example of the descriptive modelling
techniques, the flow models of Meadows (Meadows, 1991) can be mentioned.
Multi-actorMulti-factor from the outsiders perspective (level II)
The next level (11) to be distinguished is the multi-actor or stakeholder environment. Solving
the problems in a purely technical way is not always the best way to do this. Many parties and
actors are involved and should be taken into account. The central research theme in these kinds
of analyses is the identification of the actor-related factors. Which factors are found to be
important by whom? Methods commonly used are network relations models (Lindenberg,
1983), game sessions derived from the game theory (Myerson, 1991) and cognitive mapping
(Eden, 1998). Most of these techniques have a strong background in the social sciences. The
methods are usually applied from one global or central scope and try to formulate one causal
model. Differences in meaning are just explained by the differences of actors interests, instead
of other mental construct thinking of the actors. Therefore, at this level of decision analysis it is
to be concluded that although the techniques provide us a total rich picture with insight into the
actors and related factors, most of the reasoning is based on axioms, which could be far
divergent from several individual actor perspectives. We can say that the obtained insights and
solutions have a bounded rationality, because of this omission.

Multi-actorhfulti-factor from individual stakeholder perspectives (level III)

At this level (111) the main aim is to understand the real perception of the actor. Therefore, we
try to identify its process environment towards other actors, and try to understand its actions
and reactions. If men define things are real, they are real in their consequences, according to
Thomas. In the literature, a lot of research addresses to the same phenomenon: How to define
the individual perception of an actor in order to have insight into his or her strategical
behaviour (Wildavsky, 1993) (Hoppe; 1998). Cultural theory, configuration analysis (Klabbers,
1991), Q-method (McKeown, 1988), depth-interviews and metaphors (Hoppe, 1998) are

(Modelling City Logistics) 39

methods that try to provide insight into the perceptions and interactions. Although these
modelling techniques provide us more detailed insight into the individual behaviours, the
interaction processes between the actors will behave differently. Again we have to admit that
the bounded rationality has returned. This time it is not shown at the analyst level, but at the
individual actor level. Each actor will learn from his experience and try to strengthen its current
position in the interaction processes. Information exchange between the actors is submitted
strategically and selectively in order to influence the perceptions of the individual actor and
vice versa. Therefore we have met the next level of complexity.
Multi-actorlMulti-factor with hidden agendas (level IV)
As we have seen before some actors can behave strategically. Providing selective information
and the use of hidden agendas are tactics of actors to operate strategically and generate an actor
environment with an enormous complexity. Saying this, doing that and thinking such are
key phrases that are characteristic of these environments. At this level the complexity reveals
itself with a high intensity. Ambiguity, uncertainty, interdependence, relations and societal
constraints are the elements that could have an influence on actors, factors and relations among
them. As far as we know techniques to analyse this full complexity are not available. Therefore,
at this level (IV)of analysis researchers and analysts have just one way to handle this
complexity: they have to rely on their gutfeelings.
(City) Logistics deals at level IV
Competition in the freight transport and logistics market is strong and profit rates are low. The
companies in this service-industry behave strategically. As consequence of third-partylogistics contracts have a limited term of validity, which could imply that companies could be
in business competition today, and tomorrow operating together as business partners.
Currently, many companies are taken over as well. These takeovers are results from strategic
reasoning and acting of the main logistics companies. Not only the companies operating in the
logistics market behave strategically, but also even authorities do. Many local governments
are in competition with others trying to attract companies to invest in their area. In order to
have a more comprehensive image of the actors understanding we have to choose the most
detailed level of analysis: multi-actodmulti-factor situation with hidden agendas.

2.7.2 The Dana Toolbox

Some people perceive that analyses based on gut feelings could be typed as rather inferior and
methodologically poor. Associations with crystal balls and playing cards are easily made. Still,
the solutions and results to be found with our gut feelings will provide us many times an

40 ( C i v Logistics)
adequate intervention in complex situations. We cannot always explain why specific
interventions work, but one thing is for sure, it is certainly beyond coincidence.

In order to express our gut feeling analysis more precisely, we have developed a tool called
Dynamic Actor Network Analysis (DANA). Actually, the tool is set up as an open database in
which several perceptions of actors can be submitted. A Windows environment with graphical
support makes the tool user-friendly and suitable for having the causal diagrams (perceptions)
drawn by individual actors. An important feature of the tool is the query generator. All the
individual perceptions can be queried in various ways, which provides us valuable insights, not
only related to the network of actors, but also to sharpen our own reasoning. The underlying
modelling language is based on the network paradigm and embodies concepts from cognitive
mapping and linguistic approaches to approximate reasoning.
The description of a DANAnalysis starts with the all-embracing definition of an arena.
Arena :=(name,(actor), (factor),(relation),(rule), (analyst view))

(2.8)

An arena could be interpreted as the playing field of actors, which implies the actors

themselves, factors or important issues, relations between the actors, and rules to obey. As part
of the arena we also added the analyst perception in order to compare, for instance, the analyst
view with some actor perceptions or with states of a next step game-simulation. The definition
of an actor is represented as:
Actor := (name, {attributel,(private thought),(public iioice))

(2.9)

The attributes have to be interpreted as factors or instruments that are unique for this actor. The
instruments are defined as tools of actors for directed influencing other factors. The actor
definition provides us also the opportunity to deal with the strategical behaviour of an actor, i.e.
the multi-actorlmulti-factor with strategical actor perspectives. The private thought contains a
construct of the main goals and thinking issues of an actor. The public voice is meant to
express the way an actor communicates regarding his goals and thinking. Both elements are
typed as a perception. The description of perception follows the definition of Ramathian and
Hiat (Ramathian, 1989) by making a distinction in factual assumptions, causal assumptions,
and teleological assumptions. In our modelling language we prefer to use a more common
terminology: facts, links and goals:
Private thought := ((factor)3(link), (goal])

(2.10)

Public voice

(2.11)

:= ((factor], {link), (goal))

(Modelling Ci? Logistics) 41

An actors assumptions will always relate to factors, some of which may be defined as specific
attributes of himself or other actors. Factors represent how an actor perceives the current state
of his surroundings:
Factor := (name, state)

(2.12)

State may be specified either as the value associated with the factor, or as an expression
involving a relational operator/director, or as a user-defined operator/director by the analyst.
State := value I (operator/director, value)

(2.13)

Links are causal assumptions in the sense that they represent how an actor thinks changes in his
surroundings to be brought about. These assumptions typically have the form of an implication:
Link := iflfactor, operator, 1,alue) then (certainq, factor* operator/director, i,alue), reiersible
(2.14)
With the term certainty we can ascribe a likelihood to the causal expression. The analyst can
use phrases such as possibly, probably, and certainly. Reversible is a boolean field
indicating whether this causal relation can be turned over. The last elements of the perceptions
are the goals. Goals represent what changes an actor considers to be desirable or undesirable:
Goal := (insistence,factor)

(2.15)

Since an actor may pursue some goals more eagerly than others, we introduce yet another
fuzzy linguistic construct, insistence, for which the analyst may substitute phrases such as if
possible, preferably, or absolutely. The insistence of an actor on a goal is to be used in
evaluation mechanisms whenever the actor has to choose between alternative usages of
instruments.
Relation := {(name, gpe. actorl,actor2, transitiise, reflexive, rei,ersible, i,alue))

(2.16)

Relations can have serious influences on actors, i.e. on actors perceptions. The type of relation
indicates whether a relation is based on hierarchical ideology, or on sympathy or trust. The
attributes transitive, reflexive and reversible are a matter of course. The last term of relation is
value, which could be interpreted as the intensity of this relation. The conceptual modelling
language is still under construction. but a frequent use of this language in our tool gives us a
rich base to testify the expressiveness of language. In the next session we will demonstrate
some topics of the modelling environment on the basis of a city distribution.

42 (CigLogistics)

2.7.3 Perceptions Modelling in City Distribution

The historical centres of our cities offer a convenient atmosphere for shopping and leisure
activities. The quality of the living environment is affected by the noise and air pollution of
trucks. Spinning motors while (un)loading the truck cause emissions of several damaging
environmental exhausts. In the early 90s the concept of city distribution centres became an
issue of growing interest for Dutch municipalities. Centralised distribution, organised at a
distribution centre at the edge of a city, was thought to provide a firm basis to improve the
quality of the inner city. Many positive effects were anticipated from the introduction of the
city distribution centre-concept: congestion reduction, gain of time, reliable delivery,
economies of scale and noise reduction. In 1997 the 'real-world' implementation of city
distribution centres in the Netherlands gave rise to some critical questions about its functioning.
On the whole, it is to be concluded that the total size of demand provided for the city
distribution centre was overestimated. Governmental, logistic, managerial and juridical
incentives have not resulted in a substantial switch towards new distribution concepts, mostly
because the behaviour of forwarders, government and shopkeepers has not been analysed
thoroughly. Superficially evaluating the concept of city distribution centres, one can conclude
that the concept seems to fail.

Figure 2.5 The Municipality's perceptioflublic Voice

(Modelliiig C i n Logistics) 43

With the DANA toolbox we had been able to investigate the perceptions of the actors.
Analysing the differences in perceptions opened our minds for some serious pitfalls in the
policy measures taken by the local municipalities. More precisely, most of the municipalities
introduced in their city time-windows. Only during these specific time frames freight carriers
are allowed to enter the inner city and deliver to clients. Outside these time-windows freight
carriers are not allowed to deliver and they will await a severe penalty if they are caught by the
municipal police. As we can see in Figure 2.5, the perception of the municipality was that this
should lead to an intensified use of the public distribution centre. This statement can be read in
the link statement as follows:

IF <Muizicipalin> does <set time wiizdows> THEN <Use ofyublic D O will <cei-taiizly>
<increase>
Analysing the freight carriers public i,oice perception gives us a quite opposite way of
thinking. Generally the carriers plan the routes most efficiently over more than one city. In
most cities the time-windows are rather identical, which means entering the city after 7 a.m.
and leaving the city before 11 a.m. To meet these time constraints, the carriers are forced to cut
efficient routes in two or more sub-routes. As a consequence of cutting the routes the carrier

Figu c 2.6 Freight Carrier perspective/ Public Voice

44 (Cily Logistics)
will have to use more trucks to satisfy the (same) demand. Besides this effect the congestion
around the cities at this specific time-interval is certainly at a maximum. Therefore, the
introduction of time-windows will lead to a greater use of heavy trucks and will also indirectly
to increased congestion. This thinking is expressed in Figure 2.6.
The link-statements related to the factors Heavy trucks and set time windows are defined as
follows:

IF <Municipality> does <set time windows> THEN <Heaty trucks> will <certainly>
<increase>
In consultation with the municipality the carriers also indicate that use of the public
distribution centre will be more likely when the carriers have to use more heavy trucks. This is
reflected in the next statement:
IF <Carrier.Heavy-trucks> does <increase> THEN <Municipality. Use-of-PublicDC> will
<probably> <increase>
While the instrument is causal related to the individual goals of both actors, there might be
reason to suspect that a conflict situation will happen, especially when the (colouring) intensity
of the goals (see for instance Quality of living environment in Figure 2.5) is high. The intensity
of colouring is representative for the degree of aiming.
Until now we have worked out only a small part of a multi-actodmulti-factor analysis from the
individual stakeholder perspectives (level 111). Evidently we are able to point out some conflict
situations. The strategical roles played by the actors have not yet been analysed. Therefore, we
extend our analysis to a level N-analysis. As a part of this way of analysing the perceptions of
the actors are divided in a private thought and in a public voice. In Figure 2.7 the public and
private perceptions of the carrier are illustrated.
While the competition in the transport market is strong, no carrier is willing to give up some
clients and to experience a reduction in his market share. Their main goal is to maintain their
market share by all means. As a counter reaction to the instrument <set time windows> the
carriers will ignore this measure. By ignoring the time windows they can be more released
from the time pressure and they are able to control their use of heavy trucks more adequately.
The controlling argument is defined as follows:
IF <Carrier> does <Ignore time wiitdows> THEN <Carrier.Heaty-trucks> will <certairzl~>
<slightly> <increase>

(Modelling City Logistics) 45

Figure 2.7 Freight Carrier perspectivePrivate thought

In practice, the conflict really happened. The carriers did not invest in new trucks and did not
want to drop some of their demands. The same amount of trucks was identified on the roads in
and around the cities. Even far beyond the time-windows large trucks were loading and
unloading around the heart of the inner city. Mostly parked at the frontiers of non-authorised
distribution zones, the truck drivers unloaded their trucks by hand with a hand truck. Due to
these situations, the time pressure for the carriers became extremely high. Also the temporary
parking trucks caused a lot of hindrance for local residents and shopkeepers. Also the police
were not prepared for such a workload to regulate and solve the related traffic problems.
With this limited consideration about the use of time-windows in the city distribution we have
showed how we can model individual actor perceptions. If we will take into account the
strategic behaviour of the actors. a level-IV-analysis will be necessary to gain full insight in the
consequences of taking specific measures. Specially, the distinction between what is said, the
public voice. and what is rhoughr, the private thoughts give us opportunities to identify
possible conflicts and contra-reactions between the actors.

46 ( C i y Logistrcs)

2.7.4 Embodiment of Actor Modelling

With the supportive tool DANA we have been able to identify some serious pitfalls in the
introduction of city distribution centres. The graphical interface, i.e. the causal relations
between factors, instruments and goals, helps to clear and sharpen the assumptions of the
actors perceptions. In circumstances with many actors involved, the database firstly serves as
our big memory, and secondly gives us the strong opportunity to ask the individual actor
perceptions for factors and goals in accordance with each others, contradictions in goals, and
possible causal effects by the use of the specific instrument.
During the development of DANA we have thought over the positioning of our conceptual
language versus other techniques. The described levelling with an increasing degree of
complexity gave us a framework to position and to compare decision analysis techniques.
While we are modelling perceptions of multiple actors, the tool has been made applicable for a
multi-user environment. With these extensions our tool can be classified to the field of group
support systems. Knowledge and techniques derived from this soft-OR (Checkland, 1990)
will be incorporated to strengthen tool and methodology.
Reasoning from the Thomas-theorem, it is impossible to develop a general single model that
integrates all the aspects relevant to the actors. Therefore, the development of an overall
methodological transportation framework, containing multiple interrelated models capable of
representing perceptions on qualitative factors and uncertainties, is needed. The development
of an approach to build these interrelation models must fill the gap between logistic decision
making processes on the one hand, and the specification of the design contents on the other
hand. The new models must be easy to use and understand, with user-friendly capabilities, such
as graphical representations of the systems under analysis. Sijbrands (Sijbrands, 1993) clarifies
that the task to support strategic logistic issues seems to be simple; however, the way to
process aizd support is the main research objective to be tackled in future.
Therefore we suggest a new approach that comprehends three interrelated parts: dynamic actor
network analysis, performance measurement and logistics modelling.
Dynamic Actor Network Analysis
The approach starts with the description of the multi-actor situation. For that purpose, we
developed an actor modelling language, which supports the representation of different actorperspectives (Bots, 1999). In general, there is no lack of theoretical notions of networks and
actors within policy science (See also Section 2.7.3). However, there is a lack of practical aid
for empirical research on basis of the actor-model and its network-approach. A simple and

(Modelling Cily Logistics) 47

accessible (ICT-supported) policy-technology for research into networks and actors is in fact
still absent (Twist, 1998).
Every actors perception is modelled in terms of factual, causal and teleological assumptions.
The factual assumptions represent how an actor perceives the current state of his environment.
The causal assumptions represent which changes will occur in the perception of the actor. A
causal scheme is a possible representation, but the language supports these relations usmg ifthen commands. The last type of assumptions are the teleological assumptions, which
represents the actors view on his desirability. Both the causal assumptions and the teleological
assumptions are interpretations, respectively the relations and weights of importance, of the
actor on factors. Due to the formal description of these perceptions in a database (See also
Section 2.7.2), several queries can be executed. For instance, a question like which actors have
conflicting goals on a specific factor or use different definitions for a factor, could be brought
to our special attention in the design of a logistic concept.
Performance measurement
The perception-based factors are the normative values of the performance indicators and form
the connection to the design of the logistics concepts. The connection is comprehended in the
definition of a performance indicator. Performance indicators are quantifiable factors which
enables a company to measure performances in relation to a pre-defined norm (Goor, 1993).
In our approach the pre-defined norm forms the normative element which is a reflection of an
actors perspective on a factor. The value part of a performance indicator can be derived from
the logistics modelling. Finally, the measurement of the value part against the normative part
forms the definition of the performance indicator.
Logistics modelling
The last part of our approach is directed towards the calculation of the value parts of the
performance indicators. Based on the appointed factors of importance in the network analysis,
interpreted from the analysts perspective, dedicated models are developed to measure the
impact of various logistic concept variants. The great challenge is to develop a model at such a
level that is, on the one hand comprehensive and easy to understand for the actors involved and,
on the other hand, sufficiently detailed to validate in practice.

(Cigi Logistics with ITS) 49

CITYLOGISTICS
WITHITS
3.1 FUNDAMENTAL
CONCEPTS
Intelligent Transport Systems (ITS) involve the application of advanced technologies to help
reduce the costs of transport systems. ITS makes, skilful use of advanced electronic and
communications technologies to merge people, vehicles and roads into integrated, intelligent
systems (Nissan, 1998).
Using advances in electronic and communication technology, ITS has the potential to:
(a) Reduce freight distribution costs
(i) increase productivity of local delivery vehicles
(ii) increase reliability of commercial vehicle operations
(iii) increase safety
(b) Increase the capacity of urban freight systems (without providing additional traffic
infrastructure).

ITS have two key elements, intelligence and integration. Intelligence involves gaining
knowledge through data collection and information processing. Integration relates to
connecting and co-ordinating the key elements of the system.
Gains in efficiency through reduced delay and congestion costs can be achieved by developing
integrated information systems. This will also lead to reduce environmental and social impacts
associated with the movement of freight within urban areas.

ITS provides a wide range of opportunities for developing effective City Logistics schemes.
With limited funds and space available for new roads in many cities, there is a real need for
road more sophisticated procedures for efficiently using existing transport systems. With the

50 (CigLogistics)
growing use of "Just-in-Time" transportation there are increased requirements on carriers to
increase performance as well as flexibility (OECD, 1992).
City Logistics based ITS also rely heavily on horizontal data integration that involves the
exchange of data and information between organisations. This requires a high degree of coordination and commitment. Computer systems must be compatible across organisations.
ITS generally consist of the following 3 elements:
(a) Data acquisition
(b) Data processing
(c) Information dissemination
This chapter discusses how advanced technology can be developed and used to perform these
tasks to create successful ITS for City Logistics.

3.2 DATAACQUISITION
An important component of ITS is the collection of data that provides input for management

and operating procedures. Typically a wide variety of data is required for a City Logistics ITS.

3.2.1 Types of Data

ITS usually involves integrating various types of data that can be characterised by its
variability, certainty and source:
(a) Variability
(i) Static (constant)
(ii) Dynamic (changes frequently)
(b) Certainty
(i) Deterministic (certain)
(ii) Probabilistic (random)
(iii) Fuzzy (ambiguous)
(c) Source
(i) Public sector (Road or land authority)
(ii) Private sector (Carriers or customers)

(CigLogistics with ITS) 5 1

Variahilig

i System Speclficatioiz

1 Certainty

1
~

Transport network geometry &

regulations
Land use patterns

Static

Static

Deterministic

Travel times - historical

Deterministic

Probabilistic

Incidents (incl. crashes,

weather & special events)

Table 3.2 Data for distribution management

Variability
Certainty
Fleet
Vehicles (incl. capacity)
Vehicle location

Static
Dynamic

Deterministic
DeterministiclFuzzy

Customers
Location - historical
- real time
Time windows - historical
- real time
Demands - historical
-real time

Static
Dynamic
Static
Dynamic
Static
Dynamic

Probabilistic
Deterministic
Deterministic
Deterministic
Deterministic
Deterministic

Data relating to the transport and land use systems is usually collected by the public sector
(road or planning authorities).
Data relating to the nature of deliveries is usually collected and maintained by the carriers
(Table 3.2).
Data regarding the location of vehicles can be either deterministic (if at depot or customer or
using GPS) or fuzzy (in transit). Customer data that is received before trucks are dispatched
are static and deterministic in nature. However, if these are updated in real time as this
information is received they are considered dynamic.

52 ( C i q Logistics)
Roadway regulation information for freight vehicles involves weight, height, parking, time
and turn bans.
Dynamic or real-time information involves data being continuously collected using dedicated
detection technology:
(a) Traffic (e.g. travel times, volumes and incidents)
(b) Weather
(c) Vehicle location (e.g. GPS)
Static information includes the road network elements (e.g. speed limits).

3.2.2 Requirements and Responsibilities

Establishing and maintaining a data collection program for an ITS based City Logistics
scheme involves defining the requirements and responsibilities for those involved. Several
issues need to be addressed:
(a) Recording
(i) frequency of updating
(ii) storage
(iii) retrieval of data
(b) Ownership
(i) Pricing
(ii) quality control (integrity) of data
(c) Privacy
(i) confidentiality
(ii) security (especially vehicle location information)

3.2.3 Technology
Currently, mobile telephones and CB radio are generally used to communicate between
drivers and control centres or customers. However, there is a wide range of advanced
information technology now available that can automate data collection and communication.
Various technologies now exist for the automatic collection of truck data. Sophisticated
measuring devices can measure a range of attributes of trucks, including length, weight
(Bergan et al, 1995; Karuo and Koyasu, 1995) and speed. These recordings can be linked to
monitoring centres and used for planning, enforcement or fleet management purposes.

(Ci? Logistics with ITS) 53

3.2.4 Travel Time
ITS have already been developed to automatically collect vehicle travel times (Kurosaki er al,
1993). Technology incorporating vehicle license plate recognition using image processing
techniques have also been developed to collect and predict vehicle travel times in real time
(Takahashi et al, 1996).
Multiple electro-magnetic loop sensors imbedded within the road pavement can be used to
automatically classify trucks into standard configurations based on length and number of axles
(Vincent, 1986).
In Japan radar vehicle detectors are used to discriminate vehicle types and CCD cameras used
to capture license plates and these are matched using real time image processing to calculate
travel times (Traffic Bureau. National Police Agency, 1998). Infrared vehicle detectors
(optical beacons) are also used to detect vehicles using near infrared-light to perform two-way
optical communications with vehicles.

3.2.5 Automatic Vehicle Location (AVL)

Real time vehicle location or position data allows the dynamic assignment of jobs (See
Section 8.2) as well as customers to be advised on the current location of their goods and
updated arrive time. There are several methods that can be used to automatically collect travel
time information for trucks, including using specialised equipment within vehicles, such as
Global Positioning Systems (GPS) or electronic tags.
GPS is a object tracking system developed by the US Department of Defence. It is satellite
based and available world wide, at all times and at low cost. It widely used in many areas of
transportation, including aviation and shipping. GPS allow the dynamic location of a vehicle
to be determined using satellite technology. The co-ordinates of the location of a vehicle are
provided in real time. Although it is still quite expensive, GPS is now the leading technique
for automatic vehicle location. Differential GPS can achieve a position accuracy of less than
10 metres, which is sufficient for VRS purposes. To compensate the unstable performance
within urban areas, gyroscope based dead reckoning are often utilised in conjunction with
GPS. GIS provides locational coordinate data. This can be displayed on digital maps (e.g.
GIS).
Other alternative methods of AVL include roadside beacons, specialised mobile radio towers
and geo-synchronous satellite networks for both communication and tracking purpose.
However, the position accuracy of these techniques are inferior to GPS. Individual companies
can purchase equipment that allows their fleet to be monitored in real time.

54 (CivLogistics)

3.2.6 Network Performance

A variety of network performance data can be collected to monitor the performance of urban
traffic networks:
(a) Congestion (e.g. average speed and travel times)
(b) Incidents
(c) Road maintenance
Electronic tags installed on vehicles can be detected by induction loops or other electronic
scanning equipment as trucks pass detectors. Travel times can be determined by comparing
multiple readings of the same vehicle at different locations in the network (Kloot, 1999).
Many cities now have some form of automatic travel time data collection system for
congestion management or performance monitoring purposes.
Melbourne is proposing to collect travel time information on its arterial road network using
electronic vehicle tags and readers located at intersections (Kloot. 1999). Several hundred
thousand motorists have already purchased eTags for use on the privately funded toll road
City Link. Antennas operating at the CEN standard of 5.8 GHz will be attached to traffic
signal gantries at the approach for eventually 200 intersections. Nearby tag readers will record
the signal from antennas and these are sent to a central computer for matching using the
SCATS traffic signal communications system.

3.2.7 On-board
As well as speed and distance advanced in-vehicle technology allows the dynamic monitoring
of a range of other operational parameters relating to vehicle performance eg. sudden braking
incidents and load information. This can facilitate real time rescheduling of vehicles as well as

enhancing safety performance (Section 8.2). Electronic vehicle log-books can automatically
monitor vehicle performance as well as incorporating drivers notes (Laakso and Rauhamaki,
1999).

3.3 DATAPROCESSING
Data processing is an essential component of a City Logistics ITS. Data must be converted
into information. Information becomes information only when they are communicated in
forms and at times suitable for use in a particular decision, (Bowyer and Taylor, 1985). Thus.
information needs to be provided to decision makers at the appropriate time and in an
understandable form.

( C i h Logistics with ITS) 5 5

Information is transformed data, that is:
(a) Processed
(b) Structured (organised)
(c) Accessible
(d) Understandable
(e) Comparable
(f) Useable
(g) Relevant
(h) Communicated to inform
Data processing generally includes one or more of the following procedures:
(a) Verification
(b) Summarisation
(c) Integration
(d) Prediction
Often specialised procedures need to be developed to perform these tasks. Their nature will
depend on the specific functions of the system.
Checking the data for any obvious errors (including missing or invalid data) is an important
component of data verification. Editing procedures may need to be developed. Acceptable error
rates and data accuracy levels should be specified.
Summarisation involves aggregating data, usually over a time or spatial dimension. Integration
involves combining data from a variety of sources. Relational structures need to be defined.
Prediction requires mathematical models to be developed.
The data processing procedures necessary for an ITS for City Logistics also requires a system
for managing the information to be developed. Databases need to be designed and constructed.
This involves specifying file structures (e.g. records and attributes) and human interface
procedures, including the layout of forms.
Procedures for ensuring that there is adequate physical storage for data as well as backups and
security need to be created. Relational structures allow databases to be integrated.
Models are required to be developed for predicting performance and future conditions:
(a) Operating costs

56 (City Logistics)
(b) Level of service
(c) Travel times
This information is typically used in route guidance systems.

3.4 INFORMATION
DISSEMINATION
The dissemination of information to facilitate efficient movement of goods in urban areas is
an important component of ITS based City Logistics schemes. This involves understanding
the decision-making environment of the users of the information. Consideration of the
information needs of potential users will influence the form and timing of its dissemination.
Appropriate use of existing and available technology should be considered. There are a variety
of ways of disseminating information to decision makers. A number of issues need to be
considered:
(a) Who are decision makers (or users of information)?
(i) drivers
(ii) dispatchers
(iii) fleet managers
(b) Why is the information required (application)?
(i) fleet management
- vehicle routing & scheduling
- vehicle tracking
- driver communication
(ii) vehicle management
- route guidance
- vehicle operation
- emergency information
(c) How
(i)
(ii)
(iii)
(iv)

can the information be presented (i.e. what medium)?:

Internet
phone
fax
truck stops

(v) TV
(vi) radio

(CiyLogistics with ITS) 57

3.4.1 Technology
Two-way communication is often needed to satisfy users information needs. Due to recent
technological advances, there are a number of methods available for transmitting information
to commercial vehicle operators, e.g. commercial vehicle radio broadcasting, mobile
telephone, ISDN or Internet.
Radio frequency (RF) communication, including trunk radio, civil band (CB) radio and
mobile telephone, are the main means of two-way information transmission between vehicle
fleets and depots. For information exchange between other sectors, such as those with road
authority or customers, a variety of facilities, such as standard telephone, teletext, leased lines
or Internet, can be employed.
The ability of decision makers to receive and understand information will often be constrained
and will vary according to the user type and the location where it can be disseminated:
(a) Users
(ij dispatchers
(ii) drivers
(iii) managers
(b) Location
(i) control centre
(ii) roadside (e.g. VMSj
(iii) in-vehicle (e.g. VDU)

3.4.2 In-vehicle computer

In-vehicle computer systems have a number of benefits for commercial vehicle operators and
have good potential for being an effective tool for City Logistics (Hellaker, 1996; Suzuki,
1999).
They can automate many of the administrative tasks undertaken by drivers, leading to
increased efficiency. In-vehicle computer systems can calculate fees, monitor performance
and generate reports. Processing of documentation and invoicing can also be undertaken.
Communication with dispatch centres can process customer requests in real time permitting
rescheduling of customers. The vehicles location, amount of cargo, progress can be sent to the
control centre using mobile telephones.

58 (CivLogistics)
Navigation systems can be used to predict arrival times. This can increase load factors and
reduce the amount of travel as well as the number of trucks required.
There are also a number of safety benefits associated with in-vehicle computer systems. The
computer can receive business information without the driver having to use the telephone.
Warnings can be given to drivers for dangerous driving, speeding, sudden braking or rapid
acceleration. Monitoring these events can also assist with driver education.

In vehicle computers can automatically log information relating to vehicle and driver
performance, including, monitoring operational efficiency fuel consumption, vehicle speeds,
drivers work time, feedback into maintenance and scheduling systems. They can aid the
efficiency of vehicle control and goods by monitoring waiting times at customers and depot
for loadingfunloading goods.
In-vehicle units capable of recognising the position of vehicles have the potential to facilitate
road pricing. Tests were conducted on a zone entry and exit road pricing systems in Hong
Kong (Kakihara, 1999).
Navigation is currently based mainly on street maps and the drivers local knowledge
Information on real time traffic conditions is largely obtained from traffic bulletins on radio,
CD radio or mobile telephone. However, in-vehicle navigation systems have a number of
benefits, including savings in time and costs, reduction in driver workload, more reliable
scheduling and simpler co-ordination. A nationwide survey of 4,000 fleet operators in
Germany identified requirements for truck navigation systems (Just and Krug, 1999). Real
time information on traffic and weather conditions was rated as high to very high. Market
penetration was considered to be related to the cost of the units, reliability, sensitivity to realtime information and ability of drivers to understand the advice given.
Communication between vehicles. customers and the control centre can take many forms:
(a) From vehicles
(i) location data
(ii) vehicle status (load factor)
(iii) status of cargo
(b) To vehicles
(i) orders
(ii) announcements (e.g. emergencies)
(iii) traffic and weather information

(City Logistics with ITS) 59

(c) To customers
(i) location of goods
(ii) expected arrival time
Route guidance providing driving instruction using in-vehicle computers can be based on a
number of criteria:
(a) Distance
(b) Travel times -real time (including incidents and road maintenance)
(c) Restrictions -links & turning movements (height, weight and time)
(d) Tolls
(e) Number of turns
Route guidance systems have a number of benefits including:
(a) Increased fleet efficiency
(b) Reduced congestion
(c) Increased reliability (especially intermodal movements)
For example, the Fleetonline software developed for Arizona uses the Trafficcast forecasting
system, updated for urban areas every 30 minutes (Li et al, 1999).

3.4.3 Digital Road Maps

A digitised map of the city of Bremen, including truck route guidance network, truck
information (e.g. regulations) was distributed to truck companies for free (Just and Krug,
1999). This led to an increased number of trucks using truck routes as well as a decrease in
traffic using minor roads.
3.4.4 Internet
The Internet provides a powerful means of disseminating information for ITS based City
Logistics schemes. It allows the exchange of data across a variety of different types of
computers over wide areas. A number of Internet based systems have been developed for
improving the distribution of goods in urban areas (Wild and Moller, 1997; Wild and Moller,
1998). For example, an information kiosk in Aspropyros. Greece provides trip planning and
accommodation information.
The Internet has good potential to facilitate business-to-business (B2B) trading and
information sharing. Where electronic data interchange (EDI) was only available to a limited

60 (City Logistics)
number of companies that had specialised compatible technology, the Internet allows easy
access and communication between organisations. It provides a single standard
communication protocol for B2B information sharing. There are numerous ways the Internet
can be used as an integral part of ITS City Logistics schemes, including:
(a) Real time inventory and order monitoring across the supply chain
(b) Interaction with customers and trading partners
(c) Freight tracking
(d) Bidding systems

3.5 GEOGRAPHIC
INFORMATION
SYSTEMS
Geographic Information Systems (GIS) are computer-based procedures for storing,
manipulating and mapping spatial data. Attributes of spatial objects (e.g. streets) can be stored
in a common database. Various tools allow analysis to be undertaken of spatial networks. GIS
provide a framework for integrating traffic network and performance data allowing a realistic
representation of the traffic network to be constructed.
Existing vehicle routing and scheduling procedures do not allow a realistic representation of
the traffic network to be developed (Eibl, 1996). Road network information (e.g. turn
restrictions, clearways, parking, speed limits, designated routes, number of lanes and gradients)
can be managed within a GIS and integrated with models.
GIS provide an integrated spatial referencing system that has the potential to perform many of
the tasks required to successfully operate a Computerised Vehicle Routing and Scheduling
(CVRS) system (Calogero, 1994; Keenan, 1998). GIS can be used to efficiently store and
manipulate road attribute data (geometry and travel times) necessary for estimating minimum
cost routes between customers in urban areas. The mapping, analytical and database
management capabilities of GIS software also has the potential to aid CVRS systems. For
example, address matching (geocoding) allows the automatic identification of the location (of
depot, customer, etc.) on a computerised map. given its address to be performed. Routes can be
generated based on the connectivity of roads, legality and advisability of using certain roads.
There are a number of general ways that GIS can be integrated with vehicle routing and
scheduling systems. Firstly, GIS can be used as an interface for the routing and schedule
procedures. Various data can be imported and exported between the routing and scheduling
procedures and the GIS. In this case, the optimisation modules are run separately from the GIS.
The GIS is only used to manage the network data and display the results. This can be achieved
using a dynamic data exchange (DDE) protocol and establishing an inter-process
communication ( P C ) link between the two packages. Examples of this type of integration are

( C i v Logistics with ITS) 61

the Routesmart Software with ARC/INFO and OPCOMs vehicle routing utilities with
MapInfo.
Routing and Scheduling modules can also be imbedded within the GIS. These can be either
built-in procedures or user-defined modules. Many desktop GIS have a set of routing and
logistics procedures such as clustering, network partitioning and vehicle routing with time
windows. Most GIS (e.g. MapInfo and TransCAD) allow user defined modules developed in
other programming environments to be executed in the same way as built in procedures within
the GIs.
When establishing a GIS for goods distribution analysis the traffic network and its performance
as well the distribution network must be specified. A representation of the physical
characteristics of the main roads needs to be provided. Other physical geometric information
may also be incorporated. Travel times for links can also be included or estimated from
distances. The location of depots and delivery/pickup points must also be defined.
Characteristics including the amount of goods and specified arrival times can also be stored as
attributes.
GIS libraries can also be accessed from stand-alone programming environments allowing
specialised optimisation procedures to be integrated. Specific programs can be developed for
incorporating distribution networks into the GIs for display and modification, reverting the
network format to an appropriate structure for the optimisation models and returning the
optimal routes and schedules to the GIS environment for visualisation and analysis of the
results. An example of this type of system was developed for Melbourne using the DELPHI
programming environment and Mapobject GIS toolkit (Thompson et al, 1999). Details of the
location, demand for goods and time windows of assembly plants were managed within the
integrated software environment.
Recent applications of GIS in urban freight have been in planning hazardous routes (Kim,
1998; Kim et al, 1998) and real time fleet management (Wang, 1998; Wang et al, 1999).

3.6 EFFECTSOF E-COMMERCE

The recent growth in e-commerce and tele-shopping is a trend that will continue to influence
retailing and goods distribution patterns in urban areas. Home delivery of goods will become
more prevalent as customers demand that goods be delivered to their homes at specified times.
This will lead to smaller loads being distributed to more dispersed locations. Retail stores will
be bypassed as local couriers deliver goods directly to homes. Distributors instead of
consumers complete the final phase of the distribution chain.

62 (City Logistics)

Currently e-commerce is still limited to certain industries such as books, CD's, computer
hardware and software. However, it is predicted that Internet sales will grow in the food and
clothing sectors as well.
New supply chain management and routing and scheduling procedures will need to be
developed to accommodate new (direct-to-customer) distribution systems. This will require a
major shift in focus for logistics models, since most currently ignore the transport of goods
after reaching the store.
The effects of direct delivery systems on the social and environmental amenity of local areas
needs to be considered. Co-operative distribution systems offer the potential to reduce the
number vehicles entering local areas.
Distribution costs from warehouse to retail outlets as well as operation costs of retail stores
(rent, cashiers, stackers, etc.) form the major proportion of costs in traditional retailing in
urban areas. However, with Internet trading, distribution costs will increase as goods are
delivered from warehouses directly to home. Costs associated with picking individual orders
will become significant.
Internet sales increase the amount of travel by sellers, but decrease in amount of travel by
customers. There is also a change in the mix and types of vehicles used to distribute goods to
homes. This will led to more small trucks and vans using residential streets, producing noise,
amenity and safety problems.
E-commerce presents a new set of challenges in supply chain management. Large numbers of
small orders must be processed quickly, accurately at a low cost. There is a need to develop
new distribution systems using City Logistics to cater for the rapid fulfilment of customers
orders generated by internet sales.

3.7 CURRENTITS APPLICATIONS

A description of current ITS applications in automatic tolling and travel time information is
described in this section.

3.7.1 Automatic Tolling

Automatic tolling eliminates the need for vehicles to stop to pay tolls, avoiding delays at
tollgates. Several cities have already implemented automatic tolling technology. Automatic

(City Logistics with ITS) 63

debiting systems enable drivers to automatically pay tolls without cash and stopping at
tollbooths. Vehicles with electronic devices (cards or transponders) are able to be identified by
readers or detectors. Vehicle information is read and transmitted to a control centre for
processing. Charges are made based on the vehicles type and sections of the network used.
Sometimes variable charges are made depending on the time of day.
Japan is committed to installing electronic toll collection (ETC) at 60% of its tollgates by
2003. It is estimated that tollgate congestion accounts for 35% of all expressway congestion
(Nissan, 1998). ETC will result in considerable time and fuel savings. Given the high volume
of freight traffic travelling on expressways in Japan, there are substantial economic benefits
associated with ETC. Conventional toll collection takes an average of 14 seconds per car in
Japan (HIDO, 1996). Significant reductions are expected in personnel, maintenance and fuel
costs relating to the introduction of ETC in Japan.
ETC in Japan planned to be integrated (nationwide) but vehicles must still enter the toll gate at
a low speed around 30 kmih (Yamamura, 1998). Vehicles will be charged using roadside
radio communication technology (Fujimori, 1999). Tests have shown no safety problems and
a substantial increase in capacity.
Transurbans City Link project in Melbourne is based on electronic tolling technology. Since
opening in 1999, the Western Link toll facility has considerably reduced travel times for
freight vehicles. Motorists pay tolls while travelling at freeway speeds using an electronic tag
eTag that is mounted inside the windscreen. Vehicles are not required to stop at tollbooths
or boom gates. As tagged vehicles travel through the toll zones, an electronic transaction
occurs as the eTag is detected and a fee is deducted from the users pre-paid account. A
computerised visual recognition system is used to ensure that the vehicle class matches the
eTag type.
Overhead gantries have been erected at each toll point and when vehicles pass beneath them,
detectors communicate with the tags by microwave and automatically detect the appropriate
toll from the toll account. If a vehicle does not have a tag, the licence plate is recorded using
video. This is checked against the day pass list and if a day pass was not purchased the licence
plate is sent to the Government Enforcement Agency and an infringement notice is issued.
Separate rates are used for light and heavy commercial vehicles based on weight. Lower rates
are charged for light commercial vehicles weighing between 1.5 and 4.5 tonnes gross vehicle
mass (GMV). B-double trucks up to 65 tonnes (GVM) are able to use this facility.

64 (City Logistics)
3.7.2 Travel Time Information
Most major cities now have implemented ITS that provide some form of travel time
information. These allow users to avoid links with roadworks, incident sites and congested
areas. Some systems allow travel times to be predicted between specific locations based on
current and predicted levels of demand.

3.8 EVALUATION
ISSUES
There are a number of challenges relating to how to plan, design and successfully implement
ITS for City Logistics. These should be addressed in order to create effective ITS. Issues
relating to the usability, financial viability, political acceptability, competitive neutrality and
horizontal compatibility are described below.

3.8.1 System Usability

Freight carriers often are small business with little computerised technology. Many small
companies do not have access to advanced information technology (Laakso and Rauhamaki,
1999). A lack of resources and knowledge as well as inadequate educational background can
inhibit the utilisation of ITS. Audio and graphics based information systems offer good
potential to help simplify the interface with new technology.

3.8.2 Financial Viability

ITS initiatives need to be financially viable for both the providers of the technological
infrastructure as well as the users. There should be a reasonable return on investment to justify
the acquisition costs. Often a reduction in operating costs by shippers and carriers can be
realised. Demand and cost models need to be developed to forecast the financial impacts of ITS.

3.8.3 Political Acceptability

New initiatives relating to the application of advanced technology often requires support from
administrators. With ITS based City Logistics schemes the following values often need to be
addressed:
(a) Safety
(b) Privacy
(c) Freedom
(d) Confidentiality
Initiatives should not threaten but enhance these values.

(Cirq' Logistics with ITS) 65

3.8.4 Competitive Neutrality

Care must be paid to ensure that equal access is maintained to markets with the provision of
ITS City Logistics services. Free market principles ought to be upheld with administrators
creating transparent processes when dealing with the private sector.

3.8.5 Horizontal Compatibility

The integration of information systems between organisations requires trust and commitment.
Exchanging and sharing information between companies can be a sensitive issue. However,
co-operation is vital for the successful implementation of ITS based City Logistics schemes.

(Demand and Supply Models) 67

DEMAND
AND SUPPLY
MODELS
4.1 OVERVIEW
The movement of goods comprise one of the major sources of urban traffic problems, such as
traffic congestion, negative environmental impacts, energy consumption and so on. Social
impacts of goods movement and freight transport should therefore be quantitatively assessed,
and this can be undertaken using demand, supply and impact models. However, much less
research has been undertaken on modelling the demand andor supply of urban goods
movement than that of passenger movement. This is partly because traffic congestion has been
often highlighted in urban planning, and passenger movements have been considered to play
more important role than goods movements. More plausible reasons for this may include the
fact that it is more difficult to model the demand of goods movement.
There are many factors effecting the demand for freight that makes it difficult to model. The
urban freight systems are highly complex due to decisions being made by a number of persons,
including individual vehicle owners, fleet owners and a network of sub-contracting
arrangements by forwarders (Meyburg, 1979; Yeomans and Balce, 1992). In addition, there is
enormous diversity in the range of predictions that can be made within the freight system
(Harker, 1985; Winston, 1983). Central to all models is the concept of forecasting the
behaviour of agents involved in the movement of goods. The economic agents involved in the
movement of freight are defined as producers, consumers, shippers, carriers, potential carriers
and government. The movement of goods is assumed to be largely influenced by the decisions
and interactions between these agents.
The modelling of goods movement in urban areas has relied heavily on approaches developed
in the area of passenger demand modelling (Meyburg and Stopher, 1974; Starkie, 1971).
Furthermore, the vast majority of freight demand models developed have followed the
aggregated four-stage model, which is traditional approach for representing passenger demand.
This approach involves four estimation processes: generation and attraction, distribution, mode
split and assignment.

68 ( C i v Logistics)

In the case of inter-city freight transport, prediction has been made by applying other
approaches: e.g. econometric, spatial price equilibrium and network equilibrium approaches.
Harker (1985) reviews the major methodologies involving these approaches.
Econometric modelling has been mainly applied to investigate the shipper-carrier-government
relationship. Estimates of industry wide or firm specific cost or production functions have
typically been determined. Spatial price equilibrium modelling incorporates a network
approach and looks at the interactions between producers, consumers and shippers. Nodes in
the network are designated as either producing or consuming regions for commodities or both.
Links connect these regions either directly or with transhipment nodes. Demands and supplies
are estimated within each region. Equilibrium flows are determined based on the transport cost
and price of commodities. Takayama and Labys (1986) review the ways to formulate spatial
price equilibrium models. Freight network equilibrium models utilise a network structure to
represent the movement of goods and mainly focus on the interactions of shippers, carriers and
potential carriers.
There are two basic categories of demand models, those that focus on commodity flows and
those that focus on truck trips. Both types of models have vastly different data requirements.
Rimmer and Black (1981) present a conceptual framework for investigating urban goods
movement. They define the urban goods process as a logistical chain between producer and the
ultimate consumer. Emphasis is given to the linkages between transfer terminals,
manufacturers, wholesalers, retailers and householder or institution as final consumer. The
links between economic units (land use activities) create the demand for urban goods
movement. The role and perceived problems of a number of different actors including
pedestrians, car drivers, truck drivers, truck owners, shippersheceivers, residents, highway
authorities, urban planners and government are outlined.
This chapter deals with conventional approaches representing the demand and supply side of
urban goods movement: generation, distribution, mode split, simulation, travel times and
generalised costs. The following sections present a detailed description of demand and supply
models relating to urban goods movement and urban freight transport.

4.2 GENERATION
Freight generation models can be generally classified by their spatial unit (land use or zone)
and flow unit (goods or trucks). Regression models have been widely used to develop
relationships that quantify the influence of variables relating to the intensity of production
activity. In regression models, the variable to be forecasted is the dependent variable and the
independent variables are causal variables. A classification of regression models by their

(Demand and Supply Models) 69

functional forms representing the functional relationship between the dependent variable and
the independent variables is given below:
(a) Simple linear
There is only one independent variable (i.e. a simple regression), and the functional
relationship is assumed to be linear.
(b) Multiple linear
The number of independent variables is greater than one (i.e. a multiple regression), and
the functional relationship is assumed to be linear.
(c) Non-linear
The functional relationship is assumed to be non-linear (e.g. an exponential function).

In the case of multiple linear regression models, their general functional form can be given as
below:

y=a,+a,x, +a,x,+...+anxn+E

(4.1)

where,
y : dependent variable
a2 : regression

coefficient

x!

: independent variables (or explanatory variables)

: disturbance term

In the case of freight generation models, the dependent variable represents the freight
generation. Since dependent variables cannot be completely explained by the independent
variables used, disturbance term is available for representing such an error.
Regression coefficients represent the relationship between the dependent and independent
variables. The values of these coefficients can be obtained from a dataset involving the value of
dependent variables and independent variables experienced. A measure of correlation between
predicted value of dependent variables with the resulting regression coefficients and their
actual value is given by a multiple correlation coefficient. The accuracy of prediction is better
as the value of this coefficient is closer to 1.
The demand for freight is however, often affected by factors external from the site, area or
region being studied. These may include economic, technological and government policies.
Regression models that relate freight generation to land use activities cannot account for these
external factors. Hence, the intensity of activity represented by land use variables is unable to
account for the affects of external factors. Regression models are only able to represent urban

70 (CiQ Logistics)
goods generation patterns at a certain point in time. Although it is well accepted that regression
models are too simplistic to represent the complex market segments for freight services,
commodity flows organisation and operation of the road haulage, they can play an important
role in the long-term strategic planning process.
Meyburg and Stopher (1974) suggest that freight land use classification should be based on
homogeneity with common freight generation activities and generation of freight vehicle
movements. They also recommend that models classify goods by type and include a measure
of intensity of use (i.e. floor area or employment), account of the freight handling capabilities
and restrictions and incorporate a measure of accessibility. At the site or activity level, Maltby
(1970) presented a number of relationships for light and commercial vehicle attractions and
generations for manufacturing plants based on their employment and floor area.
Ogden (1991) presents some typical generation rates based on floor area for a number of types
of developments for courier vans, light rigid, heavy rigid and articulated trucks. It was
acknowledged that there was wide variation from site to site. Ogden ef al. (1981) describe
some general generation characteristics of 73 companies from an interview survey conducted
in Melbourne. Although no analytical relationships were determined a number of broad
generalisations for several industry groups were made. This study showed that truck generation
patterns are highly variable between companies and depend upon the needs of clients and the
specific markets they operate in.
Rimmer and Black (1981) developed regression models for zonal truck demands for both light
and heavy vehicles based on area, employment types, number of households, employment
types and land use types. Similar models for heavy vehicle productions and attractions were
presented by Watson (1975) and Freier (1981).
Ogden ( I 977) presented a number of simple regression based models from a large commercial
vehicle survey undertaken in Melbourne. Using zonal data, relationships of truck trips
generated and attracted by trip purpose were estimated for eight trip destination purpose groups.
Regression equations for the amount of goods generated by eight commodity types were also
produced. A number of socio-economic explanatory variables for each zone, representing
employment, population, number of households and resident labour force were tested.
Significant variables were identified and coefficients estimated. Separate models for seven trip
purposes including retail, wholesale and industrial delivery were developed.
Other approaches apart from regression models have been utilised for estimating freight
generation and attraction. Macroeconomic models, for example those incorporating
input-output linkages, have been used for representing inter-regional freight movement.
Growth-factor methods are often used for forecasting freight trip ends, as used in the case of

(Demand and Supply Models) 71

forecasting future passenger trip ends

4.3 DISTRIBUTION
The gravity model has been the traditional approach for modelling the distribution patterns of
urban goods movement. The production constrained gravity model has been widely used to
estimate the spatial patterns of goods being transported (Rimmer and Black, 1981). A
generalised cost function has been used in a number of studies, including Columbia (Kresge
and Roberts, 1971). In the case of urban goods movement, the gravity model can be formulated
as follows:

where,
k : commodity type index
45 : commodity flow (or truck trips) from zone i to j for commodity type k

p:

: balancing (calibration) factors

0: : supply and demand for commodity k at zone i (or j ) , respectively

generalised transport costs for commodity k between zone i t o j

parameter
listribution of urban truck trips and commodity flows in Melbourne were analysed using a
gravity model (Ogden, 1978). Models were developed using commercial survey data.
Commodity flows represented by weight were represented as being attracted to destinations not
produced from origins using a singly constrained gravity model. Airline distance was used to
represent impedance for commodity flows while travel time was used to represent impedance
for truck trips. Models were calibrated for various trip purposes and commodity types. This
involved estimating the negative exponent of the impedance function. Generally higher
exponent values were estimated than typically found to represent personal travel since trucks
tend to travel more within local areas with long distance travel in urban areas generally being
uncommon. The highest values were for retail trips and wholesale delivery. Manufactured
products were lowest. Reasons for this included that transport costs are a lower component of
the total costs for higher value products and other service factors may influence their
distribution. Thus spatial separation would not be such an important factor in influencing their
origin and destination patterns. Waste products had the highest exponent value since low value
or zero-value products would have significantly shorter trip lengths to minimise the transport
and collection costs.

72 ( C i c Logistics)
Ashtakala and Murthy (1988) developed a set of gravity models for various types of
commodity flows between cities and towns in Alberta. Production constrained gravity models
were developed using an impedance function based on distance and an exponent parameter.
Regression analysis was used to test agreement between the OD flows predicted by the gravity
model and the OD flows estimated from the survey. Optimal values of the impedance
parameter were estimated for a number of different commodity types.
The Bureau of Public Roads model was used for estimating light and heavy commercial
vehicle inter-zonal traffic (Rimmer and Black, 1981). Park and Smith (1997) describe a
procedure for incorporating truck classification count data into the estimation of origin and
destination patterns in Wisconsin.
Mathematical Programming (e.g. linear programming) is widely used by the freight industry in
minimising the total direct transportation costs (in money terms, or sometimes in terms of
generalised costs), subject to supply and demand constraints within a city or region. Here,
companies attempt to minimise their distribution costs while satisfying the demands of their
customers. Multi-mode and multi-product assignment models have been developed at a
regional and national levels (Drissi-Kaitouni, 1990).

In general, the formulation of distribution modelling based on linear programming (LP) is as

follows:
Minimise

(4.3)

subject to

c 4 i j = Dj

(4.4)

Cqij= oi

(4.5)

This problem with a linear cost function is called the transportation problem. This type of
problem can be solved easily with short computational times. Although, LP is a good model for
freight planning for a specific carrier or f i i , for urban freight planning, it is not so useful
because it is often difficult to simultaneously represent the flow of various commodities. The
gravity model has therefore been used for predicting future distribution patterns of urban goods
movement due to its flexibility: the value of y can represent the relative importance of cost
under the supply and demand constraints.

(Demand and Supply Models) 73

4.4 MODESPLIT
Mod2 selection procedure is necessary when alternative modes are available for freight
transport between any two points. The best mode will be selected by shippers to transport their
goods considering various attributes of the modes available:
(a)
(bj
(cj
(d)
(ej
(g)
(h)
(i)
(i)

Accessibility
Availability (infrastructure, service, equipment)
Capacity
Customer service
Cost, freight rate, price
Distance
Flexibility (scheduling)
Frequency
Inventory
Loss and Damage

(k)
(1)
(mj
(n)

Reliability (transit time, transhipment time)

Safety
Security
Speed (transit time, terminal time, transhipment time)

(f)

Mode selection also depends on the characteristics of goods to be transported. A variety of

characteristics of goods will be taken into account when selecting the best modes:
(a) Density
(b) Shipment size
(c) Time window
( 4 Type
(ej Value
(0 Volume
(gj Weight
Almost all intra urban goods movement is by road since road transport generally provides a
higher level of flexibility and financial viability than rail. This is important as urban freight
movement tends to be over short distances with short lead times. Rail will also almost always
require a road leg since rail terminals are limited. The additional time to load and unload
goods can be significant. Trains are restricted to timetables that are limited in frequency due to
the large nature of haulage generally required for their economic viability.
Meyburg and Stopher (1974) suggest that different types of trucks should be considered as

74 (Ciry Logistics)
modes in the analysis and prediction of freight demand. Here, delivery vans are considered to
be different modes than semi-trailers. Truck operators must consider the characteristics of the
goods, driver qualifications and a number of possible legal and institutional constraints. In
addition, the factors influencing mode choice are largely dependent upon the commodity being
transported (Zlatopper and Austrian, 1989). A number of general factors including reliability,
quality, cost, speed, length of haul, shipment size and freight rate have been identified as being
important (Gray, 1982; Allen er al., 1985).
The majority of previous studies on freight transport mode choice have revealed considerable
differences in the estimated results across commodity groups. Aggregation over such
heterogeneous commodity types can cause bias in the estimation of models as the same model
structure and the same values of unknown parameters are assumed for all observations. This
will result in different consequences of any policy change for different commodity types (Nam,
1997). The logit model with disaggregate data, one of the most widely applied discrete choice
models, has been widely used for passenger mode choice analysis. The superiority of the
Box-Cox logit model has been recently shown over the linear logit model. Few applications of
the Box-Cox logit have however been found in the area of freight transport: shipper's choice
between own-account and third-party commodity transport and inter city freight mode choice
in Canada (Picard and Gaudry, 1998).
Application of disaggregate approach is not limited to mode choice. Freight transport
essentially includes a variety of decision-making: type of commodity and its volume to be
transported, origin and destination. Apart from the shipper's decisions, carriers have to choose
the route travelled for pickup/delivery. The mechanism of these choices can be described using
discrete choice models, such as binary choice models, multinomial choice models and
multidimensional choice models (Ben-Akiva and Lerman, 1985).
It is widely accepted that decision-makers make their choices of mode; shipment size,
frequency and supply market. simultaneously, not in sequence. Recently models for
representing the combined choices of mode and shipment size have been proposed (Abelwahab
and Sargious, 1992). In these models, the interaction between combined decisions is
represented.

4.5 ASSIGNMENT
Route choice is the carrier's decision regarding which route is to be used to pickup or deliver
goods. The choice of route by drivers within an urban network may be constrained by the
vehicles size. Driver knowledge will affect the number of alternatives considered as well as
their perceived costs. There are a number of general factors that tend to influence the route

(Demand arid Supplx Models) 7 5

choice behaviour of truck drivers in urban areas including driver, vehicle and route. Driving
effort both physical and psychological has an important influence in the route choice of truck
drivers (Stem e t a / . , 1983).
General factors affecting the route choice of truck drivers were identified in a study of the
impacts of the opening of the Gateway Bridge (a toll facility) in Brisbane (Bowyer and Ogden.
1988). A survey identified the following factors in order of importance:
(a) Occupant costs (time value and meal money)
(b) Gateway Bridge toll
(c) Quality of service to customers (e.g. satisfying time constraint)
(d) Vehicle operating costs
(e) Driver stress
(0 Industrial harmony
This study indicated that the route choice process of truck drivers is quite complex and
contains many factors that have not generally been incorporated in route choice algorithms.
A survey of 100 freight vehicle drivers in Brisbane revealed that most drivers found less
congestion and the shortest route to be the most important factors affecting route choice
(Bitzios and Ferreira, 1993). The effect of toll roads was more pronounced for owner-drivers
compared with company drivers. Most drivers stated that less congestion, shortest route, less
turns and signals and the fastest route were extremely important factors. Drivers of articulated
vehicles placed a greater emphasis on road width, road vertical alignment, turns, signals and
congestion that was the case with rigid vehicle drivers. The smaller vehicles were more likely
to choose the fastest and shortest routes. Light commercial vehicles placed much more
emphasis on time and distance. Only 8 percent of drivers stated that they did not choose the
route they used. Some 92 percent of drivers believed that there were only a maximum of three
routes available. The majority of drivers (63%) believed that there were only two major routes
available.
A logit model for predicting truck volumes on routes was proposed in a freight modelling study
in Sydney (Yeomans and Bake, 1992). This approach relates the probability of a driver
selecting a route to the relative cost or travel time for using that route compared with other
alternatives. Separate models for different vehicle types were recommended. It was suggested
that separate models could also be developed for different commodity types and weight of
goods as well as the weight of the vehicle.
Tsukaguchi and Vandebona (1999) developed procedural models to represent the dynamic
aspects of the pick-up process in Osaka City. These models incorporate different shipper

76 ( C i v Logistics)
characteristics and operating procedures. They also account for the route choice and site choice
processes used by truck operators in the field.

In general, freight transport systems can be represented by a network of nodes and arcs. Arcs
represent the transport links or routes between nodes, which can be centroids, freight terminals,
depots, ports, airports or intersections. Spatial price equilibrium modelling, for example,
involves a network-based approach for intercity freight transport. Transportation network
modelling is often used in urban transport planning. It is well recognised that transportation
network analysis is an effective approach for investigating traffic flow patterns.
Urban goods movement influences traffic flow patterns, and can cause traffic congestion and
negative environmental impacts in urban areas. Traffic conditions on road networks, in turn,
substantially influence on urban goods transport because almost all intra urban goods
movement depends on road-based vehicles. Therefore, traffic flow models are important for
modelling City Logistics. Traffic conditions on road sections can be predicted using traffic
assignment techniques as part of traffic demand estimation procedure. Traffic assignment
procedures involve two types of models in order to predict link flows, route choice models and
traffic flow models.
Traffic flow models represent the relationship between traffic demand and traffic conditions.
Traffic demand is usually represented by a trip matrix. Traffic congestion is estimated using
properties of links including speed-flow relationships. These models must be able to
incorporate the effects on travel times due to the dynamic and spatial aspects of traffic flow on
the road network. Traffic flow on road networks can be represented using two types of models,
traffic assignment models and traffic simulation models.
Traffic assignment models can be classified by the physical characteristics of links and
stochastic elements in route choice. The former relates to the capacity restraint of links, and the
latter to the difference in individual perceptions of best routes and objectives of travel.
Traffic links are usually assumed to have a capacity restraint. In this case, the user equilibrium
traffic assignment approaches have been adapted to evaluate the traffic flow on road networks;
stochastic user equilibrium approach if stochastic properties are considered in driver's route
choice, deterministic user equilibrium approach, if not. The simplest case is when links have no
capacity restraint and stochastic elements are not taken into account. In this case, the
all-or-nothing assignment technique can be applied to identify the traffic conditions on the road
network. Patriksson (1994) reviews the formulation and solution algorithms for traffic
assignment techniques.
Many City Logistics initiatives require data regarding the traffic conditions on the road

(Demand and Supply Models) 77

network for different times of the day for evaluation purposes. For example, real-time traffic
conditions are necessary for assessing ITS-based City Logistics initiatives since driver's travel
behaviour will dynamically change in response to traffic information (e.g. dynamic route
choice). Dynamic vehicle routing and scheduling systems may not be able to be established
without the data associated with real-time traffic information.
However, dynamic traffic assignment models, presently, can only be applied to a limited range
of applications due to their theoretical properties. Traffic simulation models are therefore more
suitable for predicting and representing the dynamic change in traffic flow.
A number of urban transport models permit an additional penalty for stopping at intersections
as well as travel time to be incorporated within the assignment procedure for trucks. For
example, in EMME/2 a penalty changes the calculation of the least cost route for trucks, but
does not alter the travel times on links for trucks, allowing the model to output values of travel
times for trucks on the preferred routes (INRO, 1996).

4.6 SIMULATION
MODELS
Mathematical optimisation models have been widely used to solve problems in urban freight
transport. They have the advantage of identifying the problem structure because they typically
involved specifying an objective function and a set of constraints.
Mathematical optimisation models, however, essentially involve developing an abstract
representation of a real-world problem, and consequently their solutions can only represent
those in the world defined by the objective functions and constraints, namely those not in the
real world. Simulation models have the potential to better deal with real-world problems.
Simulation provides a powerful tool for modelling traffic and transport systems (Young et al.,
1988). Applications of discrete event simulation in City Logistics have however been limited
due to large scale of urban traffic networks and vast number of vehicles operating within the
system. Simulation is generally used in terminal management and intersection capacity
analysis. In particular, simulation provides a means of investigating the performance of
customer allocation strategies in the dynamic scheduling of vehicles in the courier industry.
Simulation modelling was recently used to predict the benefits of using historical travel time
information with stochastic programming procedures (See Chapter 7).
A simulation model was constructed for the Chicago Area Transportation Study to evaluate
terminal clustering and primary truck route design plans (Southworth et al., 1983). GoodTrip is
a disaggregate model that estimates goods flows and urban freight traffic based on logistical

78 (CivLogistics)
chains (Boerkamps and van Binsbergen, 1999). This model is based on representing choices in
the supply chain. The interactions between the spatial organisation of activities, goods flows.
traffic flows and infrastructure are modelled. GoodTrip was used to compare the performance
of urban distribution centres and underground systems with traditional systems for food retail
and book distribution in Groningen, in the Netherlands.
Traffic simulation models have been used to represent and predict the traffic conditions on the
road network, especially in investigating the dynamic characteristics of traffic flow. Traffic
simulation models are classified into two types, micro-simulation models (e.g. NETSIM,
Mahmassani et al., 1990) and macro-simulation models (e.g. CONTRAM, Leonald et al.,
1978). Micro-simulation models simulate the behaviour of each vehicle and hence require a
large amount of computation time. The macro-simulation models represent traffic flow as a
group of vehicles, thus provide a less realistic representation of actual traffic conditions.
Hybrid traffic simulation models (e.g. BOX, Fujii et aL, 1994), which involve both a
microscopic and macroscopic representation of traffic, have been developed recently.
A hybrid dynamic traffic simulation model has been used to predict the impacts of several City
Logistics schemes (See Section 6.4). This model is based on a modified BOX model where the
origin and destination of each vehicle is defined and trucks are converted to passenger car units
to determine travel times on traffic links. Route choice behaviour within the network is based
on drivers perceptions that are related to the predicted travel times. This model allows the
arrival time of specific vehicles at assigned nodes (customers) to be estimated. Groups of
vehicles travelling on sections of links are represented as a flow rate that is dependent on
density. This model allows a detailed representation of truck movements and their interaction
with passenger cars to be performed with passenger vehicles represented at an aggregate level.
This type of model provides a feasible approach for investigating the impacts of City Logistics
schemes for urban traffic networks. More detailed description of BOX model is presented in
Section 6.4.
Simulation modelling provides a powerful tool for predicting the benefits associated with ITS.
The effects of various types of network performance information (e.g. travel times) on the costs
of distributing goods in urban areas can be estimated (See Sections 3.5 and 7.3).

4.7 TRAVELTIMES
The aim of City Logistics is to develop and effectively implement measures able to reduce the
total social, economic and environmental costs of the distribution of goods in urban areas using
road-based vehicles. Both traffic congestion and environmental impacts are strongly influenced
by travel times (or speed), which depend on traffic flow. Transportation costs as well, are

(Deniarzd arid Supply Models) 79

influenced by the traffic conditions on the links of the road network. The performance of the
road system must be predicted to estimate the demand and impacts of City Logistics schemes.
Models representing the relationship between traffic demand and travel times are therefore
required.

A number of urban transport models allow traffic demand and travel times to be estimated
using traffic assignment procedures. This is usually undertaken by focusing on the flow and
times on specific road links. The best approach is likely to be direct observation, but practical
difficulties may arise in making observations. Therefore, to complement direct observations,
theories of traffic flow are used (Bell and Iida, 1997).
The relationship between traffic flow and link cost is often represented using link cost
functions, which are sometimes called link performance functions. An overview of link cost
functions is presented by Branston (1976), including BPR function and Davidson function.
Link cost relates strongly to driver's route choice and plays an important role in traffic
assignment. It is well accepted that the perception of travel time and other cost elements for
traffic links may differ between drivers. Most existing traffic assignment models however,
consider travel time to be a primary influencing factor of link cost. Therefore, the majority of
link cost functions focus on the relationship between travel time and traffic flow on a link. Link
cost functions can be classified into two types; those that rely on the assumption that traffic
congestion can stop the traffic flow without considering the effects of traffic control devices
(e.g. signals) and those that assume that the link flow is dependent on intersection controls. The
first type of link cost function is represented by a function that has two travel times for any
given traffic flow (Figure 4.I).
Another is that applied if a signalised intersection at the end of a link is assumed to influence
the traffic flow on the link (Figure 4.2). In this case, delay time at a signalised intersection
rapidly increases as traffic volume on the link approaches its capacity of that intersection;
traffic flow finally stops when the intersection reaches its capacity limit. This type of link cost
function is typically represented using Davidson function (Davidson, 1966). The mathematical
formulation of Davidson function is as follows:
fo("o)='oo~~+J"o

where,
: free travel time on link a (minute)
Va : traffic flow on link a (vehicledday)

KO : traffic capacity of link a (vehicledday)

J

: delay parameter

/(KO-"A

(4.6)

80 (City Logistics)

Here, the value of the delay parameter depends on the characteristics of links. Figure 4.3 shows
the shape of Davidson function for several values of the delay parameter.
Basic models of user equilibrium traffic assignment approach can be formulated as non-linear
minimisation problems with a set of convex feasibility constraints. Link cost functions must
strictly be monotonically increasing so that this type of model can be solved uniquely. User
equilibrium traffic assignment models therefore, cannot directly use the above-mentioned link
cost functions: those asymptotically approaching to link capacity and those of higher order.

Traffic flow

K
O

Figure 4.1 Relationships between travel time and traffic flow on a link without stopping at

Traffic flow
K O
Figure 4.2 Relationships between travel time and traffic flow on a link with stopping at
signalised intersection

(Demand and Supp/y Models) 8 1

0
500

1000

Traffic flow
Figure 4.3 Davidson function (roo= 10, K O = 1000)

150

ha

=500

500

1000

Traffic flow
Figure 4.4 BPR function

(roo= 10, a = 2.62 and

p = 0.5)

The BPR function, developed by Bureau of Public Roads (BPR) of the United States (Bureau
of Public Roads, 1964), is a strictly monotonically increasing function and has been widely
used for user equilibrium traffic assignment problems due to its superior operationality. It has
two parameters of a and p (equation 4.7).

These parameters were originally set at a = 0. I5 and

p = 4, respectively. In general, since the

characteristics of roads may be different among countries, the original values of these
parameters will not be always available for all countries. Modified BPR functions (Steenbrink,

82 (Cic Logistics)
1974), developed in the Netherlands, has the same functional form as the original BPR
function, but different parameter values: a = 2.62 and p = 5. Figure 4. shows the shape of
modified BPR function for typical values of traffic capacity.
Intelligent Transport Systems (ITS) enable the direct observation of vehicle travel times.
Dynamic traffic flow models, which automatically collected travel time data, are expected to
be developed for the near future.

4.8 GENERALISED
COSTS
Generalised costs perceived by decision-makers, consist of all elements considered to be
important by decision-makers. In general, there are many types of costs that can be
incorporated, including money, time, distance, or impact (mental, physical). A generalised cost
provides a measure that combines these elements. Generalised cost functions for freight
demand can be formulated as follows (e.g. Kresge and Roberts, 1971):

where,

fzi : out-of-pocket charge of product k for using a service from zone i t o j

a,: : coefficient of term rn for product k
V,f : explanatory variable, such as, travel time between zone i andj, waiting time on customers,

delay penalty on customers, variability of travel time, probability of loss or damage to

goods in transit

In general, the values of these explanatory variables depend on the mode used and to some
extent on the commodity being transported. These coefficients also rely on the value of
commodity. Recently supply chain management systems have tried to minimise these elements
together with stocking costs to realise the higher level of service including urban Just-In-Time
pickup/delivery. Variables influenced by traffic conditions on the road network in the area
being studied, such as, travel time, delay penalty and variability of travel times, are strongly
related to the increase in inventory costs. The values of coefficients for these variables are
therefore proportional to the value of the goods.
Costs incurred in transporting goods are associated mainly with routes, terminals and vehicles.
Total costs of transporting goods are generally composed of transportation costs and facility
costs. The transportation costs involve vehicle operating costs and fixed costs of vehicles. The

(Dentand and Supply Models) 83

vehicle operating costs are the sum of time costs and penalty costs for early and delayed
vehicle arrivals at customers. The fixed costs involve various expenses for vehicle,
administration, driver, insurance and taxes. The facility costs are also composed of variable and
fixed costs, including costs of facility construction, facility maintenance, equipment installation,
equipment maintenance and transport within facilities.
These cost functions are often assumed to be linear, but are more realistic if they are non-linear
(e.g. exponential functions or quadratic cost functions) (Jara Diaz, 1982). However, the
complexity of the modelling increases when non-linear functions are used. A non-linear facility
cost function was used in planning the location of freight terminals (See Section 8.2).
Transportation rates can be analysed using spatial price equilibrium models, which have been
widely applied in the area of transportation economics. The initial model was developed by
Samuelson (1952) and Beckmann et al. (1956) and Takayama and Judge (1964) extended this
model. These models presented the delivered price and quantity traded in each spatially
separated markets using supply and demand functions. They also showed that some factors of
demand and supply may influence the change in delivered prices within each region while a
change in transportation rate can influence the variation of delivery prices between regions.
Harker and Freizs (1986) extended these models to those that can be applied to intercity freight
flow prediction.

(Impact Models) 85

IMPACT
MODELS
5.1 OVERVIEW
Impact models provide a core modelling function for predicting the impacts by implementing
City Logistics initiatives. These models are used to estimate a wide range of impacts:
(a) Social impacts by alleviating traffic congestion and crashes
(b) Economic impacts due to changes in fixed costs and operation costs
(c) Environmental impacts in terms of C 0 2 or NO, emissions and noise levels
(d) Financial impacts by reducing costs to carriers and shippers
(e) Energy consumption by changing the amount of energy used

5.2 SOCIAL AND ECONOMIC

MODELS
The volume of goods that flow in a given market is the sum of individual shipper decisions.
These are influenced by the specific inputs and outputs of each production process.
Input-output analysis is a method for representing the structure of urban economies and
predicting usage rates of commodities by industry and firm size. It estimates the interaction
between industries by identifying the output of firms as well as the inputs required to produce
outputs.
Input-output tables show how the output of each industry is distributed among other industries
and how much input to that industry from other industries is required to produce that output.
Each element represents the sales from one industry or sector of the economy to the other.
Commodity flows between industries in monetary units are represented. Procedures can be
applied to estimate the demand of commodities for all firms in an area (Hutchinson 1974,
Roberts and Kullman 1979).
Demetsky (1974) presented a framework for summarising the movement of small goods within

86 (Ciy Logisria)
an urban area. A series of operations allows an input-output summary of commodity flows to be
established. Service zones are established within the study area and the origin and destination
patterns of small goods represented in terms of temporal, volumetric and spatial dimensions to
show the variation in demand for transport. Relationships were then developed between the flow
of goods and activity units to determine input-output tables. The interaction between zones and
industry groups were represented.
A combination of the input-output approach with the four step modelling process was recently
used in the Portland Commodity Flow Study (Taylor and Button, 1999). The model categorises
commodities by type, market segment and entry mode, then allocates an arrival facility and from
that, a destination facility. Commodities were converted to vehicle equivalents, then assigned to
the network and compared to vehicle classification counts.

5.3 ENVIRONMENTAL
MODELS
5.3.1 Overview
Environmental models predict the effects on the environment by implementing City Logistics
measures. Models are normally used to estimate the following three impacts.
(a) Noise
(b) Vibration
(c) Air pollution (NOx, COz, SPM, etc.)
There has been growing concern with regard to these three types of environmental impacts and
hence there is a need to predict them (Kroon er al. (eds.), 1991). The noise, vibration and
emissions of NOx and SPM generate effects on the local environment near roadways. COz
emissions also affect global warming of the earth. Recently more attention is paid to these types
of environmental issues.

5.3.2 Noise
There are a number of factors that affect the noise level generated by traffic at a given distance
from the road as indicated in equation 5.1.

where,

L, : noise level

(Impact Models) 87
V : average travel speed

Q : traffic volume

a, : proportion of large vehicles

I : distance from the road to the prediction point

ad: index of attenuation with distance by diffraction

The noise level is normally represented by the sound intensity level (dB) as shown in equation
5.2.

where,
I : sound intensity

I,: reference sound intensity (= lo.'* W/mZ)

The volume of large vehicles is an important factor, since large commercial vehicles produce
much larger noise than small passenger vehicles. The attenuation by diffraction of noise in
equation 5.1 depends on the structure of the road; the embankment and the elevated concrete
structure.
For predicting noise levels at a distance I from the road, we need to estimate the average speed
of vehicles, the traffic volume and the proportion of large vehicles using the road. Normally
traffic simulation is used for estimating these values. Macroscopic traffic simulation is more
suitable for large-scale networks, for example, for the entire main road network of a large city.
However, for estimating the environmental effects in a small district of a city or a specific link or
intersection, a microscopic model, which deals with the behaviour of each vehicle, is more
appropriate.

The traffic environment improvement manual in Japan (1989) provides the following equation
for estimating the traffic-induced noise.

I
1
L,, = Lc -8 - 2010g,,l+ l O I O g , , ( ~ -ttanh27~ -) + a , + a ,
d

LI = 87 + 0.2V + lolog,, (a, + lOa,)

where,

(5.3)

(5.4)

88 (City Logistics)
L,, : mean value of traffic-induced noise level (dB(A))

Lw : average power level caused by a vehicle (dB(A))

V : average travel speed ( k d h )
a, : proportion of small vehicles

a,: proportion of large vehicles (a,+ a2 = 1)

I : distance between the noise source to the prediction point (m)
d : average headway of vehicles (m)

d = 1000~

Q : average traffic volume (vehiclehour)

ad: index of attenuation with distance by diffraction (dB(A))

a, : index of other causes (dB(A))
For predicting the noise level, a survey of background noise level is required.

5.3.3 Vibration
Traffic-induced vibration sometimes produces serious problems to people and houses built on
weak ground. In addition, ground vibration can interfere with the performance of high precision
machines.
Traffic-induced vibration levels are related to a number of factors (equation 5.5).

where,

4: vibration level
V : average travel speed of vehicles

Q : traffic volume
u2 : proportion of large vehicles

a, : index of flamess of the road surface

a, : index of dominant frequency of the ground
as: index of road structure
a, : index of attenuation with distance of vibration
The vibration level is defined in equation 5.6.

(Impact Models) 89

(5.6)

where,

ae : effective acceleration revised considering the human reaction to vibration

a, : reference acceleration (= I 0.' m/sec2).

The flamess of the road surface is the main factor influencing vibration levels. If the pavement
surface is smooth enough, there may not be serious vibration problems on stiff ground. However,
on weak ground, which is generally composed of alluvial clay, heavy trucks can produce serious
vibration problems for people and buildings. Buildings can sometimes tilt due to the differential
settlement caused by traffic-induced vibration.
The dominant frequency of traffic-induced vibration is related to the bearing capacity of the
ground. Alluvial weak ground has a lower dominant frequency from the fact that vibration with
lower dominant frequency propagates further (Taniguchi and Sawada, 1979).
The type of road structure, for example, ground level, embankment or elevated concrete
structures also affect the vibration level. The attenuation with distance of vibration is related to
the dominant frequency of the ground and the soil type.
For predicting the effects of vibration, it is required to estimate the average travel speed of
vehicles, the traffic volume and the proportion of large vehicles. Besides these parameters,
information concerning the flamess of the road surface, the dominant frequency of the ground,
the road structure and the ground soil type is also required.
The traffic environment improvement manual in Japan (1989) presents the following equation
for estimating the traffic-induced vibration.

where,

L ,,: traffic-induced vibration level 10% lower than the maximum (dB)
Q' : equivalent traffic volume per 500 seconds per lane (vehicle/500 sec/lane)

Q, : traffic volume of small vehicles (vehiclehour)

Q2: traffic volume of large vehicles (vehiclehour)

90 (CiwLogistics)
V : average travel speed of vehicles (km/h)
M : number of lanes of roads in both ways

a, : index of flatness of the road surface

a, : index of dominant frequency of the ground
as: index of road structure

a, : index of attenuation with distance of vibration

c,,c2,c3,c4:
constant

5.3.4 Air Pollution

Emissions of toxic gases including NOx, CO2 are related to several factors (equation 5.8).

where,

L, : density of gases
V : average travel speed of vehicles
Q : traffic volume

7V : type of vehicles
1 : distance from the road to the prediction point
W : width of road
u : wind velocity
H : height of source
The emissions of toxic gases are highly dependent on the speed of vehicles. For example, the
typical curve for the emissions of NOx from vehicles indicates that the emissions decrease as
speed increases for low speeds and reaches a minimum when the speed is between 60-70 km/h.
For higher speeds emissions increase as speed increases.
Large trucks with diesel engines emit 15-20times more NOx than passenger cars. Therefore, the
vehicle type considerably influences the emissions of exhaust gases.
The diffusion of exhaust gases can be obtained by solving diffusion equations in three
dimensions, which incorporates the natural diffusion of gases and the effect of wind.

As an example, Ooishi (1996) estimated C 0 2 emissions associated with fuel consumption. First
fuel consumption can be estimated by the following equations.

(Impact Models) 9 1

For large truck

f, = __
5390 - 11.03V + 0.0758V2 + 587.6

For small truck

f, = 5442 - 1.194V + 0.0117V2 +81.2

For bus

f,=

7164 -13.00V

+ 0.1008V2 + 611.7

(5.1 1)

For passenger car

f, = 3569 -1.706V

+ 0.0128V2 + 105.2

(5.12)

(5.9)
(5.10)

where,
f,: fuel consumption (cm3/km)
V : average travel speed of vehicles (km/h)

Then, the C 0 2 emissions can be calculated by

Ec = f, , u,

(5.13)

where.
Ec : COz emissions (g-C)
U , : CO: emissions per unit fuel consumption (g-C/ cm3)

The typical value of U , is given as 0.623 (g-C/ cm3) for gasoline and 0.730 (g-C/ cm3) for diesel
by the National Institute of Science and Technology Policy in Japan (1992).
5.3.5 Applications

Kraus (1998) presented a model for estimating the length of trunk tours of freight vehicles for
evaluating the environmental and economic impacts of distribution systems. This model
determines the tour length from a central warehouse to customers related to the average distance
between them, the size of order and the vehicle capacity, average vehicle load, and the average
number of customers in a tour. The test results had good accuracy compared with the real tour
length.
Taniguchi and van der Heijden (2000a) presented a model for evaluating City Logistics
initiatives in terms of COz emissions. The model is composed of two sub-models, a model for
vehicle (pickupidelivery truck) routing and scheduling problem with time windows (VW-TW)
for each company as well as a dynamic traffic simulation model for the fleet of pickup/delivery
trucks and passenger cars on the road network within the city. They estimated the impact on C02
emissions as well as the cost reduction by implementing City Logistics measures. They

92 (Cig Logistics)
concluded that co-operative freight transport systems are most effective in reducing COz
emissions when freight demand increased.
A model that predicts air pollution and noise generated by freight vehicles at intersections was
developed by Ma (1999). This model was used to estimate the environmental capacity of a major
freight route in Osaka. The effects of controlling access during peak hour and promoting low
emission vehicles were determined.

5.4 FJNANCIAL
MODELS
5.4.1 Overview
Financial modelling is important for evaluating the feasibility and profitability of a project
related to City Logistics measures. Most public works projects can be evaluated based on
cost-benefit analysis. For example, building new expressway with the high benefit cost rate can
be accepted from the economic point of view. However, a project related to City Logistics
measures also requires good profitability as well as a high benefit cost rate, because it is
intrinsically related to private logistics activities. For example, even if a new underground freight
transport system can be regarded as feasible according to the cost-benefit analysis, it cannot be
adopted unless a high level of profitability can be demonstrated.

Thus financial models play an important role in determining whether or not to adopt City
Logistics measures. Sometimes profitability analysis reveals that City Logistics measures
require financial support from the public sector. The reasons for this are that projects related to
City Logistics measures often cannot internalise the positive external economy and require a
large initial investment.
Indeed the benefits caused by City Logistics measures. by improving the environment often go
to residents, while shippers or freight carriers do not receive them. If a project requires a
substantial initial investment, private companies hesitate since they do not like to take large
risks.

5.4.2 Cost-benefit analysis

Cost-benefit analysis is a common method used for evaluating public sector projects (Boardman
ef al., 1996). Here, the costs of construction and maintaining the system are estimated. The social
benefits generated by the system are also estimated over the projects life. Both the costs and
benefits are converted to the present value based on a social discount rate. The present value of
costs and benefits can be calculated using equations 5.14 and 5.1 5.

(Impact Models) 93

(5.14)

where,
C
,, : present value of costs
C, : cost for i th year

n : projects life
r : social discount rate

(5.15)

where,

B,, : present value of benefits

B8: benefits for i th year
Then the benefit cost ratio can be estimated using equation 5.16.

(5.16)

If the benefit cost ratio given by equation 5.16 is over 1, the project is considered feasible from
an economic point of view.
The net present value can be given by equation 5.17

NPV = B,, - C,,

(5.17)

Projects with higher values of NPV are regarded as being more desirable.

94 ( C i t Logistics)
~
The internal rate of return can be defined as the discount rate that satisfies equation 5.18.

(5.18)

The internal rate of return (IRRj indicates the discount rate when the annual benefits and costs
during the project period are equal. The higher the IRR is, the faster the investment can be paid
back. Therefore, a project with a higher value of IRR is a preferable project.
Costs are generally related to the construction, maintenance and operation of the system as well
as interest. For estimating the construction costs of new systems such as underground freight
transport systems, surveys of similar transport systems are required. The operation of automatic
control systems requires higher initial investment in computer hardware and other control
devices. But once the automatic systems are installed, they allow more flexibly in operating the
system than manually controlled systems.
Social benefits are composed of direct and indirect benefits. Direct benefits associated with City
Logistics measures include:
(a)
(b)
(c)
(dj

Time-related benefits
Travel-related benefits
Crash-related benefits
Environment-related benefits

Indirect benefits include:

(a) Opportunity for jobs
(b) New location of companies
(c) Increase in revenue of the municipality
City Logistics measures can generate a reduction in vehicle kilometres travelled or in the number

of vehicles used for operation, which result in a travel time reduction on the road network. The
reduction of travel time is beneficial for both passenger cars and freight vehicles. These are
referred to as time-related benefits. Normally times can be converted to monetary estimates
using value of time parameters.
City Logistics measures can also decrease vehicle operating costs including, vehicle running
costs, drivers' costs, petroleum costs and tolls. Sometimes a reduction in damaged goods can be
obtained. These benefits represent travel-related benefits.

(Impact Models) 95

Large trucks using urban streets are sometimes involved in serious crashes. Some City Logistics
measures are effective in reducing the total vehicle-kilometres travelled by large trucks, which
may decrease the number of crashes. A reduction in road crashes can save human lives and
serious injuries. In addition, fewer crashes also alleviate traffic congestion.
It is generally difficult to quantitatively evaluate environment-related benefits. However, the
emissions of NOx and COz can be converted to the monetary terms by estimating the costs of
making them harmless.

A quantitative evaluation of the indirect benefits listed above is more difficult than that for the
direct benefits. Therefore, the indirect benefits are often stated qualitatively. Modelling the
indirect effects of City Logistics measures requires further investigation.
5.4.3 Profitability Analysis
Since private companies are involved in logistics activities, the profitability of a project related
to City Logistics measures is vital for evaluating whether or not to adopt it. Profitability analysis
can be undertaken in four stages:
(a)
(b)
(c)
(d)

Set a price for service

Predict the demand based on the price
Calculate the income
Estimate the profit

The price for using a new City Logistics based service is an essential factor when undertaking
profitability analysis. The price can be considered a parameter, since it is often difficult to
identify the appropriate price for a new service. Once the price is set, the demand based on this
price and the level of service can be predicted. Sometimes stated preference surveys are
conducted to estimate the demand for new systems that have not yet been developed or
implemented elsewhere. The predicted income and profit highly depend on the accuracy of the
predicted demand levels. Special attention should be paid to fully informing people, who answer
questionnaires in stated preference surveys, regarding the characteristics of the new system.
The internal rate of return can provide useful information for evaluating the profitability of a
project, because projects related to City Logistics measures are often undertaken by a third party
supported by private as well as public sectors. The internal rate of return is an appropriate
indicator for evaluating a project that involves a mixture of input from both the public and
private sectors.

96 ( C i y Logistics)

5.4.4 Applications
One good example is the case of underground freight transport systems projects. An evaluation
of the new underground freight transport system in Tokyo using cost-benefit analysis concluded
that this project was economically viable (Ooishi and Taniguchi, 1999). It was estimated that this
system would result in substantial reductions in travel times, emissions and crashes. However,
the profitability of this project was questionable with the study concluding that at least 50% of
the construction costs would be required to be provided as assistance to ensure its financial
viability. Details of this study are presented in Section 5.6.

5.5 ENERGY
CONSUMPTION
MODELS
5.5.1 Overview
City Logistics measures will significantly affect the energy consumption of freight vehicles by
improving and rationalising urban freight transport systems. Models are required for estimating
the change of energy consumption by implementing City Logistics measures. Since trucks carry
most of the goods transported in urban areas models for estimating the fuel consumption by
motor vehicles need to be developed.

5.5.2 Factors Influencing Fuel Consumption

There are a number of factors that influence the fuel consumption of motor vehicles.
(a) Running pattern
- Average travel speed
- Driver behaviour
(b) Road conditions
- Surface quality
- Wetness of road surface
- Bends
- Gradients

(c) Vehicle conditions and other effects

- Vehicle weight
- Engine size
- Fuel type

(Impact Models) 91

Type of transmission
Manufacturing tolerances
Engine and vehicle wear
Poor maintenance
Cold start and cold running

Weather conditions
Use of auxiliary equipment

Everall (1968) proposed the following equations for estimating fuel consumption as a function
of a vehicles average travel speed.

f,= k,

+ kV + k,V

(5.19)

where,
f, : fuel consumption per unit distance
V : average travel speed of a vehicle
k, ,k , , k, : constants

The constant k, in equation 5.19 is associated with the fuel consumed per unit distance to
overcome rolling distance of vehicle and is approximately proportional to the vehicle mass
(Evans and Herman, 1976). The second term in the equation 5.19 is associated with various
time-dependent frictional loses and is approximately proportional to the idle fuel flow rate. The
third term in the equation 5.19 represents the aerodynamic resistance, which a vehicle meets
during running. This term is less important in the range of low average speed of vehicles.
According to running tests on actual roads (Yamada, 1980), fuel consumption is minimised at
average travel speeds between 50-60 km/h.
Driver behaviour has a considerable effect on the fuel consumption. In general, gentler
acceleration and braking leads to less fuel being consumed.
The Public Works Research Institute in Japan performed experiments in relation to the effects of
road surface quality on fuel consumption on a test track. It was found that the quality of the road
surface significantly affects fuel consumption. For example, fuel consumption on an unpaved
road was about 22% larger than on a sealed pavement. The wetness of a roads surface also affects
fuel consumption, with lower fuel consumption occurring with dry pavement conditions.
Roumegoux (1979) used computer models to study the effect of bends on fuel consumption for
inter-urban conditions. This analysis showed that traversing a bend in a 1-ton vehicle with lateral
acceleration of 0.2G consumed an extra 1 1itrellOOkm of fuel compared with a straight road.

98 (City Logistics)

Pelensky et al. (1968) studied the effects of gradient and proposed the following equation.

f, = 87 + 0.45r + o . 0 8 ~ ~ g e

(5.20)

where,
t : travel time
m : vehicle mass
g : acceleration of gravity

8 : gradient angle
Various factors concerning the vehicles condition also influence fuel consumption. The vehicles
weight significantly affects fuel consumption, since the rolling resistance is proportional to the
weight on each axle.
It is clear that larger engines consume more fuel. The weight and engine displacement of a
vehicle is designed so that the vehicle has the best performance, or the optimum use of engine
power and fuel. Therefore, expressing fuel consumption as a function of engine displacement is
almost equivalent to expressing fuel consumption as a function of vehicle weight.
It is generally recognised that diesel-engined vehicles consume less fuel than gasoline-engined
vehicles, particularly when the engine is cold.
Vehicles with an automatic transmission consume more fuel than those with manual
transmission. This is due to the loss of engine power in the transmission system and the
narrowness of the range of gear ratios. Recently continuous variable transmission has been
developed and is expected to realise better fuel consumption than manual transmission as well as
more comfortable driving.
Poorly maintained vehicles consume more fuel. Defective engine tuning, such as faulty ignition
timing, non-optimum carburetion, dirty filters and slack belts, increase fuel consumption.
The ambient temperature is also an important factor affecting fuel consumption. Any fall in the
ambient temperature causes an increase in fuel consumption even when the engine is hot. This is
due to greater heat losses and greater viscosity of the oil in the transmission system.
The engine not only drives the vehicle but also provides power for operating auxiliary and
accessory equipment, including air conditioning, power steering and power brakes. If the vehicle
has a number of accessory items installed and operating the effects of these items on the fuel
consumption cannot be neglected.

(Impact Models) 99

5.5.3 Models for Estimating Fuel Consumption

Yamada (1980) developed a model for estimating fuel consumption using on-road test data. The
model as shown in equation 5.21 includes type of pavement, stopped time, average gradient, use
of air conditioning and average travel speed for expressing the fuel economy, which is the
inverse of fuel consumption. Factors that have high correlation with other factors were removed.

f, = 6.372-0.716 rg -0.1 93tp-1.392 go - 1 . 4 1 2 ~+ 0.138V - 0.001V2

(5.21)

where,

f, : fuel economy (Wlitre)

I

: = 1 for gravel road, = 0 for asphalt pavement

fs : stopped time (min.)

g, : average gradient (5%)

a : = 1 when air conditioning is on, = 0 when air conditioning is off

V : average travel speed (km/h)

By partially differentiating equation 5.21 with respect to the average travel speed and equating it
to 0, the speed, which realises the best fuel economy, is found to be 69km/h.
Kanzaki et al. (1986) developed a computer model for estimating fuel consumption by
simulating the power transmission system of vehicles. This model has an advantage that fuel
consumption of vehicles under every possible running condition can be estimated without
running tests. In this model, engine torque and engine rotation under a given running condition is
firstly determined. Equations representing the operation of the power transmission system of
vehicles are used to calculate the torque and the rotation. Engine maps are finally employed to
determine the fuel consumption corresponding to the torque and the rotation identified.
Kraus (1998) estimated fuel consumption as a function of order size and distance from depot for
a 40-ton vehicle, based on data from the German environment authority.

100 (Cizy Logistics)

5.6 CASE STUDY (THE NEW UNDERGROUND

FREIGHTTRANSPORT
SYSTEM FOR TOKYO)
5.6.1 The Make-up of The New Underground Freight Transport System Network and The
Amount of Traffic Expected to Use This Network (Ooishi and Taniguchi, 1999)
A precondition for estimating the effects of constructing the new system, it was necessary to
know the make-up of the network, and to find out the amount of traffic, which can be expected
to use this new system. Therefore, giving priority to the roads which are anticipated to be used
often, plans for three stages of construction were devised, and based on this, analyses were
conducted on the effects of constructing such a system and the profitability of this system. The
roads to be constructed at each stage are shown in Figure 5.2. The primary road to be
constructed will be 70 km long, with the secondary road being 102 km long by adding 32 km,
for a total of 201 km in the entire network (the tertiary road of 99 km long to be constructed).
The results of the calculations of the anticipated traffic volumes using the roads to be
constructed at each stage are indicated in Table 5.1. Here, based on census data for road traffic
in 1990, the percentage of unloaded trucks travelling as opposed to loaded trucks was 45%.

Figure 5.2 Network of the New Underground Freight Transport System to be constructed at each
stage

(Impact Models) 101

Table 5.1 Anticipated traffic volumes for roads to be constructed at each stage

5.6.2 The Overview of The System

In Japan, unlike most countries, a high proportion of vehicles using the main roads are trucks.
Fifty-two percent of the total vehicles using the roads in the wards of Tokyo, which are of the
same or a greater scale as the metropolitan roads, are trucks. Also, the share of trucks on the
Tomei Expressway between Tokyo and Nagoya is approximately 70% during the day, and
reaches a maximum rate of 90% at night.
The amount of goods distributed tends to increase with economic development. Furthermore,
due to advances being made in services, including just-in-time shipping, and increased
frequency of small shipments, there has been an increase in the amount of traffic in the form of
highly mobilised trucks. With an increase in truck volumes come such environmental problems
as those related to traffic, including congestion and vehicles being parked on roadways, as well
as such problems as noise and air pollution. In addition to traffic-related problems of trucks, the
shipping industry in Japan is also faced with problems, including difficulty in securing a young
labour force, and the aging of its existing labour force.
Measures for dealing with traffic in the form of trucks are important for activating the economy
and improving the environment. There is a limit to the extent to which problems can be solved by
merely expanding the width of roads under city planning carried out by the metropolitan
government and continuing with already existing measures for dealing with traffic on roadways,

102 (City Logistics)

including restrictions on automobile emissions. It is for this reason that a new system is needed
for drastically addressing the various and assorted problems that have to do with road
distribution. Therefore, the Ministry of Construction has conceived a new underground freight
transport system as described below.

(a) Having vehicles use electricity as energy, so that they do not give off toxic gas
(b) Having vehicles travel on special separated roads, to avoid noise problems along the side of
roads
(c) Reducing road traffic in the form of trucks, by switching a suitable proportion over to the
new system
(d) In cities where it is difficult to gain additional space, having traffic travel through
underground tunnels, utilising the benefit that these vehicles do not produce toxic gas
(e) Having vehicles driven automatically, thus eliminating the need for people to drive them
(f) Making regular shipping possible
(g) Providing on occasion conditions for the transport of freight
Methods of transport that could be considered for the new underground freight transport system
described above include conveyor belts, tubes, containers, piggyback and dual-mode truck
methods. We have considered as conditions for selection the fact that it would be possible to
appropriately deal with social and economic problems, including lessening the traffic on regular
roadways and reducing air pollution, while maintaining the special superior characteristics of
automobiles, including the fact that they can bring people from door to door and can always be
utilised, and with it being possible to supplement superior characteristics of railway systems,
including being able to handle large capacities, regularity and energy conservation. As a result,
since it would be possible to combine the superior characteristics of automobiles with the special
superior characteristics of railway systems, the Ministry of Construction chose the dual-mode
truck method that would require a minimum of time and labour for loading and unloading
freight.
The new underground freight transport system is a system by which dual-mode trucks (DMT)
travel on regular roads outside of the system and on special railways within the system to carry
freight.

In this system, in order to be able to handle small shipments a weight capacity of 2 tons was set,
and freight is continuously carried from the place of origin to the destination. Also, within the
system trucks are driven automatically using as energy electricity collected from the guide way,
and on regular roads the trucks are operated by human drivers using batteries as the energy
source. The speed of travel within the system is 45 km/h,which is twice as fast as the speed on
regular roads, and driverless operation is possible. Furthermore, in addition to forming a network
in the city, this system includes a terminal for the transfer freight from one truck to another so

(Impact Models) 103

that trucks from the city can get to other urban areas smoothly. An image of this new
underground freight transport system is shown in Figure 5.3.

;--

main track

..

pallet style automatic conveyance

office

Figure 5.3 New underground freight transport system

5.6.3 Effects of Constructing The New Underground Freight Transport System and
Cost-benefit analysis

Effects of constructing the new undernound freight transport system

By considering the purpose of the project, the effects of constructing this new system are
travel-related benefits, time-related benefits, reduced COz emissions, reduced NO, emissions,
and a reduction in the number of traffic accidents. These are mainly attributable to a reduction in
the number of vehicles used for transporting freight and a greater speed of travel on the roads, as
a result of constructing the new system.
Estimates of the travel-related benefits were made using formulae for basic travel cost units by
vehicle type (Ministry of Construction, 1992), which was used in the 11th Five-Year Plan for
Road Improvement. The time-related benefits were estimated by calculating the difference based
on whether or not this new system had been constructed. The time-related cost estimates were
obtained by multiplying the total travel time on roads, which was obtained by dividing total
traffic volume on the roads by the average speed of travel, by the time-related value (The Japan
Synthetic Institute, 1995) by vehicle type. Also, the effect of reducing NO, emissions was
calculated based on the amount of NO, emissions and the cost of countermeasures for reducing
these emissions before and after the construction of this new system. The effect of reducing COz

104 (Ciq Logisrics)

emissions was also calculated based on the amount of COz emissions and the fixed cost for C 0 2
(Research lnstitute of Innovative Technology for the Earth, 1994) before and after the
construction of this new system. The effect of a reduction in traffic accidents was calculated by
multiplying the sum of total traffic volume by trucks before and after the construction of this new
system and by dual mode trucks (DMT) on the roads, by the rate of truck-related accidents
involving people and the average amount of payment of automobile liability insurance and
voluntary insurance. The annual benefits in these stages of construction based on this
methodology are presented in Table 5.2. More than 90% of the benefits are time-related benefits.
It was estimated that at the time of the construction of the primary network these benefits would
in total be approximately 910 billion yen a year, and approximately 980 billion yen a year at the
time of the construction of the secondary network, and approximately 1.2 trillion yen a year at
the time of the construction of the entire network.

Table 5.2 Benefits in the stages of construction

Cost-benefit analysis

In addition to evaluating this new system, based on the construction costs and the
above-mentioned construction-related effects, through a cost-benefit analysis for each of the
networks constructed at the individual stages, an analysis was also made from the perspective of
additional investment (for example, investment for additional roads for development from the
primary network to the secondary network). The construction costs for this new system are
indicated in Table 5.3, referring to similar construction costs, with the following used as
hypothetical conditions for an analysis:
(a) The construction period is 10 years, with the annual cost at 1/10 of the total construction cost
(b) The analysis period is a total of 60 years, from the time that the construction is commenced
until 30 years after the entire network has come into use

(Impact Models) I05

(c) The social discount rate in Japan is generally 4% to 6% and the rate used in this study was

5%
(d) As indicated in Figure 5.4, the investigation was conducted for 3 construction and usage
patterns
The results of the calculation of the benefit-cost ratio at each stage of construction based on the
above-mentioned conditions are shown in Table 5.4. As a result of conducting the construction
process from sections with high demand for a conversion into this new system, the benefits in the
primary network are 4.6 times the cost entailed for constructing and operating this new system.
Also, the benefit-cost ratio drops as the network is expanded, and even if an entire network were

to be constructed, it is anticipated that this would result in a sufficiently good effect with a
benefit-cost ratio of 3.5.

Table 5.3 Construction, maintenance and management costs for the networks at each stage
of construction

Infrastructure
(100 million yen)
Construction Other than
costs
infrastructure
(100 million yen)
Total
(100 million yen)
Vehicle purchase costs
(100 million yen)
Other maintenance and management
costs (100 million ven/vear)

Primary ISecondaryj

Tertiary
(entire network)

13.154

17,831

3 1,398

5,193

6.478

10.440

18.347

24,309

41,838

2,016

2,291

3,391

349

414

66 1

Note: The roadways, depots and transfer facilities are part of the
infrastructure, and everything else is classified as other than infrastructure, as
far as the construction costs are concerned.

106 ( C i v Logistics)

Period of Evaluation

t
2WO

Cnstruction
of primary
network onlly

2020

2010

2040

2030

205U

2060

i-------s.

-------

Beginning of Coming into use

Cnstruction
construction
through secondary
.... ....
network

Cnstruction
through
tertiary network

-------

....- ....

.- - .- - .-

Figure 5.4 Construction pattern and a usage pattern at each stage of construction

Primary
Secondary

4.6
4.1

Investigation concerning additional investment

The following examination was conducted in order to see whether or not the construction has
been effective in proportion to the investment, for the additional investment in the process of
going from the primary network to the secondary network and from the secondary network to the
entire network, at each stage of the construction plan. Here. with difficulty in securing
profitability for toll roads during recent years, there have been extensions in the periods for
collecting investments, and rather than the system itself not being able to be used for 30 years,
since it is of a make-up by which it can be used for more than 50 or 60 years, like the subway in
Tokyo, the analysis period will be set at 70 years after this new system has come into use. Here a
comparison is made as to the additional costs for constructing the additional portion of the
network, and those increased benefits, being incorporated in the benefit-cost ratio.
The results of the cost-benefit analysis for the additional investment calculated based on
above methodology is indicated in Figure 5.5. This shows that for the primary network
benefit-cost ratio has exceeded 4 even for the 30-year period after coming mto use and for
periods of 40 and 50 years, respectively, after coming into use, there was an increase in

the
the
the
the

(Impact Models) 107

4.00
3.00
2.W

Primary network
Secondary network (added)

Figure 5.5 Benefit-cost ratios for portion of added investment

5.6.4 Analysis of The Profitability of The New Underground Freight Transport System
Income and expenditures of the new underground freight transport system
The calculation of the income and expenditures for this new system, as well as an analysis of
profitability for the networks at each stage of construction is discussed in this section. First of all,
as far as the tolls for this new system are concerned, a toll of 180 yen has been adopted on
condition that traffic volume, which is expected to use this new system, is large and the total
amount of income from tolls will be at a maximum. Based on this, the numbers of vehicles,
which are expected to use this new system at each stage of construction, as well as the income
from tolls, are indicated in Table 5.5. Also, the expenditures for each stage of construction are as
indicated in Table 5.3.

108 (City Logistics)

Table 5.5 Income from tolls for each stage of construction

I
Number of loaded trucks using the system
(100 million truckdyear)
Total vehicle-km by loaded
Roads
trucks
((100million
.. .
I
~unaergrounal
vehicle-kdyear)
Income ( 1 00 million yedyear)
1

Primary Secondary

Far I

entire
network)

0.48

0.55

0.81

0.96

1.10

1.62

9.00

10.23

16.10

1,705

1,936

3.039

However, if the cost for DMT is renewed every 6 years, the other expenditures for the
maintenance and management costs for each stage of construction are calculated at 68.5 billion
yen a year, 79.6 billion yen a year, and 122.6 billion yen a year, respectively.
Evaluation of the project from the perspective of self-support accounthc
According to the above-mentioned plan at the stages of construction, an analysis of profitability
has been conducted for a case in which the operator is responsible for the total amount of the
construction costs. As shown in Table 5.6, the calculation results indicate an internal rate of
return of 3.8% even for cases of construction in the primary network only, which means that the
profitability is poor. For the construction of the entire network, the internal rate of return is 2.6%,
and the profitability is expected to be even worse. Also, with an extension in the analysis period
to 50 years after the entire network has come into use, changes have been estimated in the
internal rate of return. As indicated in Table 5.7, although the internal rate of return gradually
increased, there wasnt any large increase in profitability.

Table 5.6 Internal rate of return for each stage of construction

I
Tertiary (entire network)

Internal rate of return

2 6%

(Impact Models) 109

Table 5.7 Changes in the internal rate of return if the analysis period is extended

Primary
Secondary
Tertiary
(enti re network)

Analysis Period
Up to 30 years 1 Up to 40 years
after the entire
after the entire
network comes into network comes
use
into use
3.8%
4.1%

3.270

3.5%

Up to 50 years
after the entire
network comes
into use
4.2%
3.1%

2.6%

3.1%

3.3%

Study to assure the soundness of the project

Through the above-mentioned study, with the operator of this new system assuming full
responsibility for all construction costs, even if the number of years during which repayment is
made is extended to a certain extent, it is indicated that there would be difficulty in making the
repayment through income received in the form of tolls from users.
However, since the effects of such things as reducing traffic congestion and environmental
improvements due to the construction of this new system would be substantial when compared to
the costs related to construction and operation, it is thought that assistance to the operator of this
new system for construction costs would be justifiable. And so, with the analysis period of the
entire network at 30 years after the start of usage, in order to assure the soundness of the project,
below is an inquiry into what amount of assistance would be appropriate.
Figure 5.6 indicates changes in the internal rate of return while changes are being made in the
assistance rate if assistance was provided for a portion of construction costs for the infrastructure
(roadways, depots, transfer facilities), among the construction costs of this new system.
Supposing that 100% of the infrastructure construction costs were to be provided as assistance,
the internal rate of return for constructing the entire network would be 10.49.Also, even if 50%
of the infrastructure construction costs were to be provided as assistance, the internal rate of
return would be improved by about 5%, and would exceed the 5% social discount rate (adopted
in this study), which is used as a criterion for analysing profitability, so that it could be said that
this would result in a situation whereby the project would be financially viable.
On the other hand, even with low-interest financing which would be suitable to the internal rate
of return, it is thought that the soundness of the project could be assured and it would thus be
possible to conduct this project.

110 (City Logistics)

Internal rate of return (9%)

r
10.0

8.0
6.0
4.0
2.0

0.0

I
0

10

20

30

40

SO

60

70

80

90

1 0
1%)

Assistance rate for infrastructure construction costs

4
Construclioii of primarj

iielwork orif!

.Cl -

Construction up lo secondrr) network

-&-

Construction up to tertinr? networkientire networh)

Figure 5.6 Comparison of the internal rate of return, with a change in the assistance rate for the
infrastructure

This New Underground Freight Transport System in the wards of Tokyo would entail a high
benefit-cost ratio and be a useful project from the standpoint of society and the economy. Even if
an analysis of profitability were to show a low internal rate of return, it should not be concluded
that conducting this project would be unreasonable, as it is considered that it is necessary to
judge whether or not this construction is reasonable from the cost-benefit analysis, as well as to
judge whether or not public assistance should be granted, and to what extent the assistance for
construction costs should be given, from the analysis of profitability.

(Vehicle Routing and Scheduling) 111

VEHICLE
ROUTING
AND SCHEDULING
6.1 OVERVIEW
Vehicle routing and scheduling procedures provide core techniques for modelling City
Logistics. The significant features of truck operations in urban areas are:
(a) Pickup/delivery trucks travel in a chain by visiting a number of customers
(b) Several pickup/delivery trucks are usually operated as a group by a company. Therefore, it
is possible to reduce number of trucks used by some measures

(c) Each customer specifies a time window to be visited by the pickup/delivery trucks
Vehicle routing and scheduling problems ( V W ) involve an optimisation process of assigning
customers to trucks and determining the visiting order of customers and routes of vehicles. The
basic information needed for the V W is the location of customers, road network conditions,
travel times, traffic regulations, etc. In addition to this basic information, other specific
information for each customer, including the daily request for carrying goods, the designated
time windows, designated driver, is given to identify the optimal visiting order and the route
for each vehicle. This process can be manually done by skilled planners, but this is a very time
consuming task. However, the recent computerized system can quantitatively evaluate the VRP
by estimating the total transport costs, the accumulated running distance for each vehicle, the
accumulated operation time and the load factor of each vehicle.
Considering these features, some VRP models have been developed in the operations research
area. Solving the V W is difficult if the number of customers is large, because this problem
belongs to NP-hard (Non-deterministic Polynomial hard) class problems. Therefore, heuristic
approaches are required to obtain the approximate solutions for practical problems.

6.2 TRAVELLING
SALESMAN
PROBLEMS
Travelling salesman problems (TSP) are the basic problem for VRP (e.g. Lawler e f al., 1992).

11 2 (City Logistics)

This section describes the formulation of the TSP and techniques to identify the approximate
solutions using heuristic methods.
The TSP can be described as follows:
There are

IZ

cities (customers) and the cost cy to travel from the city i to the city j

is

given. A salesman (vehicle) starts from the home city (depot) to visit each city exactly once
and return to the home city. The problem is to find the optimal route (visiting order of
customers) that has the minimum total travel cost.
TSP can be formulated mathematically:

Minimise
Subject to

Xi,

=1

Vi E N

J=!

cc '
,ES,EI,

'tJ

VS

c N ( S # $ . S #N)

where,
2 : total travel cost
cy : travel cost from the city i to the city j

x~,: = 1 if travelled from the city i to the city j

= 0 otherwise

N : set of the number of cities = { 1, 2. 3. .., IZ }

S : nonempty subset of N ( not equal to N )
Equation 6.2 indicates that there is only a single route from the city i to other cities. Equation
6.3 shows that there is only a single route to come to the city j . Equation 6.4 indicates that
the route should visit each city exactly once and continuously (Hamiltonian circuit). The
equation 6.4 can be replaced with the following equation.

(Vehicle Routing arid Scheduling) 1 1 3

Both the equations 6.4 and 6.6 denote that no sub-tour is allowed as shown in Figure 6.1.
IS, represents the cardinality. The route shown in Figure 6.1 visits 4 cities but

1 1 x,, = 0
IES

and the equation 6.4 does not hold. S i n c e 1

=2

and

1x1- 1 = I , the

JE,h'\S

equation 6.6 also

l e S ,ss

does not hold

S
City

N\S

0:

Figure 6.1 Sub-tours that are not allowed in TSP

TSP is a combinatorial optimisation problem. It is very difficult to identify the optimal solution
of TSP with the large number of cities. If the entire set of solutions were enumerated (complete
enumeration method) that involves calculating the cost for all possible routes to determine the
best solution of TSP, the number of calculations would be about n! (Since the number of routes
is (iz-I)! and the number of calculating costs is i t ) .

If the basic computation time on a digital computer was 10.' seconds, the computation time for
solving TSP in case for 11 = 10 would be lo!* lo-*=0.036 seconds. However, when rz increases,
the computation time will increase dramatically as shown in Table 6.1. It is actually impossible
to solve large scale practical TSP using the complete enumeration method. Thus heuristic
methods must be used to identify approximate solutions. The application of Genetic
Algorithms (GA), Simulated Annealing (SA) and Tabu Search (TS) to V W will be discussed
later in this chapter.

114 (Cit?;Logistics)

Table 6.1 Rapid increase of computation time for solving TSP with the increase
of number of cities n

6.3 VEHICLEROUTINGAND SCHEDULING

WITH TIMEWINDOWS
6.3.1 Overview

The vehicle routing and scheduling problem with time window (VRP-TW-F) is defined as
follows. A depot and a number of customers are defined for each freight carrier. A fleet of
identical vehicles collects goods from customers and delivers them to the depot or delivers
goods to customers from the depot. For each customer a designated time window, indicating
the desired time period to be visited is also specified. For example, in the case of collecting
goods, vehicles depart from the depot and visit a subset of customers to pick up goods in
sequence and return to the depot to unload them. A vehicle is allowed to make multiple routes
per day. Each customer must be assigned to exactly one route of a vehicle and all the goods
from each customer must be loaded on the vehicle at the same time. The total weight of the
goods for a route must not exceed the capacity of the vehicle. The problem is to determine the
optimal assignment of vehicles to customers and the departure time as well as the order of
visiting customers for a freight carrier. V RP-TW-F explicitly incorporates the departure time of
vehicles as a variable to be determined.
Figure 6.2 demonstrates an example of vehicle routing and scheduling problem. Here, the
number of trucks ( m ) is equal to 5 and the number of customers ( n ) is equal to 8. The optimal

(Vehicle Routing and Scheduling) 115

solution is for the freight carrier to use 2 trucks out of 8 to visit 8 customers in the order shown.
We need to take into account the additional points to make the vehicle routing and scheduling
problem more realistic.
(a) Each customer has a time window to be visited by a truck
(b) There are multiple depots
(c) Travel times vary dynamically

u customer
Figure 6.2 An example of the vehicle routing and scheduling problem

6.3.2 Formulation
Here, we describe the vehicle routing and scheduling problem with time windows of customers
(Taniguchi and van der Heijden, 2000a; Desrosiers et al., 1995). The model in this section uses
one (forecasted) value to represent the travel time between the depot and customers. The model
thus is referred to as Vehicle Routing and scheduling Problem with Time Windows Forecasted
(VRP-TW-F) model. The objective of the VRP-TW-F model is to minimise the total costs. The
mathematical formulation is given below.

116 (City Logistics)

Minimise

Subject to
no 2 2

(6.10)

CN,= N

(6.11)

(6.12)
(6.13)
(6.14)
(6.15)

where,
(6.16)

where,

C ( t , , X ) : total cost (yen)

t, : departure time vector for all vehicles from the depot

t, = {f/,,, I=l, m

X : assignment and order of visiting customers for all vehicles

X = { X/I I = 1, m )

x, : assignment and order of visiting customers for vehicle I

xi= { n ( i ) I i = 1,N, }
n ( i ) : i th customer visited by a vehicle
d ( j ) : number of the depot (= 0)

(Vehicle Routing and Scheduling) 117

N,: total number of customers visited by vehicle

no:total number of d ( j ) in x i

rn : maximum number of vehicles available

c,,! : fixed cost for vehicle 1 (yen /vehicle)

6 , (x,): = 1; if vehicle I is used

= 0; otherwise
C,,,(t/,o,x i) : operating cost for vehicle I (yen)
CP,/(t/,o,x)): penalty cost for vehicle 1 (yen)
c,,! : operating cost per minute for vehicle 1 (yen /min)
ti,.(,,

: departure time of vehicle 1 from customer n(i)

T ( < , n ( / , , n ( i ) , i ~ ( i +:l average

))
travel time of vehicle 1 between customer n(i) and n(i+1)

at time

f,n(,l

t c , n (:, loadinglunloading
)
time

at customer n(i)

c ~ , ~(2)( ,:)delay penalty cost per minute at customer n(i) (yerdmin)

ce,n(i)
( t ): early arrival penalty cost per minute at customer n(i) (yerdmin)

N : total number of customers

D(n(i)): demand of customer n(i) (kg)
t;,o : last arrival time of vehicle I at the depot
ts : earliest time for starting truck operations
re : latest time for starting truck operations
W, (x,): load of vehicle I (kg)
WC,,
: capacity of vehicle I (kg)

Equation 6.7 represents the total cost to be minimised. The first, second and third terms of the
equation 6.7 represent the fixed cost of trucks, the operation cost of trucks and the early arrival
and delay penalty, respectively. Note that there are two types of costs in operating trucks; the
fixed cost that is independent from the operation time and the operation cost that varies with
operation time.
The total cost is composed of the vehicle costs, tax and insurance costs, etc. Whereas operation
costs are composed of the fuel, maintenance, personnel and tolls, etc. The penalty cost
composed of the early arrival penalty when arriving at customers earlier than the designated
time and waiting, as well as the delay penalty when arriving late at customers.
Figure 6.3 shows the typical penalty function for early arrivals and delays, where (t,',,,-ti,,,)

118 (CivLogisrics)

denotes the designated time window to be visited by a truck. The early arrival penalty can be
equal to the operation cost, since this is the waiting cost of trucks. The delay penalty depends
largely on customers, commodities and the urgency of delivering. The delay penalty in most
cases involves breaking contracts, or being disadvantaged in the next transaction instead of
actually paying money. Therefore, some methods to quantify the delay penalty are required.

cost
A

Figure 6.3 Penalty function for early arrival and delay

6.3.3 Heuristic Approach for Solving VRP-TW

The VRP-TW is one of combinatorial optimisation problems of NP-hard class (Reeves, 1993).
It is difficult to identify the exact solution of VRP-TW for large-scale networks. Therefore,
heuristics techniques have to be used to determine optimal solutions for VRP-TW.
Recently, a number of meta-heuristic methods that have been applied to the VRP-TW,
including Genetic Algorithms (GA). Simulated Annealing (SA) and Tabu Search (TS) (See
Section 2.6).
Genetic Algorithms (GA)
GA is a method for searching for an optimal solution based on the evolution of living
organisms. The GA technique initially determines a chromosome for an individual. The

(Vehicle Routing and Scheduliizg) 119

chromosome is composed of multiple genes. The actual structure of chromosome is

represented as genotype.
Figure 6.4 shows an example of genotype for VRP-TW. The genotype is a coded string with 7
genes in the figure that corresponds to the movement of a truck. This genotype demonstrates
that the truck starts depot, which is represented by 0, and visits customers 5, 2, 1, 3, 4 in this
order and returns to the depot (0).

Figure 6.4 An example of genotype for VRP-TW

The GA technique is based on generating a number of populations, where individuals within a

population represent a solution to the problem. Subsequent generations are determined by
procedures where parents are selected and new individuals (solutions) are produced based on
processing characteristics of the parents. This involves multiplication, crossover and mutation.
The multiplication is done using the fitness of individual. If the individuals with higher fitness
are solely preserved, it may converge to a local optimum. To escape such local optimum the
GA technique adopts crossover and mutation. The crossover rate and mutation rate should be
appropriately determined for each specific problem.
Simulated Annealing (SA)
SA is a neighbourhood search technique based on the analogy with the physical theory of how
material cools within a heat bath. The laws of thermodynamics describe that the probability of
an increase in energy as given by equation 2.4 in Section 2.6. The SA technique applies the
idea to the combinatorial optimisation. That is, the energy in thermodynamics can be replaced
to the objective function of the optimisation problem. The cooling parameter should be
determined for a particular problem.
Tabu Search (TS)
TS is a neighbourhood search technique that makes systematic use of information on past
variable exchanges to generate new solutions. The best solution in the neighbourhood can be
chosen unless it is in the tabu list. The tabu list keeps a record of the number of times that an
exchange was selected to determine a new solution. Both tabu tenure and aspiration criteria are

120 (City Logistics)

parameters that should be determined for a particular problem.

6.3.4 Model Application

Comparison of heuristic algorithms
Taniguchi et al. (1998) applied three meta-heuristic approaches. GA, SA and TS to VRP-TW
and compared the performance of these three techniques. Figure 6.5 shows a simple test road
network used for comparing the performance of the heuristic techniques. Travel times between
nodes (customers) for all vertical and horizontal links were set at 12 minutes and 18 minutes
respectively. One depot is located in the centre of the network and 10 customers were randomly
selected from all other nodes in the network. The maximum number of trucks was limited to
ten. The weight of goods to be picked up at each customer was randomly distributed between

250 - 2,000 kg. The width of time windows (t:,,, -trt(i,) as shown in Figure 6.3 was 1 hour.
The penalty was set to a very large value after 1 hour after the end of time window. The delay
penalty cost was set at 5 times than that of the waiting time cost of trucks, which is
c ~ , ~=( 5, )ce,n(8)in Figure 6.3.

Parameter estimates for the three heuristic techniques were determined using a benchmark
problem whose exact solution is known. The following methods and parameters gave best
solutions for the problem described above.

18 minutes
A

12 minutes

0node
Figure 6.5 Test road network

(Vehicle Routing and Scheduling) 121

(GA)
Method of mutation: Deletion and Insertion
Crossover rate: 0.7
Mutation rate: 0.14
(SA)
Methods of changing the array of number: Exchange
Control parameter: 0.999

(TS)
Method of move: Exchange
Tabu tenure: 20
These procedures and parameters were used to compare the three heuristic techniques. Figure
6.6 shows the average performance of the three techniques when applied to ten problems that
were specified by randomly selecting ten customers (nodes) from the test network (Figure 6.5).
The graph shows the average discrepancy for each technique versus computation time. The
discrepancy is defined as the difference between the value of the objective function and the
best-known solution. A personal computer (MICRON MILLENNIA XKU, CPU Pentium I1
300MHz, Memory 64 MB) was used to perform the calculations.

d-GA

SA
+TS
-D

4 8 16 32 64 128 200
computation time (sec)

Figure 6.6 Comparison of performance in three techniques

122 (CivLogistics)
Good solutions were found by all three techniques. TS reached the best-known solution with
the shortest computation time. However, it is difficult with the TS technique to determine the
appropriate tabu tenure that is suitable for a specific problem. Relatively small changes in the
value of objective function were achieved for the GA. The solutions were largely affected by
the selection of random numbers for generating the first population and the crossover and
mutation procedures. SA gives more stable solutions with good accuracy, if the computation
time is extended to 200 seconds. The selection of random numbers for probabilistic moves in
the neighborhood search had almost no effect on the final solution. Therefore, while the
accuracy of the TS technique was best for shorter computation time periods, SA is the best
model when longer computation time was allowed.
Effects of relaxing time windows
The effects of relaxing the time windows were also examined using the small network shown
in Figure 6.5. Here, the width of the time window was set at 4 levels: 1, 2, 4 and 8 hours. For
the first level, the starting time

ric,,

was randomly selected from every hour between 9 a.m. to

4 p.m. For the second and subsequent levels, the time windows had either the same starting or
finishing time as the first level but with extended width. Based on the comparison of the 3
heuristic techniques in the previous section, SA was used to investigate the effects of varying
the width of the time windows.
Figure 6.7 shows the change in total cost, total travel time and the required number of trucks
when the width of the time windows was increased. Each of these measures decreased as the
width of time window increased. Hence, the width of time window at customers considerably
affects the performance of urban pickup/delivery truck routing and scheduling. For example
when comparing the costs where the width of time window is only 1 hour with that of 4 hours,
the total operating costs and total travel time were reduced by 13 9% and 9 % respectively. This
would not only reduce the operating costs of shippers or freight carriers, but also provide many
benefits for society at large due to the reduction in traffic congestion and environmental
problem in urban areas.
The model also estimated that the total waiting time at customers decreased from 29.4 to 0
minutes when the width of the time window was extended from 1 hour to 3 hours. This
reduction in waiting time would also have a positive effect on traffic flow. since waiting trucks
often impede traffic flow by occupying roadside space in the vicinity of customers.

(Vehicle Routing and Scheduling) 123

total travel time -C+ number of trucks

. ....... .......... .... .. ....

2.2
L

2.1

5
c T
_-- -,

>

'a

L
I,

'a

>

7 -

1.9

- 1.8

width of time window (hours)

Figure 6.7 Effects of increasing the width of time windows

In some modem logistic systems, relaxation of the width of time windows is not allowed. This
is common in the distribution of goods at large retailing store chains. However, it is considered
important to quantify the effects of increasing the width of time windows on the total delivery
costs and the number of trucks required and this information may provide shippers or freight
carriers and customers with an incentive to relax the strict time windows in the routing and
scheduling of urban pickup/delivery trucks.

6.4 VEHICLE ROUTINGAND SCHEDULING WITH TIMEWINDOWS AND

TRAFFIC
SIMULATION
6.4.1 Framework
This section presents a more complicated model of the VRP-TW combined with dynamic
traffic simulation. This type of model is required to deal with the realistic traffic situations. In
congested urban road networks, freight carriers' decisions on vehicle routing and scheduling
affect road traffic conditions. Then freight carriers need to consider updated traffic conditions,
which are generated by a number of decisions on the delivery planning of freight vehicles as
well as the route choice of passenger cars. An integrated model of VRP-TW and dynamic
traffic simulation is required to simulate such iterative procedures.

124 (City Logistics)

Taniguchi et al. ( 1 9 9 9 ~ )presented an integrated model of VRP-TW and dynamic traffic

simulation. Figure 6.8 presents the models framework. The model is composed of two
sub-models, a model for vehicle (pickup/delivery truck) routing and scheduling problem with
time windows (VRP-TW) for each company as well as a dynamic traffic simulation model for
the fleets of pickup/delivery trucks and passenger cars on the road network within the city.
The VRP-TW model determines the optimal assignment of vehicles to customers and the
departure time as well as the order of visiting customers for freight carriers, which becomes
input to the dynamic traffic simulation model. The dynamic traffic simulation model is based
on a macroscopic dynamic simulation BOX model (Fujii et al., 1994). This model estimates
the average travel time on each link in 30 minutes intervals. The VRP-TW model is then
re-solved using the updated average travel times on each link obtained from the BOX model.
Thus, the average travel times for each link are represented by a step function, in 30-minute
time intervals. The model therefore, incorporates time dependent travel times. Successive
iterations of both the VRP-TW model and the BOX model continue until a pre-defined
convergence criterion is satisfied.

I
c

Model for vehicle routing and

scheduling problem with time window
(VRPTW)
Freight
J carrier B
Freight 7

carrier^

Optimal routing
and scheduling
for each company

Average travel
time on each link

Dynamic traffic simulation model

(BOX model)

Figure 6.8 Model framework

(Vehicle Routing and Scheduling) 125

6.4.2 Dynamic Traffic Simulation

The dynamic traffic simulation model is based on BOX model (Chapter 4) that was originally
developed by Fujii et al. (1994). It is similar to the cell transmission model (Daganzo, 1994).
The BOX model is essentially a macroscopic model but because the origin and destination of
each vehicle is defined, it is actually a hybrid macroscopic/microscopic model. Vehicles are
assumed to choose the shortest path when they arrive at a node using an estimated travel time.
The BOX model consists of two components, a flow simulation and a route choice simulation
as shown in Figure 6.9. A sequence of boxes is used to represent each link. Groups of vehicles
flowing out of a box and into the next box during the scanning interval represent

Flow simulation

- Congested link
- O D trakel times

Description of link flow

both for pickup1
deliveq trucks and
passenger cars

Traffic
conditions

Driver's route
choice

Decision of route
choice based on travel
time estimation

Figure 6.9 Structure of the BOX model

the flow on links. There are two assumptions for modelling links, which the maximum flow
during a scanning interval is the same for all sections on links and no inflow and outflow is
allowed in the middle of links. A consequence of the first assumption is that only the lowest
section of a link can be a bottleneck, where a congestion queue starts. Two states of flow:
congested flow and free flow are represented. The time for a vehicle to proceed through a
congested queue T is given by

126 (Cit?;Logistics)

(6.17)
where,

F, : number of vehicles in a congestion queue

Ce : exiting traffic volume
The exiting traffic volume is the traffic volume that can flow out of the last section of a link
into the successive link. The time that is required to go through the running area without any
queue T, is estimated by

(6.18)
(6.19)

where,

L, : length of flowing area without any queue

V, : free running speed

K : traffic density
KO: critical traffic density
Q,,,

: maximum traffic volume

The modified BOX model shown in Figure 6.10 explicitly describes the flow of
pickup/delivery trucks that depart from a depot and return to the same depot. Pickup/delivery
trucks are converted to passenger car units and the first-in-first-out rule is assumed on all links.
The model was further modified to identify the arrival of specific vehicles at assigned nodes
(customers).
The simulation model described above estimates travel times on each link and allows link costs
to be determined. Drivers are assumed to compose a "cognitive map" for each link based on its
estimated link cost. Drivers then choose routes based on their minimum travel cost from the
current node to the destination using their cognitive map. It is assumed that all drivers have
some experience in driving within the defined network. The function for estimating the link
cost is:
C, = T,, + q ,

(6.20)

(Vehicle Rouririg and Scheduling) 127

where,
C, : estimated cost on link k

T,, : travel time on link k at time r

qk : disturbance term

In this study the disturbance term qk is assumed to be normally distributed with zero mean
and variance

0;

node

as represented by

Running area
without queue

! Congested area
with queue

node

2
I

BOX
model

6down stream
passenger cars
EEZ

pickup/delivery trucks converted

to passenger car units

Figure 6.10 Link representation using the modified BOX model

6.4.3 Model Application

Test conditions
The model described in the previous section was applied to a test network with 25 nodes and
40 links as shown in Figure 6.11 (Taniguchi and van der Heijden, 2000a). This network
includes three types of roads, urban expressways, arterials and streets with free running speeds
of 60 km/h, 40 km/h and 20 km/h respectively. Although this network is a hypothetical one, it
is similar to Kobe City in Japan. Therefore, the travel times vary for the same distance

128 (CivLogistics)

I
I

2 min.

14

I
I

Urban expressway (V, = 60 kmh)

____

Arterial (Vf = 40 kmh)

I
I

I
I

Street (V,= 20 kmh)

V, : free running speed

Figure 6.11 Test network

depending on the type of road used. Note, that the length of links shown in Figure 6.12 does
not precisely indicate their geometric distance. Any node within the network can both generate
and attract passenger car traffic. These nodes are referred to as centroids and are also candidate
nodes to be visited by pickup/delivery trucks. Ten freight carriers are assumed to operate a
maximum of 12 pickupldelivery trucks in this network. Each freight carrier has one depot.
Three different types of trucks, having a capacity of 2 , 4 and 10 tons respectively can be used.
However, up to four trucks of each type can only be operated by each carrier. The passenger
car equivalence rates, operating costs and fixed costs for each type of pickup/delivery truck are
based on results from recent studies of truck operations in Japan. The number of customers for
each carrier was randomly generated between 5 and 24 as shown in Table 6.2. The actual nodes
to be visited for each carrier were also determined randomly from all nodes in the network.

(Vehicle Routing and Scheduling) 129

Freight carrier
A
B
C

Depot node number

19
13
3

Number ot customers

18
17

10
20

8
22
11

Three types of time windows were permitted in this study, time windows with one hour, time
windows for a.m. (9:OO-12:OO) or p.m. (13:OO-17:OO) and no time window. The type and
starting time of each customers time window was based on a recent survey (1997) in Kobe and
Osaka area. The average travel time on each link for the scanning interval is provided by the
dynamic traffic simulation. In this study the scanning interval used was 30 minutes. When
initially calculating the optimal routes and schedules, the average travel times on each link
were assumed to be equal to the travel times using free running speeds.
Simulating current traffic conditions
Firstly, hypothetical traffic conditions were simulated to provide a benchmark for estimating
the benefits of introducing the advanced routing and scheduling systems. The optimisation
model for VRP-TW was applied to the test network. The value of the objective function for the
chosen solution was 1.2 - 1.5 times higher than that of the best solution and the average load
factor of trucks was around 20% lower than the best solution. This discrepancy was based on
the survey on the improvements found by several freight companies in Kobe City. This
solution is assumed to represent the current pickupldelivery truck operations before introducing
advanced systems. The pickup/delivery truck traffic was estimated to account for 14% of all
traffic within the network. This percentage of pickup/delivery trucks is almost identical to the
actual conditions within the Kobe area.
Effects of introducing advanced routing and scheduling system
The effects of freight carriers introducing Advanced Routing and Scheduling System (ARSS)
on road traffic were investigated. The ARSS provides optimal routes and schedules using the
VRP-TW model described in the previous section. Three cases were considered, ARSS
penetration rates of O%, 50% and 100%.In the case of 50% penetration rate, freight carriers A,
B, C, D and E shown in Table 6.2 introduced ARSS. The demand for freight transport at each

130 (Ci3Logistics)
customer was increased to 1.5 and 2.0 times the base case. This increase by 1.5 and 2 times the
base case is taken as parameters. For example, if the freight demand were annually increased
by 2%, the demand would increase to be 1.5 times the base case in 20 years and 2 times in 35
years. The demand for freight transport in the base case was hypothetically set, as the
distribution of demand coincides with the actual demand distribution given by survey in
Osaka-Kobe area.
Figures 6.12 and 6.13 show the effects of the penetration rate of ARSS on COz emissions with
the demand for freight transport. In Figure 6.13 both the change of C 0 2 emissions and the
demand for freight transport values are normalised by the value of base case. C02 emissions
were determined using an established fuel consumption relationship. Fuel consumption was
estimated using the average travel speed of vehicles on each link. This estimation was carried
out for passenger cars and pickup/delivery trucks respectively and then combined together.
Figure 6.13 indicates that the normalised C 0 2 emissions for the 100% penetration rate were
reduced by 8.3% from that for the 0% penetration rate, when the demand was doubled.
However, the normalised COz emissions increased by 14% when the penetration rate was 50%
compared with a zero penetration rate, when the demand was doubled. This is attributed to an
increased use of larger trucks that produce more COz emissions than smaller trucks. Freight
carriers generally intend to use larger trucks, which allows them to reduce the costs compared
with using smaller trucks for carrying the same amount of goods. Table 6.3 shows the travel
time for different types of trucks. Trucks with capacity of 10 tons travelled longer periods than
the other smaller trucks when the penetration rate was 50% and the demand was double the
base case. In Table 6.3 the total travel times of trucks for penetration rates of 50% and 100%
are lower than that for the penetration rate of 0% in three cases of normalised demand by 7.1 14.7%. Therefore, it can be noted that ARSS considerably helps alleviate traffic congestion.
The reduction of travel time of both pickup/delivery trucks and passenger cars is estimated to
be 1.5 - 2.3%.

(Vehicle Rourirzp and Scheduling) 13 1

+Penetration
rate = 100% . .
- 0 - Penetration rate = 0%

. .Penetration rate

= 50%

~~~

400

350

150

"

100
1

1.2

1.4

1.6

1.8

2.2

Normalised demand for freight transport

Figure 6. I2 Effects of penetration rate of advanced routing and scheduling system on change in
COz emissions with increasing demand for freight transport

~~

--D- Penetration

rate = 100%

. . t . .Penetration rate = 50%

- 0 - Penetration rate = 0%
rr

1.5
2
Normalised demand for freight transport

2.5

Figure 6.13 Effects of penetration rate of advanced routing and scheduling system on change
in normalised COz emissions with increasing demand for freight transport

132 (Civ Logistics)

Table 6.3 Change in travel time of different types of trucks by advanced routing and
scheduling system

Effects of cooperative freight transport system

There are various types of cooperative freight transport systems, for example. cooperation in
building and operating a common depot, cooperation in carrying goods by common
pickupldelivery trucks and cooperative use of information systems. Here, cooperation in
carrying goods is examined as shown in Figure 6.14. This figure demonstrates cooperation
between two freight carriers D and H, with each freight carrier having numerous customers to
visit. In this cooperative freight transport system, each freight carrier collects goods from
customers within its neighbourhood. As a result the total travel distance and the required
number of trucks will be reduced. Here, it is assumed that the VRP-TW-F model is fully used
by all ten freight carriers in both cases with and without cooperative freight transport.
Figures 6.15 and 6.16 show the effects of the cooperative freight transport system on the C02
emissions by vehicles of freight carriers D and H who participate in the system with the
increase of demand for freight transport. The figure demonstrates that the CO2 emissions
produced by freight carriers who participated in the cooperative system can be reduced
considerably. The level of COz emissions produced by these freight carriers involved in
cooperation remains at almost at the same level as the base case when doubling the demand for
freight transport, while it doubles from the base case without cooperation. The normalised C02
emissions with cooperation were reduced by 51.8% from that without cooperation, when the
demand was doubled. Figure 6.15 indicates that the C02 emissions in the base case without
cooperation are lower than those with cooperation. This is due to the increased travel times
experience by trucks with the capacity of 10 tons as shown in Table 6.4. The table also shows

(Vehicle Routing and Scheduling) 133

that total travel times for all truck types are reduced in all of the demand cases considered. This
produces benefits relating to better traffic flow conditions on the network. Table 6.5 shows the
total costs of freight carriers D and H. The total costs are reduced by 23-29% after
implementing a cooperative freight transport system for the three demand levels. Therefore,
cooperative freight transport systems are effective in reducing costs at various levels of
demand.

Freight carrier D

Freight carrier H

,-

(a) before cooperation

r----7
Freight carrier D
l

Freight carrier H

(b)after cooperation

@ :Customers nodes
0 :Other nodes

:Depot

Figure 6.14 Cooperative freight transport system by changing the customers to be visited

134 (CioLogistics)
~

+without

cooperation . . t . .with-cooperation

'2 130
-0
120
$ 110
100
.E 90
f

,g
h

80
70
60
1

1.2

1.4

1.6

1.8

2.2

Normalised demand for freight transport

Figure 6.15 Effects of cooperative freight transport system on change in COz emissions by
freight carriers D and H with increasing demand for freight transport

1.2

1.4

1.6

1.8

2.2

N o d s e d demand for freight transport

Figure 6.16 Effects of cooperative freight transport system on change in normalised COz
emissions by freight carriers D and H with increasing demand for freight
transport

(Vehicle Routing and Sclzeduliizg) 135

Table 6.4 Change in travel times of different types of trucks for freight carriers D and H
by cooperative freight transport system

Wighout coperation (min.)

Table 6.5 Change in total cost of freight carriers D and H by cooperative freight
transport system
Normalised demand for freight transport
Cost without cooperation (yen)
Cost with cooperation (yen)
Change by cooperation (%)

1
154,337
118,218
-23.4

1.5
190,135
139,257
-26.8

2
225,996
159,853
-29.3

(Vehicle Routing and Scheduling with ITS) 131

VEHICLE
ROUTING
AND SCHEDULING
WITH ITS
7.1 OVERVIEW
The development of vehicle routing and scheduling procedures using ITS has the potential for
significantly reducing the costs of distributing goods within urban areas. The integration of
information and communication systems with optimisation models will form the basis of a
large number of City Logistics schemes in the future (Kohler, 1997). This chapter describes
some initial applications of vehicle routing and scheduling using ITS.

7.2 PROBABILISTIC VEHICLE ROUTINGAND SCHEDULING WITH TIME

WINDOWS
Current vehicle routing and scheduling procedures do not explicitly account for the random
nature of travel times in urban traffic networks. New technology allows information on the
variability of travel times in urban areas to be obtained automatically (Section 3.2). As well,
stochastic programming allows the variable nature of parameters to be incorporated within
optimisation procedures.
This section describes how probabilistic models can be developed for vehicle routing problems.
Probabilistic models use information relating to the variability of travel times as part of their
formulation. A procedure for estimating the benefits of this approach will also be defined.
Example applications will be provided.

7.2.1 Just-In-Time Distribution

Recently urban freight transport faces many challenges. There is much pressure on freight to
provide higher levels of service within the framework of just-in-time (JIT) transportation
systems with lower costs.

138 (City Logistics)

Shippers are the customers of freight carriers who send or receive goods from companies or
persons. Shippers generally tend to maximise their levels of service, which includes the costs,
times for picking up or delivering, reliability as well as trailing information. It is now a
common requirement for carriers to arrive at customers within specified time windows for
pickup/delivery in the manufacturing and retail sectors. A recent survey in Osaka and Kobe in
Japan (Traffic Planning Council for Kyoto, Osaka and Kobe Area, 1997). found that freight
carriers were required operate with designated arrival times or time windows for 52% of goods
delivered and for 45% of goods collected in terms of weight. Such strict time windows have led
to smaller loads of goods being transported more frequently. The reliability of goods delivery
has become important in JIT transport systems.
Freight carriers try to minimise their costs of collecting and delivering goods to customers to
maximise their profits. They need to provide higher levels of service to customers at lower total
costs. This is especially important when they are requested to arrive at customers within a
specified time period. However, freight carriers often face difficulty in operating their vehicles
on urban roads due to traffic congestion. This has led to the inefficient use of trucks, where
smaller loads are being transported and trucks often have to wait near the location of customers
when they arrive earlier than the designated time.

7.2.2 Stochastic Programming

Stochastic programming are techniques that aim to find optimal solutions for problems
involving uncertain data (Birge and Louveaux, 1997). When the value of parameters are
uncertain, modellers tend to use the expected values of parameters in formulating optimisation
problems. However, when information relating to the nature of the uncertainty of future events
exists, stochastic programming techniques can incorporate the probability distribution within
the objective function.
The value of the stochastic solution (VSS) measures the possible gain from solving the
stochastic model that explicitly incorporates the distribution of random variables within the
problem formulation. VSS represents the value of knowing and using the distributions of future
outcomes. It is relevant to problems where the future is uncertain and no further information
about the future is available. VSS estimates the cost of ignoring uncertainty when making a
decision (i.e. determining a solution).
There have been limited applications of stochastic concepts within vehicle routing and
scheduling problems (Laporte et al.. 1992). Stochastic programming with recourse involves
specifying the objective function with planned route costs as well as explicitly incorporating
expected penalty costs, where the stochastic travel times are used to estimate the penalties

(Vehicle Routing and Scheduling with ITS) 139

incurred for arrivals at customers outside the designated time windows. The travelling
salesman problem with time windows (TSPTW) incorporates penalties for truck arrivals at
customers, outside designated time windows.
Current approaches to vehicle routing only include one value for determining the travel time
between customers. Average travel times are generally used as a forecasted single value
representation of the time taken to travel between customers. However, stochastic
programming can be used to incorporate the random nature of travel times by incorporating an
expected penalty cost in the objective function.
This section presents a probabilistic model for optimising vehicle routing and scheduling

7.2.3 Travelling Salesman Problem with Time Windows and Stochastic Travel Times
A model for representing the travelling salesman problem with time windows and stochastic
travel times is presented in this section. A stochastic programming with recourse model is
defined, formulated and tested, where the travel times between customers are considered
random variables and represented using probability distributions. The probabilistic model is
shown to be more reliable than the deterministic model, leading to significant reductions in
costs in congested conditions.
The travelling salesman problem with time windows (TSPTW) incorporates penalties for truck
arrivals at customers, outside designated time windows. Vehicles incur a penalty proportional
to the period of time they arrive outside of customers time windows. The TSPTW consists of
finding a permutation X of the set [O, I...N, O } corresponding to a tour which starts and ends at
the depot, i.e. x (O)=O and x ( N + l ) = 0, such that every customer is visited exactly once and the
total routing cost is minimised (Equation 7.1).

where,
C, :
i n ( ,n i i c , ,:

wo(,:

unit service and vehicle running cost

travel time between customers n(i-1) and n(i)
waiting time at ithcustomer visited

140 (City Logistics)

dn(,: delay time at ithcustomer visited

The arrival time at customers, t:(!,, depends on the departure time at the previous customer,
t:(l-,, , and the travel time between customers (Equation 7.2). The departure time from

customers depends on the arrival time, any waiting time and the service time (Equation 7.3).

Numerous solution procedures and heuristic have been developed for solving the TSPTW and
vehicle routing problem with time windows (Lawler et al., 1992; Daganzo, 1987; Desrosiers,
1988; Tsitsiklis, 1992, and Homaifar et al., 1993). Since one value (e.g. mean) is typically
used to represent the travel times between customers, this model will be referred to as the
Forecasted model (Psaraftis, 1995).

In practice, the time taken for a truck to travel between customers is random and influenced by
a number of factors, including the route, the vehicle, the driver and the weather experienced.
The Forecasted model fails to address the uncertainty in travel times, and hence in the times of
waiting and delay at customer sites.
In an attempt to incorporate more realism in the representation of travel times a number of
approaches have been developed. Laporte et al. (1992) presented a number of formulations of
the vehicle routing problem with stochastic travel times but the delay penalty function only
considered the overall route duration. Thus penalties for individual customers were not
incorporated.
Malandrandaki and D a s h , (1992) presented a time dependent travelling salesman problem
that treats travel times as step functions. Travel times were assumed to depend on the time of
day and time windows are represented as constraints.
Recently the reliability of goods delivery at customers has become a major issue within the
framework of just-in-time (JIT) transportation systems. Intelligent Transport Systems (ITS)
involving the application of advanced technologies to help reduce the costs of goods
distribution. ITS has the potential to improve the productivity of local delivery vehicles (Euler,
1990), by increasing the reliability of commercial vehicle arrival times. Several ITS have

(Vehicle Routing and Scheduling with ITS) 141

already been developed to automatically collect vehicle travel times (Kurosaki et al., 1993).
Historical travel time patterns within urban areas can be represented by probability distributions
and incorporated in vehicle routing and scheduling procedures.
Stochastic Programming are techniques that aim to find optimal solutions for problems
involving uncertain data (Birge and Louveaux, 1997). The random nature of travel times can be
incorporated within vehicle routing procedures to help increase the reliability of goods
distribution. Probabilistic models are those models that consider travel time between customers
to be stochastic. There are two general approaches for incorporating random travel times using
stochastic programming techniques, chance constrained programming and stochastic
programming with recourse. Both models are presented below.
7.2.4 Chance Constrained Model
Chance constrained programming involves minimising route costs while including constraints
to identify legal routes in terms of probability statements relating to the vehicle arriving within
customers time windows (Equation 7.4). If a vehicle does not arrive at a customer within a time

window it is considered to be a service failure SFn,, . This approach defines a route as being
feasible if there is a low probability or chance of the truck arriving at customers outside of the
specified time windows.

The service failure acceptability threshold 6,,,,

is the maximum probability of a truck not
satisfying the time window that is allowed. For a route to be feasible, the probability of service
failure must be below the acceptability thresholds for all customers.
Chance constrained programming however, does not allow the costs of corrective actions to be
taken into account (Gendreau et al., 1997). In order to incorporate the expected costs of not
satisfying customer time windows it is necessary to use a recourse model. This is presented in
the next section.

7.2.5 Recourse Model

The recourse model involves specifying the objective function with planned route costs as well
as explicitly incorporating expected penalty costs, where the stochastic travel times are used to
estimate the penalties incurred for arrivals at customers, outside the designated time windows
(Equation 7.5). The first term represents the service cost, while the second term incorporates

142 ( C i q Logistics)
the expected running cost as well as waiting and delay penalties. This can be simplified by
assuming that the travel times are independent (Equation 7.6).

where,

5:

a vector of random variables corresponding to the travel times. Each realisation

of 5 is called a "state of the world";

t,,,
5 i,ni,+, :

,(c

wnf,

dnf ,({ ) :
~

the travel time between the ithand (i+l)'hcustomer in the state of the world 5;
the waiting time for the ithcustomer in state of the world
the delay time for the Ch customer in state of the world

5;

5;

fni,j,nfr+li:the mean of the travel time distribution

Statistical Procedures

Now, by assuming

ffZ( j

- 1) & t f q 1. - 1 1,

1are independent random variables the probability

that the vehicle arrives at a customer can be estimated using a joint probability distribution
(Equation 7.7).

Since a vehicle arriving at a customer before the start of the time window must wait, to
determine the departure time probability distribution it is necessary to consider 3 cases (Figure
7.1).

(Vehicle Routiiig and Sclieduliiig with ITS) 143

I Departure

Arrival Time

Time

Arrival Time

Time

Figure 7.1 Departure time distribution at previous customer

Case 1

If there is very little chance of a vehicle arriving at the previous customer after the start of it's
time window:

144 (CityLogistics)

where, w is a small number (e.g. 0,001) then, the time the vehicle leaves the previous
customer can be calculated deterministically (Equation 7.9).

Otherwise, there is a reasonable chance of the vehicle arriving at the previous customer both
before and after the start of its time window, then,

In this case the departure time distribution is a mixed distribution with a discrete component
(Equation 7.12) as well as a continuous component (Equation 7.13).

(7.12)

(7.13)

If there is very little chance of a vehicle arriving at the previous customer before the start of it's
time window,

then, the time the vehicle leaves the previous customer can be calculated stochastically
(Equation 7.15).

(Vehicle Routing and Scheduling with ITS) 145

(7.15)

7.2.6 VSS Estimation Procedure

A two-stage procedure was developed for estimating the benefits (cost savings) of using the
Probabilistic model. The 1'' stage involves determining the optimal routes for both models
(Figure 7.2). Here, the different representation of travel times is used to determine the optimal
route for each model. This process allows network factors as well as experienced travel times to
directly influence the travel times input to both models.

Travel Time
Forecasts

Forecasted
Model

Travel Time
Distributions

Model

-&KalRzte

Model)
- -(Probabilisitic
-----I

Figure 7.2 Route determination process

The 2nd stage in estimating the benefits associated with the Probabilistic model involves
assessing the performance of the optimal routes determined from the 1'' stage process (Figure
7.3). This involves simulating travel conditions and calculating the costs of trucks using the
optimal routes produced from both the Probabilistic and Forecasted models. This allows the
performance of the Probabilistic model to be evaluated.

146 (CivLogistics)

-------1

Optimal Route
(Forecasted Model)

Travel Time
Distributions

Simulated
Travel Times

Route Costs
Model

Value of Stochastic
Solution (VSS)

L------J

Optimal Route
(Probabilisitic Model)

Figure 7.3 Model evaluation process

7.2.7 Case Study

An application of the modelling procedures developed for a customer data set (Table 7.1)

randomly generated to represent customers within a 20kmx20km area with the depot located in
the centre will be used to illustrate the main differences between the Probabilistic and
Forecasted models. Here, one truck starting from the depot must visit the 8 customers and
return to the depot without any load capacity constraint.

Table 7.1 Customer data set used in case study

The values of the cost items relating to this problem are given in Table 7.2.

.
(Vehicle Routing and Scheduling with ITS) 147

Truck running cost (C,)

14 (yedminute)

Waiting time penalty rate (Ce,"(])) 14 (yedminute)

Delay time penalty rate (Cd,n(l))

70 (yedminute)

The optimal routes were determined for both models using complete enumeration (Table 7.3).
For the Forecasted model a travel speed of 30 km/h between customers was used. A displaced
lognormal distribution was used to represent the travel times for the Probabilistic model. An
average travel speed of 30 km/h was used with a maximum travel speed of 40 km/h and a
standard deviation of 15 km/h. Distances between customers were estimated assuming a grid
network. The optimal route from the Probabilistic model is 2.3 percent shorter than of the
Forecasted model (Table 7.3).

Table 7.3 Optimal Routes

Distance

Model

Optimal Route

Forecasted

[0 1 5 6 4 3 8 2 7 01 78.5

Probabilistic

[O 1 6 5 8 3 4 2 7 01

76.7

The arrival times of the truck at customers assuming a travel speed of 30 km/h for both optimal
routes are shown in Figures 7.4 and 7.5. These trajectory diagrams also show the designated
time windows for each customer. It can be seen from Figure 7.4 that although the truck would
not arrive at any customer outside the designated time windows, the truck would arrive at
customers 6 and 2 just before the end of their specified time windows. This is in contrast to the
optimal route from the Probabilistic model with the truck typically arriving before or much
earlier within the designated time windows (Figure 7.5).

148 ( C i v Logistics)

"St0"lBI
NO.

80

70

60

50

40
3

30
20

G
10
5 1
0
0

60

120

180

240

300

360

420

480

540

GOO

GGO

0
720

Time (Minutes)

Figure 7.4 Optimal route from the Forecasted model (Travel speed 30kmh)

ustomei
0

No

80

7
70
2
4
60

50
Distance

40

(km)

30

20

6
10

1
0

GO

120

180

240

300

360

420

480

540

600

660

0
720

Time (Minutes)

Figure 7.5 Optimal route from the Probabilistic model (Travel speed 30km/h)

(Vehicle Routing and Scheduling with ITS) 149

Figure 7.6 shows that only a minor increase in travel times would lead to delay penalties being
incurred at customers 2 and 6. Large increases in travel time would lead to significant delay
penalties being incurred using this route. In contrast, only minor delays would be experienced
by customers 4 and 2 using the optimal route from the Probabilistic model when travel times
increased substantially.
The performance of both routes was determined by simulating travel conditions and comparing
costs of using them. Travel times were randomly generated from lognormal distributions. The
standard deviation of travel speeds was always 50 percent of the average travel speeds. The cost
components were compared when the average travel speeds were varied (Figures 7.6 and 7.7).
The waiting time increased with average travel speed for both models. Both routes exhibited
similar decreases in travel time costs as average travel speed increased. However, substantially
lower delay costs were incurred when the optimal route for the Probabilistic model was used
compared with the optimal route for the Forecasted model.

15

20

25

30

35

40

WDelay
UTravel

45

Average Travel Speed ( k d h )

Figure 7.6 Cost components using the Forecasted models optimal route

0 Travel

15

20

25

30

35

40

45

Average Travel Speed (krnlh)

Figure 7.7 Cost components using the Probabilistic models optimal route

150 (City Logistics)

The reduction in operating costs for using the Probabilistic model increases as the average
travel speed decreases (Figure 7.8). Significant cost reductions are estimated when average
speeds fall below 25kmh. These cost savings represent the value of the stochastic solution or
the benefits of using the Probabilistic model.

::
Percent 20
10

~~~

~~

0
15

20

25

30

35

40

45

Average Travel Speed (kmlh)

Figure 7.8 Value of the Stochastic Solution

7.2.8 General Results

To investigate the potential benefits to be gained by using the Probabilistic model the costs of
using the optimal routes obtained from both models were compared for 15 sets of customers.
The location and time windows for 8 customers were randomly generated for each set. The
location of customers were generated randomly from a bivariate uniform distribution within an
area of 400 square kilometres (20kmx20km), representing a medium sized urban area. The
depot was located in the centre of this area. The start and length of time windows for each
customer were randomly generated to represent the patterns identified from recent surveys of
JIT practices in Japan (Traffic Planning Council for Kyoto, Osaka and Kobe Area, 1997).
The optimal routes for both the Forecasted and Probabilistic models were determined by
complete enumeration using an average travel speed of 30kmh with a maximum travel speed of
40 km/h and the standard deviation of travel speed being 15 k m h . The lognormal distribution
was used to represent the variation of travel times for the Probabilistic model (Taylor et al.,
19961.
The costs of using the optimal routes of both models were estimated using simulation. Random
variates were generated from the lognormal distribution to represent travel times. For all
distributions the maximum travel speed was set to be 413 the average travel speed with the
standard deviation always being 50 percent of the average travel speed. For each average travel
speed level considered the costs were determined after using the optimal routes 100 times.

(Vehicle Routing arid Scheduling with ITS) 15I

These were combined for the 15 customer sets to estimate the total costs of using the optimal
routes of each model.
The total average operating costs of using the optimal routes for Forecasted and Probabilistic
models for the 15 customer sets under a variety of speed conditions are shown in Figures 7.9
and 7.10. Similar patterns are evident for both models for the travel and waiting times. However,
a substantial reduction in delay costs are incurred when the optimal routes from the
Probabilistic model are used compared with those of the Forecasted model.

w Delay
0 Travel
0 Waiting

15

20

25

30

35

40

45

Average Travel Speed (kmlh)

Figure 7.9 Costs using the optimal routes from the Forecasted model

200
150

w Delay

Thousand
Yen
100

OTravel
OWaiting

50

'1
15

20

25

30

35

40

45

Average Travel Speed (krnlh)

Figure 7.10 Costs using the optimal routes from the Probabilistic model
Significant reductions in operating and total costs were estimated when congested conditions
were encountered when the optimal routes from the Probabilistic model were used (Figure 7.1 1).
A reduction in operating costs of 4.5% and 17.3% were estimated when average travel speeds
fall to 20km/h and 15km.h respectively. Thus significant reductions in operating costs were
estimated when average travel speeds were less than 25km.h.

152 (City Logistics)

20

percent::
5

\\\\__-

--

:-

- -

-~ -

15

20

25

30

35

40

45

Average Travel Speed (kmlh)

Figure 7.11 Reduction in costs using Probabilistic model

Overall, for the 15 sets of customers there was a 6.4% reduction in the distance travelled using
the optimal probabilistic routes. These shorter routes would be more likely to experience less
variation in travel time and therefore have lower expected delay costs than longer routes. The
reduction in total distance travelled using the optimal routes from the probabilistic model
determined here would lead to significant environmental benefits. Considerable savings in
operating costs would also be experienced. Thus the Probabilistic model has potential for being
an effective City Logistics measure.
This section has shown how stochastic programming allows the variation of travel times to be
incorporated within vehicle routing procedures. This permits a more realistic representation of
urban traffic networks to be incorporated within routing procedures. A recourse model was
developed that incorporated the expected delay and waiting costs to be taken into account.
Significant reductions in routing costs were estimated using the probabilistic when conditions
became congested. Thus, the probabilistic model was shown to be more reliable.

7.3 PROBABILISTIC VEHICLE ROUTINGAND SCHEDULING WITH TIME

WINDOWS AND TRAFFIC
SIMULATION
7.3.1 Introduction
This section focuses on vehicle routing and scheduling procedures using advanced information
systems in urban areas. Freight carriers have depots and their pichp/delivery trucks depart the
depot and visit customers with designated time windows for collecting or delivering goods and
return to the depot. This section presents a probabilistic vehicle routing and scheduling model
with dynamic traffic simulation, which explicitly incorporates variable travel times. The
uncertainty of travel times affects the identification of the optimal routes and schedules of

(Vehicle Routing and Scheduling wit11 ITS) 153

pickup/delivery trucks on very congested urban roads. Recently the implementation of

advanced information systems allows freight carriers to use historical and real time travel time
data on urban roads. This section presents a model that quantifies the benefits of considering
the uncertainty of travel times in order to rationalise logistics systems and reduce the negative
impacts of goods movement on the en\Jronment.
'
Figure 7.12 shows the measured distribution of travel times by AVI (automatic vehicle
identification) on the Hanshin Expressway in Kobe City, Japan. The AVI system reads the
licence numbers of vehicles at two points of upstream and downstream of the lane to identify
the vehicle travel time between these two points. This figure illustrates that the travel times at
the evening peak time (1 8:OO-19:OO) fluctuate considerably, while those during off-peak times
do not. As several dominant peaks of travel times appear in the graph of the evening peak time,
the lognormal distribution better represents the fluctuation of travel times than the normal
distribution. If there were such wide range of variability in the travel time distribution, it would
be difficult for freight carriers to form their delivery planning using one value of forecasted
travel time. Using the historical data of travel time distribution as shown in Figure 7.12 instead
of only one value of travel time would be beneficial for delivery planning.

-~

~~

I
measured x normal distnbution +log
~

~~

normal distribution

80
70

60

rJ 50
v

x
2
- 40
a,

330

c4

20
10
0

10

12 14

16

18 20

22

24

26

28

30 32

34

36 38 4 0 42 44 46

48

50

Travel time (minutes)

(a) Travel time distribution at off peak time (1200-13:OO)
Figure 7.12(a) Measured travel time distribution by the automatic vehicle identification at
Hanshin Expressway

154 (City Logistics)

measured +normal distnbution +log normal distnbution

-

16
14

d 10
v

r6

4
2

0
2

10 12

14

16

18 20

22

24 26 28 30 32 34 36 38 40 42 44 46 4 8 50

Travel time (minutes)

(b) Travel time distribution at the evening peak time (18:00-19:00)
Figure 7.12(b) Measured travel time distribution by the automatic vehicle identification at
Hanshin Expressway

A number of researchers (e.g. Koskosidis et al., 1992; Bramel er al., 1996; Taniguchi er al.,
1998) have investigated vehicle routing problems with time windows (VRP-TW). Other

researchers have studied stochastic vehicle routing and scheduling problems (e.g. Dror et al.,
1989; Powell et a/.,1995; Gendreau er al., 1996). Most research in this area has focused on
dynamic routing and scheduling that considers the variation in customer demands. However,
there has been limited research on routing and scheduling with probabilistic travel times.
(Laporte et al., 1992; Taniguchi era/., 1999d, 1999e, 2OOOb)

7.3.2 The Probabilistic and Forecasted Model

Framework
Figure 7.13 presents a framework of the model presented in this section. The model is
composed of two sub-models; (a) a model for probabilistic vehicle (pickup/delivery truck)
routing and scheduling problem with time windows (VRP-TW-P) for each company and (bj a
dynamic traffic simulation model for the fleet of pickup/delivery trucks and passenger cars on
the road network within the city.

(Vehicle Routing and Scheduliiig with ITS) 155

Model for probabilistic vehicle routing

and scheduling problem with time
window (VRP-TW-P)

.........

T
Optimal routing
and scheduling

Distribution of
travel times on
each link

Dynamic traffic simulation model

(BOX model)

Figure 7.13 Model Framework

The model for VRP-TW-P is defined as follows. A depot and a number of customers are
defined for each freight carrier. A fleet of identical vehicles collects goods from customers and
delivers them to the depot or delivers the goods to customers from the depot. For each
customer a designated time window, specifying the desired time period to be visited is also
specified. For example, in the case of collecting goods, vehicles depart from the depot and visit
a subset of customers to pick up goods in sequence and return to the depot to unload them. A
vehicle is allowed to make multiple trips per day. Each customer must be assigned to exactly
one route of a vehicle and all the goods from each customer must be loaded on the vehicle at
the same time. The total weight of the goods for routes must not exceed the capacity of the
vehicle. This problem is used to determine the optimal assignment of vehicles to customers and
the departure time as well as the order of visiting customers for a freight carrier. VRP-TW-P
explicitly incorporates the distribution of travel times for identifying the optimal routes and
departure times of vehicles.

156 (City Logistics)

The optimal assignment of vehicles to customers and the departure time as well as the visiting
order of customers for each freight carrier, becomes input to the dynamic traffic simulation
model. The dynamic traffic simulation model is based on a macroscopic dynamic simulation
BOX model (See Section 4.6). This model estimates the distribution of travel times on each
link in 1-hour intervals. The VRP-TW-P model is then re-solved using the updated distribution
of travel times on each link obtained from the BOX model. Thus, the distribution of travel
times for each link is represented by a normal distribution, in 1-hour time intervals. The model
therefore, incorporates time dependent travel times.
VRP-TW-P Model
This section describes a mathematical model of the VRP-TW-P that was introduced in the
previous section. The model minimises the total cost of distributing goods with truck capacity
and designated time constraints. The total cost is composed of three components; (a) the fixed
cost of vehicles, (b) vehicle operating costs, that are proportional to the time travelled and spent
waiting at customers, and (c) delay penalties for designated pickup/delivery time at customers.
The model can be formulated as follows:
Minimise
(7.16)

where,

(7.17)
(7.18)

Subject to

(7.19)

2~~N
=

(7.20)
(7.21)
(7.22)
(7.23)
(7.24)

(Vehicle Routing and Scheduling with ITS) 157

where
(7.25)

c ( ~ , , x ) : total cost (yen)

to: departure time vector for all vehicles at the depot
to={t,,,~~m
= l },

X : assignment and order of visiting customers for all vehicles

X={X,~1=1,m )
x, : assignment and order of visiting customers for vehicle 1

x I = { n ( i )I i = 1, N,]
iz(i)

: node number of i th customer visited by a vehicle

d ( j ) : number of depot (= 0)

N,: total number of customers visited by vehicle 1

no:total number of d ( j ) in x i
m : maximum number of vehicles available
c,, : fixed cost for vehicle 1 (yen /vehicle)

6 , (x,): = 1; if vehicle 1 is used

= 0; otherwise

C,,, (ti,o,x,) : operating cost for vehicle 1 (yen)

Cp,,(r,,", x,): penalty cost for vehicle 1 (yen)

c,,, : operating cost per minute for vehicle 1 (yen Imin)
t,,fi(,l: departure

time of vehicle 1 at customer n(i)

~ ( t ; , , , , ) , n ( i ) , n ( i + l ) :) average travel time of vehicle 1 between customer n(i) and n ( i + l ) at

time

f,,,,,)

t c , n ( :, )loading/unloading time at customer n ( i )

p i , n ( (ti,"
i ) ,t,x,): probability in which a vehicle that departs the depots at time t,,oarrives at

customer n ( i ) at time t
( t ): delay penalty cost per minute at customer n ( i ) (yedmin)
c ? , ~((t~) :)early arrival penalty cost per minute at customer n ( i ) (yedmin)

N : total number of customers

D(iz(i)): demand of customer n ( i ) (kg)
t;,o : last arrival time of vehicle 1 at the depot

158 (CiwLogistics)

r, : starting of possible operation time of trucks

t, : end of possible operation time of trucks

W, (x,): load of vehicle 1 (kg)

We,,: capacity of vehicle 1 (kg).

The problem specified by equations 7.16

- 7.25 involves

determining the variable X , that is,

the assignment of vehicles and the visiting order of customers and the variable t o ,the departure
time of vehicles from the depot. Note, that n ( 0 ) and n ( N ,+ 1) represent the depot in equations
7.17 and 7.18.
The distribution of travel times is required in equation 7.16 for determining the expected value
of operating costs and penalty costs. This is distinguishes the Probabilistic model (VRP-TW-P)
from the Forecasted model (VRP-TW-F). The Forecasted model, VRP-TW-F adopts only one
value to represent travel time instead of a probability distribution. The dynamic traffic
simulation calculates the distribution of travel times, which can be approximated by the normal
or lognormal distribution for every hour. Then the updated normal or lognormal distribution
can be input to the probabilistic model.
Figure 7.14 shows the penalty for vehicle delay and early arrivals at customers. The time period
( f : ( , J-r;,,,)

of the penalty function defines the width of the soft time window in which vehicles

are requested to arrive at customers within. If a vehicle arrives at a customer earlier than

t j ( 8 Jit,

must wait until the start of the designated time window and a cost is incurred during waiting. If
a vehicle is delayed, it must pay a penalty proportional to the amount of time it was delayed.
This type of penalty is typically observed in goods distribution to shops and supermarkets in
urban areas. Multiplying the penalty function and the probability of arrival time as shown in
Figure 7.14 can identify the penalty of early arrivals and delay at customers for the Probabilistic
model. The Forecasted model uses a single point estimate of the arrival time of a truck at a
customer. Therefore, the penalty for early arrival and delay can be estimated by multiplying the
penalty function by the predicted amount of waiting or delay time.
The problem described herewith is a NP-hard combinatorial optimisation problem. It requires
heuristic methods to efficiently obtain a good solution. Recently several researchers have
applied heuristic algorithms such as Genetic Algorithms (GA) (e.g. Thangiah et al., 1991),
Simulated Annealing (SA) (e.g. Kokubugata et al., 1997) and Tabu Search (TS) (e.g. Potvin et
al., 1996) to obtain approximate solutions for the VRP-TW-F. The model described in this
paper uses a GA (See Section 2.6) to solve the VRP-TW-P. GA was selected because it is a

(Vehicle Routing and Sclieduliizg with ITS) 159

Penalty (yen/min.)

Arrival time
Probability of arnval t i h e
I

I
1
I

Penalty of early arrival aFd

delay (yedmin.) I
~

A@val time (min.)

II

I
~

Amval time (min.)

Figure 7.14 Penalty of early arrival and delay at customers for the probabilistic model
heuristic procedure that can simultaneously determine the departure time and the assignment of
vehicles as well as the visiting order of customers.
Dynamic Traffic Simulation Model
The dynamic traffic simulation model is based on the BOX model that was originally
developed by Fujii et al. (1994) (See Section 4.6). The modified BOX model explicitly
describes the flow of pickup/delivery trucks that depart from a depot and return to the same
depot. Pickup/delivery trucks are converted to passenger car units and the first-in-first-out rule
is assumed on all links. The model was further modified to identify the arrival of specific
vehicles at assigned nodes (customers).

160 (City Logistics)

The travel times on each link vary within the day. The output of the BOX model is the updated
distribution of travel times on each link. The distribution of travel times representing the
interval of one hour was formulated by using the four-hour data on travel times. For example,
the distribution of travel times representing the time interval, 8:OO-9:00a.m. can be formulated
by using data between 6:30 - 10:30 a.m. Then this empirical distribution was used in the VRPTW-P model.

7.3.3 Application to Test Network

Test Conditions
The model described in the previous section was applied to a test network with 25 nodes and
40 links as shown in Figure 7.15. This road network is comprised of same type of roads with
free running speed of 40 kmlh.Any node within the network can generate and attract passenger
car traffic. These nodes are referred to as centroids and are also candidate nodes to be visited by
the pickup/delivery trucks. Ten freight carriers are assumed to operate a maximum of 12
pickup/delivery trucks in this network. Each freight carrier has one depot at the location
indicated in Table 7.4. Three different types of trucks, having a capacity of 2, 4 and 10 tons
respectively can be used. However, up to four trucks of each type can only be operated by each
carrier. The passenger car equivalence rates, operating costs and fixed costs for each type of

@2.67km

4 = 10.7km

Figure 7.15 Test road network

(Vehicle Routing and Scheduling with ITS) 161

pickup/delivery truck are based on results from recent studies of truck operations in Japan as
shown in Table 7.5. The number of customers for each carrier was generated randomly between
14 and 22. The actual nodes to be visited for each carrier were also determined randomly from
all nodes in the network. The freight demand at each customer was determined based on the
distribution of freight demand in Kobe City.

Table 7.4 Location of freight carriers' depots and number of customers

1 1 I

Table 7.5 Characteristics of pickupldelivery trucks

Capacity
Passenger
of i k equiv:;rate
(ton)
(pcuhehicle)

Operation
cost
(yedday)
14.02
17.54
23.27

Fixed
cost
(yedday)
10,417.5
11.523.1
13.789.7

Three types of time windows were permitted in this study, time windows with one hour, time
windows for a.m. (9:OO-12:OO) or p.m. (13:OO-17:00), and no time window. The type and
starting time of each customers time window was based on a recent survey in Kobe and Osaka
area. The dynamic traffic simulation provides the distribution of travel times on each link for
the scanning interval. In this study the scanning interval is 1 hour. When initially calculating
the optimal routes and schedules, the average travel times on each link were assumed to be
equal to the travel times using free running speeds.

162 (CigLogistics)
Dynamic traffic simulation requires information on passenger car behaviour, as well as optimal
routes and schedules of pickup/delivery trucks, produced by the VRP-TW-P model. This
includes the departure time and visiting order of customers. Passenger cars in this study include
actual passenger cars and trucks other than those that are considered in the optimal routing and
scheduling model. Passenger car origin-destination (OD) tables for every hour were estimated
using traffic generation rates at each centroid and the probability of 0 - D choice. The number of
passenger cars for each hour was generated using a temporal demand pattern based on the
traffic census conducted in Kobe City.

v a r i a t i o n o f c a r generation

Case 1
6500 veh. /day
(car generat ion)

j, 1

case 2
5450 veh. /day
(car generat ion)

1[

Accumulation o f
t r a v e l times
I

Case 3
4315 veh. /day
(car generation)

E v a l u a t i o n o f f r e i g h t c o s t s and
t h e environment
Figure 7.16 Flowchart for the calculation
The model described here uses an iterative
time provided by the dynamic simulation
calculation procedure. Here, the generation
vehicleslday k 10% (175 vehicledday) at

procedure for day after day. Therefore, the travel

fluctuates with the day. Figure 7.16 shows the
of total passenger cars was varied between 4375
all nodes. After 10-days operation, the iterative

(Vehicle Routing and Scheduling with ITS) 163

procedure was stopped. The fluctuation of travel times on each link was within 5%. At the end
of 10th days operation, freight carriers produced their optimal planning routes and schedules.
They then encountered three different traffic conditions, cases A, B and C. The total generation
of passenger cars per day was 6500, 5450 and 4375 (vehicledday) at all nodes for cases A, B,
and C respectively. The generation for case C was the same as in the previous 10-day operation.
This generation was uniformly located at all nodes. Case A was more congested than case B
that is more congested than case C, with the average speeds of 30, 35 and 38 km/h for cases A,

B and C respectively.
Results
Table 7.6 shows the change in total costs for the 10 freight carriers for the three traffic
conditions. This indicates that the probabilistic model reduced the total costs compared with
the Forecasted model in all cases. The reduction of total costs from Forecasted model increases
in cases A and B with the higher level of congestion than case C. It means that freight carriers
can obtain more benefits from using the Probabilistic model when the traffic congestion
becomes worse. The value of stochastic solution (VSS) is defined as the reduction in total costs
from using the Probabilistic model instead of the Forecasted model. The VSS for the three
cases was between 11% and 17%, which is considerable.
Table 7.6 Change of total costs for 10 freight carriers

I Model I

costs

I Case 1 I Case 2 I Case 3 I

operation cost

-0.8

10.5
change (%)

-23.7

11.0

-34.5

-29.4

Table 7.6 also indicates that the delay penalty decreased by 24-35%' using VRP-TW-P model,
which means that it provides better service to customers by reducing the risk of delay. However,

164 (CityLogistics)
the early arrival penalty increased by 8-1 1%. The fixed cost in all three cases increased by
approximately 5% and the operation cost remained at the same level as the VRP-TW-F model.
The small increase in the fixed cost is due to the slight increase in the number of trucks used by
10 freight carriers as shown in Table 7.7. Here, a one 2-ton truck was removed and two 4-ton
trucks were added in the optimal routing of the VRP-TW-P model.

Table 7.7 Change in the number of trucks (10 freight carriers, case A, B, and C)

I Model
VRPTW-F
model

VRPTW-P
model

I Type of vehicle I Number of vehicles1

2 ton truck
4 ton truck
10 ton truck
total
2 ton truck
4 ton truck
10 ton truck
total

3
14
18
0
5
14
19

Let us examine why using the VRP-TW-P model can decrease total costs. Figures 7.17 and
7.18 show an example of the optimal solution generated by both the VRP-TW-P and VRP-TWF models at the end of 10-day operation in Figure 7.16. In Figures 7.17 and 7.18 the horizontal
lines, which reach both ends of the graph, indicate the depot and the other horizontal lines
indicate the time windows of customers. These figures demonstrate that trucks tend to arrive at
customers earlier within the time window by using VRP-TW-P model rather than VRP-TW-F
model. For this particular case, as shown in Figures 7.1 7 and 7.18, the total delay time by VRPTW-P model was 99 minutes and that for VRP-TW-F model was 907 minutes, which is ten
times larger than VRP-TW-P model. The optimal operation in VRP-TW-P model in this case
used two trucks, whereas VRP-TW-F model used one truck. In this way, VRP-TW-P model
provides the routing and scheduling planning to avoid delays at customers.

(Vehicle Routing and Scheduling with ITS) 165

140
120

,/

E 100

8K
.+
m

.LO
U

80

60

40
/

20
0 - '

Figure 7.17 Example of optimal diagram of operating trucks (VRP-TW-F model, 10 ton truck)

140
120
h

E 100

8S

80

.+
m

.L"
U

60
40
20
0
5:OO

7:OO

9:OO

11:OO

13:OO

15:OO

17:OO

19:O

time

Figure 7.18 (a) Example of optimal diagram of operating trucks (VRP-TW-P model, the first
10 ton truck)

166 (CivLogistics)

20

,--.
E 15

a
0
8

:10

.-

-0

5
0

Figure 7.18 (b) Example of optimal diagram of operating trucks (VRP-TW-F model, the
second 10 ton truck)

Table 7.8 shows the total time after arriving at customers until the end of time window. This
total time of arrivals before the end of the time windows using VRP-TW-P model is 20.5%
longer than that using VRP-TW-F model. Thus, the VRP-TW-P model allows drivers to avoid
the delay at customers in case of increasing travel times due to unexpected traffic congestion.
This leads to the reduction of delay penalty and the total costs. The reason why the VRP-TW-P
model gives large periods between the arrival time at customers and the end of time windows is
attributed to that the delay penalty shown in Figure 7.14 is set much larger than the early arrival
penalty. The specific value of the delay penalty for 2 , 4 and 10-ton trucks was 5 times larger
than the early arrival penalty of 4 ton trucks. Therefore, trucks need to arrive earlier to avoid
the high delay penalties. On the other hand, when using VRP-TW-F model, a single value of
travel time is used instead of a distribution. No penalty is imposed on trucks unless they arrive
after a end of time window. Then trucks can arrive at a time very close to the end of time
windows.
Table 7.8 Total time arrived before end of time windows (10 freight carriers)

I Total time to spare (min.)

VRP-TW-F model I
Change

(%'o)

19,469
20.5

(Vehicle Routing and Scheduling with ITS) 167

Table 7.9 Total travel time

unit: minutelday

Table 7.9 shows the total travel time. The total travel time using VRP-TW-P model is slightly
larger than that using VRP-TW-F model. The actual running time both for passenger cars and
trucks does not increase so much, but the waiting time of trucks increased by 12-1570. This is
due to the tendency of the VRP-TW-P model to determine optimal routes where trucks arrive at
customers early to avoid delay penalties.
Table 7.10 shows the predicted change in CO2 emissions for both the VRP-TW-P and VRPTW-F models, based on the average running speed of trucks. The figure indicates that the C02
emissions by trucks decreased by approximately 6% using VRP-TW-P model compared with
VW-TW-F model, whereas the COz emissions by passenger cars remained almost the same
level for both models. The reduction of COz emissions by trucks was mainly due to 10-ton
trucks. As shown in Table 7.9 the travel time of 10-ton trucks decreased using VRP-TW-P
model, which contributed to reduce the total COz emissions by trucks, because the unit
emission rate of 10-ton trucks is larger than smaller trucks. This is the reason for the reduction
of C 0 2 emissions by trucks was achieved in spite of the increase of total travel time of trucks.

168 (Ciiy Logistics)

Table 7.10 C02 emissions

VRP
-TW-P
model

Change (%)
2 ton truck
4 ton truck
10 ton truck
Subtotal of trucks
Change (%)
Total
Change (%)

1.4
0
34
723
756
-6.5
24661
2466
-1.2

0.1
0
32
683
715
-6.4
20291
2029
-2.3

0.0
0
31
662
693
-6.3
1680
-2.7

The total COz emissions by passenger cars and trucks decreased by 1-3% using VRP-TW-P
model compared with VRP-TW-F model (Table 7.10). Therefore, incorporating the uncertainty
of travel times using VRP-TW-P model not only allows freight carriers to reduce their total
costs, but also improve the environment in terms of reducing C02 emissions.

7.3.4 Conclusions
This section presented an integrated model of probabilistic vehicle routing and scheduling for
urban pickup/delivery trucks and dynamic traffic simulation. The probabilistic vehicle routing
and scheduling model incorporates the uncertainty of travel times. After applying the model to
test road network the following conclusions were derived.
(a) The total cost for freight carriers decreased by 1 I-17% using V R P - W - P model compared
with VRP-TW-F model. The main reason for the cost reduction is that trucks tend to arrive
at customers earlier to avoid delay penalties using VRP-TW-P model. This leads to the
lower risk of delay when the travel time is increased due to unexpected traffic congestion.
The lower likelihood of delay provides better service to customers.
(b) Incorporating the uncertainty of travel times using VRP-TW-P model not only allows
freight carriers to reduce their total costs. but also improves the environment in terms of
reducing COz emissions by about 6% compared with VRP-TW-F model. This means that
the probabilistic approach in vehicle routing and scheduling can generate benefits to

av
residents and the community by improving the environment as well as to freight carriers by
cost reduction.
The section presented a case study of a probabilistic vehicle routing and scheduling model
using small test network. Further applications of this model to larger real scale networks is
required to fully validate this model.

7.4 DYNAMIC
VEHICLE ROUTINGAND SCHEDULING WITH REAL TIME
INFORMATION
7.4.1 Introduction
Dynamic vehicle routing occurs when the inputs required to solve the routing problem are not
known with complete certainty before vehicles are dispatched (Psaraftis, 1995). Here,
information changes whilst vehicles are distributing goods. Sequential updating of routes and
schedules occurs when new information is received. Policies prescribing how routes should
evolve as input is received in real time must be determined.
Dynamic vehicle routing and scheduling (VRS) is currently prevalent the courier industry as
well as in the distribution of fuel to retail outlets and households. However, as the availability
of real time information increases due to developments in information and communications
technologies other types of industries will use dynamic vehicle routing and scheduling systems.
Common types of real time information:
(a) System Performance
Travel times (congestion, incidents & breakdowns)
(i)
(ii)
Service times
(iii)
Waiting times
(b) Customer Demand

(i)
(ii)

Location
Time window

(iii)
(iv)

Amount of goods
Priority

(c) Vehicle

(i)

Location

170 (CiqLogistics)
(ii)

Load status

Static VRS procedures have been extensively researched in the past. However, there have
been limited investigations into procedures for dynamic VRS since there have been many
technological obstacles that have inhibited their implementation.
However, recent
developments in communications and information technology have made dynamic vehicle
routing and scheduling technically feasible. Increasing pressures placed on carriers by shippers
and receivers of goods within JIT distribution systems have made dynamic VRS more
attractive.
With the growing use of "Just-in-Time" transportation there are increased requirements on
carriers to increase performance as well as flexibility. Moreover, the rapid advances in related
technologies, such as computer systems, communication networks, Electronic Data
Interchange (EDI), Internet and Global Positioning Systems (GPS) have led to a renewed
interest in researching and implementing dynamic vehicle routing and scheduling systems.
Dynamic VRS is an information intensive decision-making process that integrates various data
from a variety of sources, including road or municipal authorities, freight carriers/drivers,
customers or retailers, etc.
Generally speaking, geographic and traffic information are the most important knowledge for
optimising urban distribution systems. However, the information required for a dynamic VRS
for commercial vehicle requires more manifold and heterogeneous information. Information
can be certain or uncertain (random or fuzzy), constant or variable (with respect to the period of
the operation or decision making). An analysis and classification of the various types of
information required for a dynamic VRS will increase the understanding of the nature of the
system as well guide the accommodation and processing of such information.

7.4.2 Classification of Variability

The variability of information in dynamic VRS is the major characteristic distinguishing a
dynamic system from its static counterpart. For dynamic VRS, most information is only
partially known in advance and may change during the time of operation.
Static information generally remains unchanged or changes negligibly during the VRS
decision-making process. Historical data are typical static information and they are very
valuable for dynamic VRS. They can be used together with predicted demands or conditions to
produce initial plans or even determine cruising areas for idle vehicles. Considerable
improvements in the rapidness of service response can be achieved in this way. Road network

(Vehicle Routing and Scheduling with ITS) 17 I

information is the largest amount of static information in such a system. It is stored and
manipulated within a GIS and provides fundamental spatial information for VRS.
For daily operations, dynamic information is collected for during the routing and scheduling
process. Some of this information is updated at fixed intervals, such as status of the traffic
system, while others may be received continually, such as customer requests. However, it is
still common practice for dynamic dispatching to be processed in batch mode. Another
category of dynamic information relates to the actual vehicle fleet, including each vehicles
location and un-used capacity. This provides the operational data for a dynamic VRS. For the
convenience of usage and maintenance, the dynamic information is generally processed using
real-time databases.
Certainty Classification
Dynamic VRS involves incorporating uncertain information within optimisation procedures.
This could be either random (probabilistic) or ambiguous (fuzzy) in nature. The classification
of such information will vary with the technology used for data acquisition as well as the type
of task it is being used for. For instance, if GPS is used for vehicle tracking, the fleet vehicle
location can be regarded as deterministic information. If however, the driver reports this
position, then it should be fuzzy. Customer location could be deterministic information if all or
some of the customers are fixed in advance. As for time windows, they can also be
deterministic or fuzzy, depending on specific customers requests.
Source Classification
Urban transportation systems are complex systems. Data from various sources need to be
integrated for a dynamic VRS. Fortunately, this is becoming easier with the development of
advanced technology and increasing role of the public sector in monitoring the performance of
urban traffic systems.

ITS infrastructure, typically provided by road authorities and traffic management centres will
provide historical or real-time traffic parameters for most dynamic transportation applications.
On the other hand, land information authorities often maintain basic road network information,
i.e. geometric features or address information. Besides these, other special information, e.g.
safety requirements or public activities may be obtained from local or regional authorities. With
the introduction of computer systems into commercial vehicle operations, many freight carriers
have established management information systems to record their daily operations. This
provides a useful starting point for dynamic VRS.

172 (CivLogistics)
Form Classification
Traditionally, input information has been in the form of spreadsheets, charts, standard database
or text files, while the output has been either map displays or text reports. Recently, the rapid
development and application of multimedia technology has led to novel forms of information
that VRS may utilise, including images, sounds, videos and animation. Using these forms of
information, human-machine interfaces can be more friendly and efficient. For example, video
images may be used to show the situation of the delivery/pick-up sites; whereas sound
recordings may be used for driver guidance or alarms. In this context, multimedia data can be
related with relevant GIS data designed for the above purposes.

7.4.3 A Framework for Information Integration

The following section describes a framework for a dynamic VRS developed for an urban
courier service (Wang et al., 1999).
Information Acquisition and Transmission
For static VRS, after task deployment, the fleet operates autonomously, i.e. without
interference from the central dispatcher. Here, traffic conditions are represented using historical
data that requires no real-time present traffic information. However, for dynamic VRS, this is
completely opposite. Operations are based heavily upon the real-time information of the fleet
and traffic that is updated frequently and data acquisition and transmission is of vital
importance.
Dynamic VRS relies heavily on the recent developments in communication and electronic
technology that includes from mobile communication, automatic vehicle location, vehicle
detection and the Internet. These were described in Chapter 3.
Database Federation
Since dynamic VRS deals with information from various sources, the formats of these different
data types are typically not the same. Hence, it is not sensible to build a homogeneous database
to accommodate all the related data. Therefore, a database federation mechanism will be more
suitable. Here, the databases are established and maintained separately according to appropriate
structures. For determining optimal routes and schedules, they are integrated together using
relational methods. After a relation is established, joined feature tables enable queries and
analysis.

(Vehicle Rourillg and Scheduliizg with ITS) 173

GIS can be used to store the road network data, including geometry, speed limits and
addresses (Section 3.4). This data is assumed to be static with regard to the operation process.
Due to storage space requirements, GIS servers are often used at the dispatching centre while
CD-ROMs are more popular for on-board computers.
Customer and fleet location databases contain a large amount of dynamic information that also
has a geographic dimension, typically a point feature. A database structure similar to that of
the road networks (GIS) database can be utilized for these two databases. These will form two
separate layers for GIS manipulation.
Dynamic traffic databases are generally maintained by and located within road authorities. It is
not good to directly merge these databases with the GIS database due to the complexity and
variability of the traffic data. This data should be interpreted and linked to the GIS database
before subsequent processing.
Historical databases, operational databases and auxiliary databases are generally established
and maintained by transport carriers as part of their management information system. They
usually contain data relating to daily operations, parameters of their fleet and depots,
customers, cost items, transport regulations and traffic conditions experienced.
System Function and Procedure
With GIS based dynamic VRS, operators can undertake routing analysis using actual (realtime) traffic conditions at various times, schedule requests from clients, reallocate last minute
orders, or monitor the location of their vehicles.
Delivery/pickup paths using different combinations of roads can be compared using GIs. The
cost for each path can be displayed in a report or in a chart for easy comparison. In addition,
depots and other facilities can be added to the cost analysis or displayed.

In short, for the management and monitoring platform of VRS, the following functions would
be facilitated:

(a) Address-matching/geocoding
(b) Path selection/ information
(c) Time and scheduling

(d) Load scheduling

(e) Dynamic reallocation (routing)
(0Report and map generation

174 (City Logistics)

(g) Management planning
(h) Cost estimation
Considering the heavy computation burden involved in dynamic VRS, a two-stage processing
and optimisation procedure is more feasible. The primary stage is based on known information
and initial plans are generated. The secondary stage considers incoming requests with the aid of
several qualitative reasoning methods.
Determination of the next customer is a constrained optimisation problem, involving a multiobjective function, including operating costs, customer satisfaction and vehicle capacity
constraints. Tabu search has been applied in a number of situations, including courier services
and express mail delivery (Gendreau and Potvin, 1998). Fuzzy logic and utility theory appear
promising approaches for developing efficient dynamic routing policies.
Dynamic VRS deals with complex information from a variety of different sources each with
different features. Information integration is the initial but essential step within a dynamic
VRS application. Proper transmission and processing of information between GIS and ITS
will provide a solid foundation for its practical implementation. This will require an
environment to be constructed where information can be linked, updated and integrated to
provide the decision support for enhancing the management of distribution services in urban
areas.

(Locatioiz of Logistics Temiiizals) 175

LOCATION
OF LOGISTICS
TERMINALS
8.1 OVERVIEW
The distribution of goods within urban areas using road-based vehicles has led to many
problems, such as traffic congestion, negative environmental impacts and high logistics costs.
This is largely due to the logistics systems that the actors involved in urban goods movement
(industrial firms, shippers and freight carriers) have individually established to meet consumer
needs. These systems may be useful for a company or a group of companies, but from a social
point of view, they have not always been effective for reducing total social costs and
environmental impacts. In order to address these problems, the re-establishment of urban
logistics systems could be an effective City Logistics initiative.
Goods movement involves several functions relating to nodes and links on a network. The
functions of links are transportation and pickup/delivery; node functions include storage,
deposit, handling, processing, assembling, packaging, wrapping and loading/unloading.
Logistics terminals are required to fulfil these node functions and to make both functions
interact with each other. Therefore, the location andor relocation of logistics terminals should
be incorporated into re-established urban logistics systems. Proposals have been made in Japan
by the public sector to construct logistics terminals in the vicinity of expressway interchanges
surrounding large cities. Such public logistics terminals are multi-company distribution centres
and also complex facilities with multiple functions involving advanced information systems,
which can facilitate the implementation of co-operative freight transport systems. These
terminals can also meet various needs in supply chain management systems. Similar ideas
relating to the planning of public logistics terminals have been proposed in the Netherlands
(Janssen and Oldenburger, 1991) and in Germany (Ruske, 1994). The concept of public
logistics terminals however, needs more intensive investigation in several areas, such as their
function, size, location, management as well as the role of public sector.
This chapter deals with models for determining the optimal location of such logistics terminals
from the point of view of City Logistics. The location of logistics terminals is also influenced
by their size. Therefore, this chapter also covers the modelling associated with the size of

176 ( C i v Logistics)

logistics terminals. A mathematical model for determining the optimal size of logistics
terminals, allowing the characteristics of truck arrival and loadinghnloading services to be
considered using queuing theory is presented in Section 8.2.
An outline of the traditional approaches for determining the optimal location of logistics
terminals is presented in Section 8.3. These have been mainly undertaken in the area of the
Facility location problem. Since the aim of constructing logistics terminals in terms of City
Logistics is to help establish more efficient urban logistics systems and reduce the total social
and environmental costs of urban freight transport, traffic conditions on the road network in a
study area play a key role to assess establishing such logistics systems. Although the focus is
also on modelling of the optimal location of logistics terminals, the model developed in Section
8.4 explicitly takes into account traffic conditions on the road network. Traffic assignment
techniques, which are usually applied within the field of Traffic Engineering, will be
incorporated into optimal location model of logistics terminals.
Considering the purpose of City Logistics schemes, the effect of constructing logistics
terminals should be evaluated using a variety of performance measures, such as transportation
costs, traffic congestion and environmental impacts. It would therefore be desirable for the
modelling of optimal location of logistics terminals to be simultaneously incorporated a variety
of objective functions. In section 8.5, the optimal location model of logistics terminals
presented in the previous section is extended to the multi-objective optimisation model.
The location models of logistics terminals described in sections 8.4 and 8.5 involve the traffic
user equilibrium assignment technique, and thus these models are highly theoretical. These
models may be applicable to practical problems but consideration should be given to the costs
of data collection and analysis when using these models. Simplified models, for example, using
the conventional incremental assignment method (e.g. Martin and Manheim, 1965) instead of
traffic user equilibrium assignment method can be used in practice to evaluate City Logistics
measures.

8.2 SIZE
Often regression models have been used to determine the size of logistics terminals within
urban areas. It must however be recognised that although these models may be good for
calculating the required size of logistics terminals of a company because they can easily
represent the relationship between the required facility size of a logistics terminal and the
amount of goods handled within it; they cannot represent the overall impact on urban logistics
systems, especially on traffic conditions on the road network. Thus, the behaviour of freight
vehicles within and outside logistics terminals should be incorporated into optimal size models

(Location of Logistics Termirzals) 177

of logistics terminals when modelling City Logistics schemes.
This section is devoted to the optimisation of logistics terminal size, represented by the number
of berths (Taniguchi et al., 1999b). Berths are established for trucks to load and unload goods
within logistics terminals and therefore relate more strongly to the traffic conditions outside
logistics terminals than other facilities. Since the optimal location model described in Section
8.4 was developed to allow traffic conditions on the road network to be taken into account, the
model for determining the optimal size presented in this section focuses on the number of
berths. The size of other facilities within the logistics terminals is also considered in this model,
with construction and management costs being incorporated into berth costs.
Optimisation of the number of berths must take into account the characteristics of truck arrivals
at berths and loading/unloading service ability provided by them so that the impact of
constructing logistics terminals on society at large can be considered. This can be represented
by queuing theory (e.g. Cox and Smith, 1961), using a trade-off between transportation costs
and facility costs within a logistics terminal. This approach is based on models that determine
the optimum number of berths at seaports (Noritake and Kimura, 1990).
In the case of logistics terminals, the inputs and decision variables required for determining the
optimum number of berths are as follows:
Inputs:
: berth cost per hour at logistics terminal (yen/hour/berth)

cb
T

: time

c,

: transportation cost per hour for each vehicle (yen/hour/vehicle)

period considered (hours)

Decision Variables:
y
: number of berths
Using the above inputs and decision variables, the following objective function can be
formulated for determining the optimum number of berths:

where,
C

: total cost at logistics terminal during period T (yen)

: total number of trucks using logistics terminal during period T (vehicle)

rz(),V) : average number of trucks in logistics terminal for the number of berths) during

period T (vehicle)

178 (CiQ Logistics)

If equation 8.1 is divided by c, T , it yields

where,
I'
: ratio of total costs and transportation cost per truck at logistics terminal for the number of
berths y during period T .

rbl : berth-truck cost ratio.

Given the value of c, T , the ratio r can be substituted as the evaluation criteria for
determining the optimum number of berths. In equation 8.2, the berth-truck cost ratio rbi can
be calculated by cost analysis performed in advance, and hence, if the number of berths y is
fixed, the ratio r is only a function of iz(y,V) representing the average number of trucks
within a logistics terminal. ( n ( y , V ) will be represented as n for its simplicity in the
remainder of this section.)
The average number of trucks in the logistics terminal is equal to the sum of the average
number of trucks in waiting and in service as given below.

where,
a : traffic intensity
it%

: average number of trucks waiting at the logistics terminal for berths if truck arrivals

follow a Poisson distribution and an exponential distribution is that for service times (i.e,,

M/M/S(m) in Kendall's notation (Kenda11,1953))

Note that in deriving equation 8.5:

where,
1 : arrival rate of trucks (vehicle/hour)
p : service rate for trucks (vehiclehour)

(Location of Logistics Terminals) 179

Here, both p and A can naturally be expressed as a function of V , total number of trucks
using a logistics terminal. nwJ can be derived from Little's formula (Little, 1961).
ItM,=k

w,

(8.5)

where,
W, : average waiting time of trucks in W

S (w) model (hours)

In the case of W S ( m ) model, the average waiting time of trucks W, is given below.

Substitute equation 8.6 into equation 8.5, then

(8.7)

Thus, the average number of trucks n can be obtained by substituting equation 8.7 into
equation 8.3.

Service times following an exponential distribution imply that loading/unloading times have no
regularity. Mechanisation or automation in loadinglunloading goods would increase the
regularity of service times, and Erlangian distribution would be more applicable to the
distribution of service times. Service times approach a fixed value as the degrees of freedom of
Erlangian distribution increase. However, in cases where truck arrivals follow a Poisson
distribution and service times have an Erlangian distribution with k degrees of freedom (i.e.,

M/Ek/S(w) in Kendall's notation), no theoretical formula has been derived concerning the
average number of trucks in a logistics terminal. Some approximation formulas have only been
proposed that relate the average waiting time of trucks in a M&/S (w) model to that in a
W
S (w) model. Cosmetatos (1976) proposed the approximation formula given below.

180 (City Logistics)

where,
W, : average waiting time of trucks in a M/Ek/S (-) model (hours)
1%

: coefficient of variation of trucks service time distribution

a
: =- ; utilization factor (lberth)
y

Note that MIMIS

(m)

is the special case of M/Ek/S

(w),

when k = l

Cosmetatos (1976) reported that any value of W, given by equation 8.9 could be
approximated with a relative percentage error less than +2% for most practical purposes. When
the service time of trucks obeys the Erlangian distribution with the degree of freedom k ,
equation 8.10 holds,

(8.10)

The following equation is obtained by substituting equation 8.10 into equation 8.9.

(8.11)

itwi,

average number of trucks waiting at a logistics terminal for berths in a M&/S (-) model,

can be formulated with equation 8.5 using Littles formula,

(8.12)

nu, =;W,
i

Thus, equation 8.13 is obtained by substituting equation 8.11 into 8.12.

(8.13)

Then, the average number of trucks in a logistics terminal in a M/Ek/S

obtained as given below.

(m)

model can be

(Locatioiz of Logistics Teinziizals) 18 1

(8.14)

Substituting equation 8.7 into equation 8.14, then yields:

(8.15)

If trucks don't arrive at random intervals but at regular intervals, truck arrivals cannot be
represented by a Poisson distribution. In this case, also no theoretical formula is available in
calculating the average number of trucks at logistics terminals. A simulation model based on
queuing theory should therefore be applied to determine the optimum number of berths.
The mathematical model described above is applied to an existing logistics terminal for
determining the optimum number of berths for pickup/delivery trucks and line-haul trucks,
respectively. Inputs required for application of this model were obtained by a survey at a
platform of a freight carrier within a logistics terminal. Table 8.1 shows the main
characteristics of truck arrivals and service times obtained from that survey.
The results show that the optimum number of berths for pickup/delivery trucks is 12, and 4 for
line-haul trucks. There were 30 berths each for pickup/delivery trucks and line-haul trucks in
the platform surveyed, but all the berths were not always used and most of them were used for
parking during off peak hours. Considering these situations, the estimated optimum number of
berths seems to provide reasonable solutions.

182 (CicLogistics)
Table 8.1

Surveyed characteristics of truck arrival and service

(at North Osaka Truck Terminal, in Japan)
Pickup/delivery trucks Line-haul trucks

Period T (hours)

15
(5am. - 8pm.)

24
(Oam. - 12pm.)

Arrival rate il
(vehiclello minutes)

3.08

0.91

0.31

0.37

9.94
Poisson
Exponential

2.46
Poisson
Erlang ( k =2)

Service rate p
(vehicle/] 0 minutes)
Traffic intensity a
Distribution of arrival times
Distribution of service times

8.3 LOCATION
Optimal location of logistics terminals can be considered within the framework of facility
location problems. Facility location problems have been studied in many fields such as
operations research, economies, mathematics, geography and computer science. Facility
location models (e.g. Hansen et al., 1987; Drezner, 1995; Daskin, 1995) involve determining
the location of one or more facilities in such a way that optimises a certain objective.
Facility location models have some common basic elements:
(a)
(b)
(c)
(d)
(e)

Space (defining the distribution of demand points and candidate sites of facilities)
Number of facilities to be located
Size of individual facilities
Number of existing facilities
Objective of decision maker

(0 Demand (distribution, actual demand, variation)

(g) Candidate sites for facilities (distribution, number)
(h) Facility user behaviour
(i) Capacity limits on facilities
These elements can be used to classify facility location models into three types: continuous
location models, network location models and discrete location models. With continuous
location models, any point on the plane is available for candidate sites; theoretically, there are
infinite number of candidate sites. Most continuous location models can be formulated as a

(Location of Logistics Teimiizals) 183

non-linear programming problem with the assumption that facility users move straight to the
facilities available. Network location models have an infinite number of candidate sites as well,
but these are only on nodes or links and facility users can only move within the network. This
model is more realistic because facility users are actually unable to move straight on the plane
but are able to move along links (the road network). Discrete location models consider a finite
number of candidate sites and determine the optimal location using the location of candidate
sites and predetermined costs incurred from demand nodes to the candidate sites.
Basic models of the facility location problems are those known as the p-median problem
(minisum problem) and p-centre problem (minimax problem); p represents the number of
facilities to be located. The p-median problem involves finding the location of p facilities such
that the total demand-weighted distance between each demand node and the nearest facility is
minimised; the p-centre problem involves finding it in a way that the maximum distance
between a demand node and the nearest facility is minimised.
Since candidate sites for constructing facilities are most likely to be only in a limited number
of areas for public sector infrastructure, a discrete location model with a given road network is
the most practical alternative for urban planning. For this reason, facility location models
relating to logistics terminals have been developed mainly within the framework of discrete
location models. Discrete location models are, in general, combinatorial optimisation problems
and NP-hard. Therefore, if these types of models incorporate more practical and complicated
conditions, every model requires its own solution algorithm based on heuristic techniques.
Discrete location models have been rapidly developed since mid-19th century, with the great
progress in computer technology and combinatorial optimisation techniques.
With discrete-type location models for logistics terminals, the optimal location is determined
from among a set of candidate sites given in advance within a study area. In general, optimal
location models for logistics terminals may be formulated using the following notation.
Inputs:
: number of candidate logistics terminal
i
err, : transportation costs from logistics terminal node i to demand node j
: facility costs at candidate logistics terminal node i
cb)
Decision Variables:
: amount of goods transported from logistics terminal node i to demand nodej
qii
: =1 if logistics terminal is located on candidate node i: = 0 otherwise
x,
With these definitions, an objective function for determining the optimal location of logistics
terminals may be formulated as follows:

184 (CivLogistics)

If till and cbl are constant, equation 8.16 will be a mixed integer programming problem, and
Branch-and-bound methods may provide an exact solution unless the number of candidate sites
is large. Many models, however, have incorporated more practical conditions into equation
8.16 to satisfy their objectives, and consequently heuristic techniques have to be adapted to
determine good solutions.
Optimisation problems relating to the location of transportation terminals have been modelled
together with the routing of goods (Hall, 1987; Daganzo, 1996). Campbell (1990) developed a
continuous approximation model for relocating terminals to serve expanding demand. Noritake
and Kimura (1990) developed models to identify the optimal size and location of seaports
using separable programming techniques.
Many useful findings have been provided by these conventional models in efficiently obtaining
optimal location and clarifying the basic relationships between the optimal location and road
networks. However, these models have not taken into account the change in traffic flow caused
by constructing logistics terminals, and hence traffic conditions on the road network and
environmental impacts cannot be incorporated in these models. In the case of constructing
logistics terminals from the point of view of City Logistics, it must be considered that logistics
terminals are only one part of urban road transportation systems. Therefore, location modelling
combined with traffic assignment techniques provides an effective tool for evaluating City
Logistics measures.

8.4 LOCATION
WITH TRAFFIC
ASSIGNMENT
8.4.1 Framework
This section presents a model for determining the optimal location of logistics terminals,
explicitly taking into account traffic conditions on the road network (Taniguchi et al., 1999b).
Figure 8.1 shows the structure of the logistics systems investigated. Goods movement is
assumed to be divided into two parts: line-haul, where long-distance transport is made by large
trucks on expressways, and local pick-up/delivery, where short-distance transport is made by
small trucks on urban streets. Logistics terminals are the connection points between line-hauls
and local pick-up/deliveries, with transhipments being performed there. The line-haul trucks
are assumed to travel only on expressways and the local pick-up/delivery trucks only on urban
streets. Consequently, the routes available for each type of truck will not be overlap with one
another before or after transhipment in the logistics terminal. This implies that the road

(Location of Logistics Teimiizals) 185

network is also divided into two parts. It will therefore be more appropriate to apply the
location model presented in this study to the logistics systems where routes available are
evidently different before and after transhipment, such as underground freight transport
systems.

Line-haul trucks
'Centroid'
for line-

++

'

Urban - - - - - - - - - - area
Centroid for pickup/
i/ delivery trucks
Interchange

L-------------

,T-,

Figure 8.1

Expressway

Urban street

Structure of logistics systems investigated

8.4.2 Mathematical Formulation

The location model presented in this section has following six features:
(a) The optimal location of logistics terminals is determined from candidate nodes that are
discretely given in advance within the road network.
(b) The optimal size of each logistics terminal is simultaneously determined taking into
account transportation costs and facility costs (such as construction, maintenance, land and
truck operation costs in the logistics terminal).
(c) A planner can determine the optimal size and location of the logistics terminals but cannot
control the distribution and assignment of truck traffic.
(d) Distribution of goods movement is given in advance for each pair of centroids for
line-haul trucks and pick-up/delivery trucks.
(e) Each truck can choose a logistics terminal depending on the traffic conditions on urban
streets.
(f) Inventory levels within logistics terminals are not considered.

186 (City Logistics)

Traffic condition
at each link

Behaviour of each company

and each truck
.Route choice

Figure 8.2 Structure of optimal location model

Figure 8.2 indicates the structure of the optimal location model for logistics terminals, which
has two levels of problems. The behaviour of each company or each truck in choosing the
logistics terminal used and route travelled are described in the lower level problem, which also
involves route choice behaviour of passenger car traffic. Both traffic modes satisfy the user
equilibrium condition on the road network. This problem corresponds to the doubly
constrained combined distribution-assignment model (Evans, 1976), which incorporates the
equal travel time principle for assignment and variable demand for distribution between a
centroid for pickupldelivery trucks and that for a logistics terminal.
The upper level problem describes the behaviour of a planner for minimising the value of an
objective function. In this model. the transportation costs of line-haul trucks and
pickup/delivery trucks, costs of travel time of all vehicles (i.e. passenger cars, line-haul and
pickup/delivery trucks) and COz emissions of all vehicles are included within the objective
function. Optimal size of each logistics terminal, represented by the number of berths, is also
included in the upper problem (See Section 8.2).
When transportation costs are adapted as an objective function, the upper level problem is
mathematically formulated using the following inputs and decision variables.

(Location of Logistics Terminals) 187

Upper level problem

Inputs:
i
: number of candidate logistics terminal
m,m' : type of vehicle
( m =l:pickup/delivery, m =2:line-haul, m ~3:passengercar)

cb,

(rn'&{1,2), r n f r n ' )
: berth cost per hour at candidate logistics terminal i (yenhourherth)

: time period considered (hours)

c,"

: transportation cost per hour for each vehicle (yenhourlvehicle)

a m : loading fator of each vehicle for m =1 or 2 (todvehicle)

Decision Variables:
: vector of location patterns of candidate logistics terminals
x
: vector of the number of berths of candidate nodes
y

(8.17)

subject to
(8.18)
Vm,i

(8.19)

Vi

(8.20)

where,

X, Y

: feasible sets of vector x and y , respectively

TC, TC' : total transportation costs with and without logistics terminals, respectively (yen)

Y,"

: number of berths at logistics terminal i for m =1 or 2

C,m

: total costs at logistics terminal i for m =1 or 2 (yen)

V,"(x) : total number of vehicles using logistics terminal i for m=l or 2 under a location
pattern x (vehiclelday)
n(f,V2")

: average number of trucks in logistics terminal i with the number of

berths yzkduring the period T for

= 1 or 2 (vehicle/day)

188 (City Logistics)

qz (x),q;;(x) : OD (Origin-Destination) traffic flow between logistics terminal

i and

centroids ( 0 or d) for m=l or 2 under a location pattern x (vehicle/day)

The objective function, shown in equation 8.17, measures the construction effect against the
construction costs of logistics terminals.
Equation 8. I8 is a constraint concerning the determination of the size of logistics terminal. The
number of berth
which minimises equation 8.18 is the optimal size of logistics terminal
(See Section 8.2). Equation 8.19 represents the required relationship between vehicles using
the logistics terminal and OD flow patterns. Equation 8.20 is a constraint that relates the
amount of goods flowing into the logistics terminal to that flowing out of it.
b','"(x),the total number of vehicles using logistics terminal i , as well as the OD flow qE(x)

and qL;(x), depends on the traffic conditions on a road network. Consequently, they can be
obtained from the lower level problem.
The numerator of equation 8.17, i.e. TC - T C ' , can be calculated by the following equation :

(8.21)

where,
\To

( x ) : traffic flow on link a on a road network under location pattern x (vehicleslday)

~ ~ ' " ( x: ) traffic flow of vehicle m on road network link a under location pattern x

(vehicles/day)
ro(Vo ( x ) ): travel time on link a on road network (minutes)

The lower level problem provides the traffic flow and travel times on the road network links.
Travel time on a load network link can be obtained using a relational expression called link
performance function, representing the travel time and traffic volume relationship. Here, the
following equation developed by Bureau of Public Roads (BPR) of the United States
(Steenbrink, 1974) is used as a link performance function.

(8.22)

where.

(Location of Logistics Terminals) 189

t,,,, : free travel time on link a (minute)

K , : traffic capacity of link a (vehicledday)

This equation is often used in practice, especially when traffic conditions are described with
user equilibrium traffic assignment technique. Free travel times on road network links should
be given in advance. In designing logistics terminals in terms of City Logistics, evaluation
should be undertaken for environmental impacts and traffic congestion other than
transportation costs. Equation 8.23 and 8.24 represent the objective functions for the costs of
travel time and CO1 emissions, respectively.

where.

TTC,TTC': total costs of travel time with and without logistics terminals, respectively (yen)
TCO,TCO' : total COz emissions with and without logistics terminals, respectively (yen) (g-C)
The COz emissions are the typical index representing the area-wide environmental load, but in
practice, environmental impacts other than COz emissions should be taken into account.
The numerator of equations 8.23 and 8.24 can be represented as the following equation.

(8.25)
(8.26)

where,
d "' : unit vehicle-kilometer coefficient of COz (g-C/vehicle/km)
V ,

: travel speed on link a (kndhour)

L,, : distance of link a (km)

The value in the numerator of the above objective functions corresponds to the benefit. Each of
them therefore represents the Benefit Cost Ratio for each evaluation index (i.e. transportation

190 (City Logistics)

costs, costs of travel time and the COl emissions). These objective functions require the same
constraints as equation 8.17 (i.e. equation 8.18, 8.19 and 8.20). Given a road network
investigated, the distance of road network links can be identified easily, but the estimation of
travel speed on these road network links needs the traffic flow estimates provided by the traffic
assignment procedure. This is incorporated into the lower level problem in the model.
The lower level problem simultaneously deals with passenger car traffic and pickupldelivery
truck traffic with both traffic modes satisfying the user equilibrium condition in the network. A
unique solution for this lower level problem is assured since the set of feasible solutions is
convex and the objective function is strictly convex (Sheffi, 1985). With some inputs and
decision variables, the lower level problem can be formulated as follows (omitting subscript x
for simplicity).
Lower level problem
Inputs:
Oo : generation at centroid o of trucks (vehicleslday)

D, : attraction at centroid d of trucks (vehicleslday)

Decision Variables:
q;, : traffic flow of OD pair
I?

(0,d

) for each vehicle, where either o or d corresponds to

logistics terminal i (vehicleslday)

: traffic flow for vehicle m on link a (vehicleslday)

(8.27)

subject to

f r:od

= 6 ,f

42 =

rlnod

cf

r'nod

Vr,od m =1

(8.28)

Vod

(8.29)

Va

(8.30)

Vo m = l

(8.31)

Vd m=1

(8.32)

Voi ,Vid ,Vod

(8.33)

Vr,od

(8.34)

(Location of Logistics Tenniizals) 19 1

where,

W~(~L)

: inverse of demand function for flow of trucks, where either o or d corresponds to

logistics terminal i
: path flow at path r of OD pair

fr;'

( 0 ,d

) for each vehicle, where either o or

d corresponds to logistics terminal i if m =1 (vehicledday)

: =1 if logistics terminal is located on candidate node

6,
6

,::

0 otherwise
: = 1 if path r of 0 - D pair

( 0 ,d

i under a location pattern x ; =

) passes link a ; = 0 otherwise

The inverse of the demand function for truck flows shown in equation 8.27 incorporates only
pickup/delivery trucks so that the model can be calculated with shorter computation times. In
this model, the travel behaviour of line-haul trucks is represented by the all-or-nothing traffic
assignment technique (e.g. Matsoukis, 1986).
Equation 8.29 preserves the OD traffic flow condition
relationship between link flow Vo and path flow

&.

Equation 8.30 represents the

f,yd. Traffic volume is given in advance for

pickup/delivery truck generation and attraction, and therefore Equation 8.3 1 and 8.32 indicates
that the lower level problem corresponds to the doubly constrained combined
distributiodassignment model.
W',(p) in equation 8.27 is the inverse of the demand function for 0 - D flow and assumed to be

as follows:

(8.35)
where,

y : parameter

rod : travel time at OD pair (0, d ) (hours)

y is the traffic resistance against OD travel time for freight vehicles. It can also be considered
a parameter representing the elasticity of freight vehicle demand for OD travel time. As
y increases, freight vehicles show a stronger tendency to choose a logistics terminal used
where shorter travel times are required from their origin.

192 (Cir?,Logistics)
Substituting equation 8.35 into equation 8.27, then

(8.36)

If there were no newly constructed logistics terminals, no pickupldelivery trucks could use
them. In this case, consequently, line-haul trucks would transport goods between interchanges
and urban centroids. This implies that the model deals with ordinary logistics systems:
line-haul tracks go into urban centroids through interchanges on expressways and tranship
goods to pickupldelivery trucks at depots within centroids, and then pickup/delivery trucks
transport goods to their customers within centroids. In this logistics system, namely in the
model without logistics terminals, traffic assignment is undertaken for passenger cars and
line-haul trucks.

8.4.3 Solving by Genetic Algorithms

The model described here corresponds to a large-scale non-linear programming problem and
there is much difficulty in obtaining an exact solution. Therefore, Genetic Algorithms (for
example, Goldberg, 1989; See Section 2.6) are applied as a solution procedure.
Figure 8.3 shows the calculation steps of the GA applied in this study. Step 1 assumes the
genotype of a chromosome with genes used to display an individual corresponding to the
location pattern of logistics terminals. In this case, if a gene is represented by 1, it means that a
logistics terminal is constructed at this location; if it is represented by 0, it means no logistics
terminal is constructed there. In this calculation procedure, the evaluation value is defined as
the value of objective function, shown in equation 8.17 (or 8.23, 8.24). The fitness of the
individual is obtained by making the evaluation value linear normalized. Step 2 generates the
first generation of certain numbers of individuals, and in Step 3, the fitness of each is
calculated. Step 4 performs reproduction where individuals are probabilistically selected and
multiplied based on the value of fitness and the elite individuals with high scores of fitness are
preserved to accelerate the calculation. In Step 5 , pairs are made from individuals and the
crossover processing is executed for them to search for a better solution. Step 6 performs
mutation to avoid being caught in a local optimal solution. Some parameters shown in Figure
8.3 were determined from trial-and-error procedures for small optimisation problems with
known their exact solutions.

--

(Location of Logistics Terminals) 193

Chromosome

(0 1 0 1 0 0 0 1 0 0

STEP 1
Initial setup
(Determination of genotype)

Example for location pattern of

terminals (2,4,8)

Number of individuals: 50

STEP 2
Generation of first population

Fitness obtained by linear

normalization of the value of
objective function

STEP 3
Calculation of fitness of each
individual

Number of elites: 10
STEP I
Performance of reproduction

STEP 5
Performance of crossover

STEP 6
Performance of mutation

STEP7
If number of generations

Uniform crossover procedure

Crossover rate: 1.0

Mutation rate: 0.05

NO -

Number of generations: 30

end

Figure 8.3 C a l c u l a t i o n s t e p s of t h e g e n e t i c a l g o r i t h m

194 (City Logistics)

The lower level problem has a similar formulation to the Evans model (Evans, 1976), and
hence its solution can be obtained using an efficient algorithm proposed by Evans. This
algorithm is composed of the same iterative procedure as Frank-Wolfe algorithm (e.g.,
Bruynooghe et al., 1968, LeBlanc ef al., 1975), which involves an auxiliary problem to search
the descent direction and a line search process towards this direction. The solution algorithm of
the lower level problem is described as follows.
Step 1 Initialisation
Select an initial feasible set of current traffic flows:

{ V:,()];

{ q;/) , { qi$i} , { V;.)}and

make the number of iterations N = 1. Vb and V:are

flow on link u for

pickup/delivery trucks and passenger cars, respectively. With these variables, the flow on
link a can be represented as:

v,

=Val

+ v,

(8.37)

Step 2 Shortest path search

With the current link costs t , ( V p ) ) ,build the set of minimum cost trees: search for the
shortest (cheapest) path for all OD pair and calculate the path costs t:! and

ty.(using

a shortest path search method, e.g., Gallo and Pallottino, 1986.)

Step 3 Direction search
Solve the following auxiliary problem that can be obtained by the linear approximation of
equation (8.36) in order to search for the descend direction:

(8.38)
subject to
(8.39)
(8.40)
(8.41)

: vector of

zld

(Location of Logistics Terminals) 195

zb, : auxiliary variable for pickup/delivery truck flow of OD pair

( 0 ,d

), where either o or

d corresponds to logistics terminal i (vehicleslday)

t;, :

travel time on shortest path at N-th iteration for OD pair

(0,

d ), where either o or d

corresponds to logistics terminal i (hours)

Load the solutions of above equation, i.e. { Z:;}
obtaining

u:

1=

link

1.W i +

flow

{ U;^}

and

and

{U:,ihl} .

{ Z;iN!},onto the

Then,

calculate

shortest paths

{Ur!}

with

u;.iN1

Step 4 Line search

Solve the following equations using the Golden Section search method to obtain the value

of parameter A .
(8.42)

subject to
(8.43)
(8.44)
(8.45)
(8.46)

Step 5 Current traffic flows calculation

(8.47)
(8.48)
(8.49)
(8.50)
(8.51)

Step 6 Converpence assessment

If current link flows have not changed significantly between two successive iterations, stop the
calculation; otherwise N + N + 1 and return to Step 2.

196 (City Logistics)

In summary, the overall solution procedure of this model is illustrated in Figure 8.4. Traffic
assignment techniques are not limited to the user equilibrium approach in this model. Other
conventional traffic assignment approaches are available. This model can also incorporate
other heuristic techniques than GA for determining the optimal location, such as Simulated
Annealing or Tabu Search (See Section 2.6).

8.4.4 Application to Road Network in the Kyoto-Osaka Area

The model described in this section was applied to an actual road network in the Kyoto-Osaka
area in Japan, as shown in Figure 8.5. This network is planned for the year of 2010 and has 16
candidate sites for public logistics terminals along with several new expressways. The network
has two centroids for line-haul trucks in East and West Japan and 36 centroids for pickup/delivery trucks and passenger cars. For passenger cars, 6 nodes outside the study area are
also included in the network. In Figure 8.5, ordinary road links represent national highways
and main local roads. The land price is high for candidate nodes that are close to Osaka and
Kyoto, which are large cities in this area. This means that the construction costs of logistics
terminals are higher at these nodes. The predicted distribution of traffic volumes for passenger
cars and freight traffic in the year of 2010 and the present amount of goods are used as inputs.
The unit transportation cost was set for each vehicle based on statistical data from that area:
c:= 2254,c:= 5055,c:= 4470. The load factor a mis determined with and without logistics
terminals separately. In the case without logistics terminal, this parameter is set at a' = 0.3 and
a*= 0.8 based on the statistics in Japan; in the case with logistics terminal, this parameter is

assumed to be set at a' = 0.65 and a' = 0.8, since the construction of logistics terminals has a
high potential to facilitate the implementation of co-operative freight transport systems
(Yamada et al., 1999). The unit berth cost chi relates strongly to land price at logistics
terminali , As a result of regression analysis using the data assocated with existing logistics
terminals in that area, the relationship between the number of berths and area required can be
represented as follows:
R, = 390 y ,

where,
R, : area required at logistics terminal i (m2)

+ 1137

(8.52)

(Location of Logistics Teintirtals) 191

Initial location patterns of logistics terminals

.........................................................................................................................
*
*

Generation and attraction

traffic volume of PD
Distribute traffic volume
of PC

Traffic
assignment

*
*

use logistics terminals

/=\I,

1
I

Traffic volume of PD and PC on each link

Travel time of PD and PC on each link
Travel speed of PD and PC on each link
Distribute traffic volume of PD

'

logistics terminals

constructing
logistics terminals

Road network
factors

T~~~~~~~

. Total travel time

Total distance travelled

..................................................................................

..................................

Fimess of each location pattern

Simulation using GA

C0,ernissions

travel time

Renewed location patterns of

logistics terminals

PD : Pickup/delivery trucks
LH : Line-haul trucks

NO

PC : Passenger cars

Optimal location

Figure 8.4 Model calculation process

198 (Ciol Logistics)

>

Expressway

Figure 8.5 Study area (Road network in the Kyoto-Osaka area in Japan)

Therefore, chi can be obtained from the following equation.

cb, =1881+

.I
( 3 9 0 . ~+1137),p,
:,
365.24. y ,

where.
p , : land price of logistics terminal i (yen/m2)
I

: interest rate; = 0.05 in this example

(8.53)

(Location of Logistics Termirzals) 199

The first term in the right side of equation 8.53 is a constant that was calculated from the costs
for construction, maintenance and equipment. In this case, this parameter is set to be 1881
(yenhourberth) based on the interview survey to existing freight carriers in this area.
Table 8.2 indicates the optimal solutions for each objective function with period in
consideration T = 24. This result shows that the optimal solution does not always agree with
each other for these objective functions though all of them may be influenced by the traffic
conditions on the road network. Every location pattern shown in Table 8.2 includes terminal 5.
This is because terminal 5 is near large cities that have a large demand for goods movement
while its land price is relatively low.

Table 8.2 Optimal location patterns for each objective function

f3 : COz

emissions

8.5 MULTI-OBJECTIVE
OPTIMISATION
The model presented in this section takes multiple objective functions into consideration using
a multi-objective programming method, and consequently alternative location patterns of
logistics terminals can be obtained. Transportation costs of line-haul trucks and pickup/delivery
trucks, costs of travel time of all vehicles (i.e. passenger cars, line-haul and pickup/delivery
trucks) and C 0 2 emissions of all vehicles are treated as objective functions. The values of these
objective functions will not only differ from each other but also be mutually conflicting as
shown in the previous section (See Table 8.2). A mathematical model therefore, needs to be
developed incorporating the multi-objective programming method.
Multi-objective programming methods (e.g. Chancong and Haimes, 1983; Sawaragi et
a1.,1985) can be generally applied to problems with mutually conflicting objective functions.
All objective functions cannot be simultaneously minimised in case when they are mutually
conflicting, and hence Pareto optimal solutions (i.e. non-inferior or non-dominated solutions)
are defined as optimal solutions for multi-objective optimisation problem. Pareto optimal
solutions are represented by x* when there is not any x (xEX) which satisfies fi(x) <fi(x*j at i =
1,...,k for X*EXand h(x) < f{x*) at arbitrary J . They are used if the value of at least one
objective function must be increased in order to decrease the value of a certain objective

200 (City Logistics)

function. In general, multi-objective programming method involves minimising k objective
functions being mutually conflicting under rn inequality constraints. This can be formulated as
follows:

subject to

x E x=

R"1 g(x) 5 01

(8.55)

where, g(x) is m-dimensional vector constraint function and X represents the feasible region
for x. The objective functions shown in equation 8.17, 8.23 and 8.24 are described as
maximisation problems, which can be easily transformed into minimisation problems as fitted
with equation 8.54.
The upper level problem presented in the previous section applies multi-objective
programming method to obtain optimal location patterns for logistics terminals. This problem
has discrete variables representing location pattern of logistics terminals. It requires a very long
computation time to obtain exact Pareto optimal solutions and if there are many candidate
logistics terminals (in practice it is impossible). Vector evaluated genetic algorithms (e.g.
Schaffer, 1985) have been therefore adapted in this model. Vector evaluated genetic algorithms
(VEGA) provide an effective method to quickly obtain approximate Pareto optimal solutions.
Figure 8.6 indicates the calculation steps of the VEGA. VEGA have the same genetic operators
as GA (See Figure 8.4), but the reproduction process in VEGA is performed based on the value
of each objective function (Step 4). Individuals are divided by the number of objective
functions, and subgroups of the population are generated. These subgroups are mixed after
reproduction (Step 5), and then crossover and mutation are performed for the reconstructed
population (Step 6 and 7). VEGA make use of the G A S characteristics that a number of
individuals search for multiple points at the same time and enable obtaining Pareto optimal
solutions at a time.

(Location of Logistics Teiniiizals) 201

START

Individual

( 0 1 0 1 0 0 0 1 0 0

TfTtTTTTTT

Step1
Initial setup

Example for location pattern of

terminals (2.4.8)

Step2
Genaration o f first poptilation

Number of individuals: 70
Fitness obtained by linear
normalization of l/(the value of
obiective function)

Calculation of fitness for each individual

Step4
Perfonnance of reproduction
Subgroup

Subgroup

Subgroup

Number of elites: 10

(based on the kalue of

ohjectne function L)

Step5
Mixture of subgroups
Step6
Perfonnance of cross over

Uniform crossover
Crossover rate: 1 .O

Mutation rate: 0.05

Step8
If number of generations
Number of generations: 50

Figure 8.6 Calculation procedure of VEGA

This VEGA-based optimal location model is applied to the same case study as used in the
previous section. Accordingly, the same road network and inputs are utilised. As shown in
Table 8.3, this model can successfully be applied. and the Pareto optimal solutions representing
alternative location patterns of logistics terminals can be obtained. These optimal solutions
agree precisely with exact Pareto optimal solutions obtained by using complete enumeration
method. Table 8.3 includes the compromise solutions as well as the optimal solution for each
objective function.

202 (CityLogistics)
The value in the numerator (See equation 8.17 , 8.23 and 8.24) was maximised for all objective
functions when the location pattern was composed of four nodes ( I , 2, 5 , 15). However, the
optimal location pattern for each objective function differs from each other. This result can be
explained by the difference in benefit and that in construction costs of logistics terminals
among the optimal location patterns shown in Table 8.3. As for transportation costs and COz
emissions, the difference in benefit among them was relatively small. Therefore, the
construction costs of logistics terminals had a large impact on the value of objective function.
On the other hand, as for travel time costs, the value of objective function was more strongly
affected by the difference in benefits than that in construction costs that was much larger than
the benefit.
It is desirable when planning the location of public logistics terminals that planners determine
the alternative location patterns of terminals in advance and then select an optimal (or
preferred) one from this set by considering the ease of land purchase and the needs of users (i.e
shippers and freight carriers). The multi-objective optimisation location model will be an
effective tool in such a situation. Further investigation will however be required to obtain a
preferred location pattern from alternative ones.

Table 8.3 Optimal location patterns

*maximum value of each objective function

(Future Perspecti1,es) 203

FUTURE
PERSPECTIVES
9.1 SYSTEMS VIEW AND FUTURE
DEVELOPMENTS
Economically well-developed countries face the problem of how to stop the increase in
mobility (Bannister, 1994). Increasing mobility generates economic development, welfare and
social freedom. On the other hand, the growing demand of mobility also causes negative effects,
which become more and more visible in terms of congestion, emission, noise, energy use,
space use, accessibility problems and social segmentation. Day by day the development of
adequate transport systems becomes more complex. There are two types of explanations for
this complexity of transport systems (van der Heijden, 1995). The first type of explanation can
be found from the system-internal complexity. More passengers and freight has to be
transported, more regulations have to be complied (e.g. with respect to engine emissions and
maximum loads), more standards have to be met (e.g. with respect to transport safety), and the
costs of infrastructure are huge. The system-external complexity can be explained by the rapid
changes in cultural, societal and political conditions. More and more citizens and societal
groups become more critically involved in discussions on public goals and decision making on
transport strategies. Therefore, we need to broaden our systems view (van der Heijden, 1997)
on transportation systems if we will orientate ourselves on future perspectives in transportation.
Basic within this view is that the transport system can be seen as a system, which consists of
four components of transportation: the potential users, potential subjects of transport (load units
& cargo), vehicles of the transport mode and the infrastructure. Furthermore, between each of
these components interactions exist which can be regarded as a market. i.e. the dynamic
interaction between potential supply and demand. In particular the following markets are
specified according to this view: the transport need market, the transport market and the traffic
market.

204 (CityLogistics)

I
i
Economic, Social, Spatial Organization of Society

Transport need market

t
Passengers and Freight

4
Transport market

Transport Modes

4
Traffic market

Infrastructure

Figure 9.1 A Transportation Systems View (Van der Heijden, 1997)

The transport need marker is influenced by the spatial, social-cultural, economic, temporal and
institutional organization of society. This societal organization generates a certain need for
transport: for instance a fast growing economy generates a relatively high transport growth rate.
Reversely, differences between regions with respect to possibilities for transport influence the
societal organization in these regions and cause different economic growth rates between these
regions. Hence, there exists a dynamic tension between the hidden need for transport and the
actual transport possibilities. This tension is influential by policy measures related to the price
for transport (e.g. fuel taxes). household time budgets (e.g. opening hours of shops and flexible
working

hours)

and

substitution

possibilities

(e.g.

tele-learning,

tele-working and

tele-shopping).
The transpol? senice marker focuses on the logistical organisation of physical transport.
Transport companies (or individuals) specify logistical services in terms of the pursued use of
certain transport modes, with a certain capacity and price, at specific times and routes. in order
to move passengers or freight. The market operates in the context of the tension between
transport means and actual transport needs, which can also be influenced by a variety of policy
measures. For instance, many policy measures aim at establishing a substantial modal shift

(Future Perspecti1,es) 205

from car and truck use to the use of more environmental friendly transport modes like trains
and boats.
Finally, the use of specific transport services results in traffic. The traffic market operates as the
dynamic relationship between the physical transport infrastructure and the required use of
various transport means. The result is certain amount of traffic in specific parts of specific
transport networks at specific time period. The tension in the traffic market is created by the
limited availability of transport infrastructure capacity as compared to the required use of it.
Evidently, this tension can be influenced by investments in infrastructure, capacity or road
pricing.
A characteristics of this systems view is that lower levels facilitate (offer services to) the
processes at a higher level of the system. Conversely, processes at a higher level specify
functional requirements to the processes and services at lower levels. This implies that options
for improvement of the transport systems performance cannot be evaluated by merely looking
at the impacts at one systems level. Instead, a multi-level, multi-dimensional and by
consequence also multi-disciplinary approach is required. Another interesting point of this
systems view is the introduction of the idea of different markets. Within these markets. the
interacting behaviour of a variety of parties is of influence.
To describe the future developments in transportation we will primary focus on the specific
developments in the different markets. The most problematic market is the truflc nzarket.
Therefore, it is not surprisingly that most of the future developments are oriented within this
market. The directions of these developments are described below.
(a) Accommodating current traffic flows and trying to find solutions which can improve the
efficiency within this market

In this context it should be mentioned that there have been substantial developments in the
Intelligent Transport Systems (see Chapter 3). Dynamic traffic information systems allow
drivers to communicate with a control centre and gain real time information on traffic
conditions. Studies on City Logistics show that the impact of ITS development with real
time data will be very beneficial for the improvements of just-in-time deliveries (Taniguchi.
et ul., 1 9 9 9 ~ ) .
As an extended form of ITS automated vehicle guidance (AVG) systems should be
mentioned here. In general, AVG is known as; partly or totally taking over of the vehicle
driving task by information and communication technology (Marchau, 2000). Although, in
general, AVG intends to improve traffic efficiency, there are studies that forecast a decrease

206 (Civ Logistics)

in capacity for certain penetration rates of some systems (Van Arem et al, 1996). Although,
in the long run (2005-2020),the introduction of systems is expected which take temporary
control of the vehicle in case of dangerous situations with more general obstacles and/or
assist the driver in case of impeding road departure, it is anything but certain that these
systems further evolve (Marchau, 2000). Furthermore, these systems are initially expected
to be marketed for specific groups and to be used only on motorways. Further evolution in
terms of the likely majority of vehicles in the driving population using AVG systems on
secondary roads seems far away. While these roads are the main roads to be taken by the
logistics providers in the field of City Logistics this development will not have a severe
impact.
(b) Measures of regulation
Still direct measures of regulations are most effective in forcing logistics service providers
to rethink distribution networks and search for new alternatives. Measures for weight limits,
load factor controls, time windows, designated routes and zoning are most commonly
applied in many cities. Although most of the measures are mainly opposed on this market
their final effects can be measured in the other markets of our system view.
(c) Finding alternative transport systems using different infrastructures

In this field of alternative transport systems intermodal transportation systems seems to be

very prosperous to develop. Many congested areas will be intermodal opened by the
building of new waterways, railways and inland terminals. These systems will remain the
flexibility of truck transport while the final pick-up and delivery will be by truck. Due to
the loading capacities of the barges and trains the price of these transports can compete with
truck tariffs if the distances are greater than 100 kilometres. The modelling for intemodal
transportation will be discussed in Section 9.2.
A technology that can play a big role in future logistics is underground transportation.
Underground transportation is already common technology in passenger transportation
systems such as the subway. In freight transportation systems the industrial application of this
technology is mainly restricted to transport by pipelines for some chemical products, oils etc.
Looking at the plans for underground transportation projects the circumstances under which
this technology is considered as an opportunity for improvement are relatively similar to the
problem environment of city distribution. Since the costs of building for underground tubes
are very high, we can derive that some conditions in the problem environment that must be
met before an underground distribution system might be suited for environmental
improvement:

(Future Perspectiiles) 207

a) large transport volumes must be accommodated for

b) the travelling distance must be relatively short
c) the surface area is very congested and the transportation causes a lot of environmental
nuisances and
dj the products to be transported represent high-value items.
Especially in the field of City Logistics these conditions can be met. The modelling
approaches of these systems are discussed in detail in Section 9.3.
In the transport market we can observe a trend towards more dedicated transport systems. On
the one hand the vehicles that are allowed to enter the inner city should have higher load factors.
In order to reach this goal logistics services will introduce segmented vehicles in their
distributions systems for final delivery to the shops. A big Dutch grocery has introduced with
great success the segmented truck to supply their shops. A reduction of 70% of transport
movements could be realised (Willemse, 1999). On the other hand for intermediate deliveries
between cities and distribution centres the importance of just-in-time deliveries become more
important as these shipments become an integral part of floating-stock-concepts. As a part of
hub-and-spokes-networks these connections should accommodate large volumes in order to
reach economies of scale in transportation. Therefore, the size of vehicles will increase, for
example there are now vessels with a loading capacity of 460 TEUs, trains will be equipped
for double stacking and trucks are allowed to carry more than two containers.
In the transport need marker information communication technology services will play an
important role. In Europe, there are currently two Internet-services available for carriers:
Teleroute and Freecargo. These intemet-services try to coordinate the exchange of LTLdeliveries and trucking capacities. In studies the opportunities to develop these services for city
logistics are evaluated for demand-driven and non-regulated transport services. The
development of these so-called virtual freight centres could have tremendous stimuli for
cooperative transport. These services are discussed in detail in Section 9.4.

9.2 INTERMODAL
TRANSPORT
SYSTEMS
9.2.1 Overview
During the past decade containerised intermodal transport has grown in importance as a
competitive alternative for the growth in road transportation in order to reduce (road)
congestion and environmental pollution. This growth will certainly continue in the next decade.

208 (Cig Logistics)

For instance, in Europe the ministers of transport of the European Committee try to coordinate
the development of intermodal transportation networks in their plan 'Trans European
Networks'. In contrast these policy plans the practice of terminal development and intermodal
transportation initiatives seem to develop more whimsical. Due to the small profit margins in
transportation the perspective of transportation entrepreneur is strongly focused on profitability
of new intermodal transportation services. Taking into account these different perspectives we
developed a three-stage modelling approach that is strongly connected to the perspectives of
the transportation entrepreneurs and the shippers who potentially will exploit these services.

In many national programmes for traffic and transport themes like accessibility and liveability
have become main issues in the development of a sustainable society. One of the important
policy strategies in order to meet elements of a sustainable society is the modal shift of road
transport to more environment-friendly modes like railroad, coast and barge transport. To
compete with road transport these transport modes are multimodal set-up as intermodal
transportation services with pick-up and delivery services by truck. According to a U.S.
General Accounting Office report '...the trucking industry also provided flexible, reliable, and
economical service, but the growth of trucking has contributed to concerns about safety,
congestion. pollution, and highway deterioration' (USGAO, 1992). Due to these facts the
attention of the Ministers of Transport for intermodal transportation is increasing and several
policy documents have been published concerning this subject.
At different levels of policymaking, i.e. the continental level, the national level. and the
regional level, we notice an ambiguity of plans for stimulating intermodal transport. At each
level policy makers base their own plans for intermodal transport focusing on different goals.
The integration of these plans, as well as the attention for the stakeholders, seems to be
forgotten. Therefore, the practice of intermodal transportation seems to develop more
whimsically, and not as expected in the policy plans.
On policy planning three specific arenas can be identified in which actors play their part in
relation to intermodal transport. Each arena can be seen as a level of consideration in which
actors are relationally joined. Each perspective of an actor on intermodal transportation is given
in terms of goals and instruments. Due to this insight of these perspectives in the specific
arenas we have developed three different models. Each model is developed in such a way that it
seeks for solutions; taking into account the main performance indicators of the arena.

(Future Perspecti1,es) 209

Policy
Levels

Transport
Modeling

Location Model
I

National
I

Costs Analysis
Regional

Figure 9.2 Modelling Approach

9.2.2 Continental level

The first arena is oriented at the actors playing a part in the continental level. In Europe, the
continental level is called the European hinterland. The main borders of this area stretch out
from the seaports: Le Havre, Zeebrugge, Antwerp, Rotterdam, Bremen, and Hamburg to
several European hinterland terminals, such as Milan, Metz. and Munich etc. The main actors
involved are shippers, European Ministers of Transport, road carriers, intermodal agents and
their carriers. Many shippers have changed their distribution structures as a result of the
removal of trade barriers. Due to the cost reducing opportunities many stock-holding units have,
in their distribution structure. been eliminated or outsourced. To maintain the same reliability
just-in-time deliveries have become more important. Therefore. a growth in transport, mainly
conceived by road carriers, has been inevitable and has caused, in interaction with motor traffic,
congestion on important highways. The congestion negatively influences the total transport
time for delivery, the reliability of the delivery, and, eventually, the costs. For this reason the
shippers are more interested in new intermodal transport initiatives. Besides this reason, some
shippers strive for a green company image by choosing environment-friendly transport like
intermodal transport. There are still some obstacles in the eyes of the shippers that have to be
solved. Their main concern is the current price of intennodal transport.

2 10 (Cip Logistics)

If intermodal transportation services will play an important role in future there is another point
that needs to be improved. The total number of terminals connected by water and/or rail
infrastructure is still too limited regarding the enormous supply and flexibility of road transport.
In Europe, the Ministers of Transport try to draw up plans for the development of Transport
European Networks (TENs)(EG/1692/96, 1996). These plans are focused on integration of the
transport modes and define the transport networks in hubs and spokes. At hubs fast
transhipment should be facilitated and subsequently transported by the spokes. The integral
management of the transport at a hub consists of physical transport infrastructure, traffic
controlling systems, positioning systems and navigation systems. The Ministers of Transport
try to co-ordinate and stimulate the development of infrastructure in each country in the
direction of the plans for TENS. They do not have an explicit authority for steering instruments
on national governments, but they can stimulate some developments by providing subsidies.
The intermodal agents and carriers have already established intermodal transport services for
long distance haulage. Large maritime container flows are transported by these agents from
harbours to far locations in the hinterland (for example from Rotterdam to Milan). Facing the
break-even distances it is possible for carriers to exploit profitable transport services. So far it
seems impossible to collect thick container flows for short distances. Because of their strong
competitiveness, price and flexibility, the road carriers still have the greatest market share as to
transport. The internal competition between road carriers is strong and therefore many carriers
operate break-even or less than break-even. As in other lines of business merges, take-overs
and creating alliances are daily news and should be interpreted as a matter of strategic
management to survive in the long run. Some road carriers experience hindrance caused by
some measures of national policies that aim at reducing the truck traffic.
9.2.3 National Level
National governments are represented by respectively the Ministry of Transport. Economic
Affairs and Environmental Affairs. All the ministries show a strong passion for the
development of intermodal transport. The Ministry of Transport tries to stimulate intermodal
transport initiatives to maintain the accessibility to important economic centres. If available in
countries, the rail and water infrastructure still has enough unutilised capacity to adapt more
traffic on these modes. The environment-friendliness of these modes appeals to the Ministry of
Environmental Affairs. The Ministry of Economic Affairs attaches importance to the economic
generating value of the hub-terminals. At these terminals a lot of transhipment is carried out
and these terminals attract companies having a good accessibility by all kinds of transport
modes. While the Ministry of Economic Affairs is strongly focused on the economic generating
value, a governmental policy can be observed being extremely focused on terminal
development within the frontiers of their countries. Combined with the knowledge of the break-

(Future Perspectiises) 21 I
even distances of intermodal transport the national policies are completely frontier-oriented
by trying to develop economic activities in their own countries.
The freedom of transport choice seems to be a more important issue for shippers. Transport
associations strive for fair competition in the transport market without governmental
interference regarding any of the transport modes. Truck carriers want to maintain their current
position as well as its total number of members. At the moment the transport sector has a keen
internal competition and many carriers even accept losses. The inland intermodal agents are to
a limited extent represented by these transport associations. These agents just develop longdistance services.

9.2.4 Regional Level

At the regional level local authorities would like to develop their cities in terms of economic
growth by providing accessible company fields. The attraction of an intermodal terminal could
be a serious alternative for the improvement of accessibility. By providing subsidies orland
raising lower ground taxes they try to attract companies to their areas. For this reason some
shippers are reconsidering locations, but their main motive for changing locations is the
accessibility and their location towards their customers positions. Shippers try to organise their
transport with high frequencies (allowing inventory reductions) against low wages and demand
a high flexibility towards ordering times. Therefore, terminal agents have to attract large freight
volumes and transport them with high frequencies on a regular basis.

9.2.5 Operational Level

At operational level the carriers do their utmost to follow the fixed transport schedule. The
terminal operator wants to use its transhipment equipment and its floor space as well as
circumstances allow. If the shipper agrees on price and transport schedule, he/she wants to be
on-line informed about the transportation progress.

9.2.6 A Linear Programming Model for Terminal Assignment

Since price seems to be the most important performance indicator at both the European and
National level, we developed a linear programming model that provides us with some insights
into price variation. The model is applied for locating shuttle-services between terminals in
Europe. A shuttle-service is a general logistic concept for intemodal transportation services.
The service operates with a fixed (time-) schedule to fixed terminal locations elsewhere and
visa versa. The continental container transport demands (NEA, 1992) between several
European regions, with origin or destination Netherlands, are optimally assigned to the

2 12 (Cig-Logistics)

transport network. Actually the model is a special case of the location model described in
paragaph 8.3. The transport network contains direct road transport connections and intermodal
transport connections between terminals. The optimisation is restricted to one year. The
objective function is defined as follows:

The objective function contains the following cost elements:

Shuttle Costs per TEU' between terminals. The costs of a shuttle connection are
determined (by the model) at the largest volume in one direction. The volume in the
reverse direction should be geared to this volume as much as possible.
Costs per TEU for pickup and delivery, and the transhipment costs per container at a
terminal.
Costs per TEU for direct road transport between regions.

(i)

(ii)
(iii)

The linear search algorithm minimises the objective function subject to the following
constraints:
b' j , k

(9.2)

b'k, j

(9.3)

The capacity of an intermodal shuttle-service connection is determined by the volume of the

largest transport flow in forward or backward direction. This restriction is necessary to model
the return of empty containers. The sum of transported containers between region i and region
I must be met for the specific demands.

x,): volume of direct road transport from region i to region I

xiJiI: volume of the intermodal transport from region i via terminal j and terminal k to region
1

X J h: maximum volume of a transport connection between terminal j and terminal k

D,, : transport demand between region i and region 1

(Future Pei.spectii>es)21 3
dOd: distance from location o to location d

cr8!: costs per TEU-kilometre for road transport between region i and region 1
c,, : costs for pick-up and delivery per TEU-kilometre from terminal

j to region i

c ~ :, shuttle
~
costs per TEU-kilometre between terminal j and terminal k

ch : costs of transhipmentlhandling per TEU at a terminal

T : collection of selected terminals
R : collection of all regions.
Applying this model formulation to the TEMII-data provides us with a large solution space.
The model contains 129 regions with a transport demand, 21 choices for intermodal rail
terminal locations, and 13 choices for intermodal barge terminal locations enabling more than 9
million alternative transport connections. To reduce the solution space some restrictions have
been added to the formulation. This LP-formulation allows us to make a network optimisation
based on costs. The results of this model indicate which terminal locations could have
competitiveness towards road transport. Furthermore, based on the incoming and outgoing
transport volumes, an evaluation of the geographical position of a terminal could be given. The
following tariffs were applied in the model.

Table 9.1 Overview of cost drivers

Cost drivers
Road transport direct
Road transport drayage
Rail Transport
Barge Transport
Transhipmenthandling rail
Transhipmenthandling barge

Price
Dfl. 1,20 per km
Dfl. 1,20 per km
Dfl 0,32 per km
Dfl 0,25 per km
Dfl 60,Dfl 50,-

Results of the linear programming model

The results of the model indicate a restricted attraction for the terminals mentioned in Dutch
policy plans. Apart from the barge terminal at Nijmegen, the other terminals located nearby the
frontiers seem to be less important. The Dutch policy plan for terminal development has
forgotten to take into account the international competition of terminals situated nearby the
Dutch frontiers, such as Duisburg in Germany and Liege in Belgium. Not mentioned in the
I

TEU = Twenty foot Equivalent Unit

214 (Cig Logistics)

Dutch policy plans are terminals for short distances, i.e. less than 50 kilometres. The model
results show opportunities for inland terminal development, i.e. Utrecht and Leided Alphen.

9.2.7 Detailed Cost Model

With the above linear programming model we obtain rough indications of possible terminal
locations. The next model is applied to identify more specific where a terminal should be
located on a detailed cost specification at regional level.

For the identified terminal initiative LeidedAlphen a detailed activity-based-costs-model

(spreadsheet-model) was developed (van Ham e t a / . 1997). In order to determine the total costs,
the costs for transhipment, the cargo-handling equipment, acreage and personnel requirements
are relevant. Each cost component will be discussed.

Table 9.2 Typical requirements for inland (barge) terminals (NEMaskoning, 1991)

Cost Drivers

<5000

TEU
Equipment :
-gantry crane
-mobile crane
-forklift truck
Personnel:
-crane operator/
forklift driver

500015000
TEU

>15000 TEU

I
1

-gatehouse
Acreage
(hectares)

<0,33

0,33-1.0

>1,0

On this basis, the cost of transhipment can be calculated. In this model, due to economies of
scale, the costs per container will decrease, but interestingly, stabilise when throughput exceeds
approximately 10 thousand containers annually.

(Future Perspecriws) 215

Dutch
florins

200
180
160
140
120
100
80
60

20
40

v
A

I
0

5000

10000

15000

20000

containers

Figure 9.3 Container handling costs (non linear) at inland terminal (Konings, 1993)

Transportation by barge

In relation to transportation by barge, two situations can be recognised. A new inland terminal
may be located near an inland waterway, where scheduled services by barge already exist. In
this case the new terminal is just an extra port of call and a fixed tariff can be arranged with the
inland water carrier. Otherwise, a complete new service must be organized. The fast majority of
the (Dutch) inland waterways can accommodate ECMT IV class-barges with a capacity of 1500
tons or 90 20-foot containers. The costs of hiring such a barge amount F1 1 million annually
(+/- 0.5 million US$).
Dray age
Local pickup and delivery of containers is usually carried out by truck. The fee charged for the
movement of a container between the terminal and the point of o r i g d destination differs from
the tariffs in long distance road haulage. For short distances, costs are determined more by time
(Fl 66.10 per hour) than by distance (F10.71 per kilometre). For long distance road transport an
average tariff of F1 1.85 per kilometre was used.

21 6 (City Logistics)
Results cost model Alphen d d &in
At the moment of this study, a new inland terminal in Alphen aan den Rijn was under
consideration. In this region several large shippers were found to be, amongst others, the
Heineken Brewery and the Swedish Electrolux Company. This initiative received a warm
welcome. Given the amount of cargo, forecasts indicate a throughput of at least 20,000
containers annually, or 80 containers every day. To get an impression of the terminal
operations: the rail mounted gantry crane needs approximately 5 hours to (un)load the barge. In
the vessel, two out of every three slots are in use.
After making location comparisons on calculated tariffs, the terminal was finally built South of
Alphen aan den Kjn, which allows daily sailings to the Port of Rotterdam with just one barge.
Special attention should be paid to minimize the number of callings at the deep-sea terminals in
the port area. Moreover, handling barges at the sea quay incurs extra costs. These extra terminal
handling charges (THC) were added as a surplus of F1 20 on the transhipment costs. The final
cost comparison between road transport and intermodal transport showed four favourable
locations in the vicinity of Heineken (south of Alphen aan den Rijn). The calculated tariffs
demonstrated opportunities of cost reductions up to twenty percent of the road tariff.

9.2.8 Simulation Model for Terminal Operations

If shippers decide to service their transport demands by intermodal transport, the management
of the operations becomes an important issue. Transport schedules have to be determined in
such a way that from the shippers perspectives the time conditions for delivery have to be met,
from the carriers perspectives the vesselhain loads have to be filled up as much as possible,
and from the terminal agents perspective the productivitys of the crane are to be guaranteed.
To visualise and measure these individual important factors of the actors the logistic processes
were modelled using a simulation model. After the arrival process of ships the loading and
unloading activities of ships is carried out by cranes. The straddle carriers bring and pick the
containers to/form the stack. Other straddle carriers bring and pick the containers to the gates of
the terminal, where the trucks are waiting to be (un)loaded. If the gates are occupied the trucks
have to wait outside the terminal.

(Future Perspectiies)217

Figure 9.4 A simulation model (built in Arena) of terminal operations

Based on our experiences with this three-stage-modelling approach we are able to indicate
locations for new terminals. With the use of these models the financial feasibility of a new
terminal can be calculated. By applying this modelling approach two identified terminal
locations have lead to serious terminal initiatives in Alphen aan den Rijn and Utrecht (Govera,
1998). The gap between the policy plans of the governments and the level of operational
processes has grown too far apart. The dynamic behaviour of the actors involved can not be
static stated in policy plans, but as Muller (Muller, 1995) mentioned: iiztermodal
traizspoi-tation is not just the hardware or equipment involved with the freight moiement, bur
the process, which becomes a major component of the systems approach to business. Thanks
to the recognition of actors in specific arenas we can build specific decision support models for
incorporating the specific factors of concern. In the models traditional sensitivity analysis can
be applied on the critical parameters. Nowadays we see a strong growth in the development of
intermodal services and their role in transportation will certainly increase in future decades.

2 18 ( C i v Logistics)

9.3 UNDERGROUND
FREIGHT
TRANSPORT
SYSTEMS
9.3.1 Overview
A technology, which can play a big role in future City Logistics systems, is underground
transportation. Underground transportation is already common technology in passenger
transportation systems such as subways. In freight transportation systems the industrial
application of this technology is mainly restricted to transport by pipelines for some chemical
products, oils etc. Underground logistics distribution systems seem to be a sustainable solution
for environmental, congestion and space problems. In highly congested areas initiatives for
underground projects have been studied over as an alternative. Initially Japan started with the
study of underground distribution systems (Koshi et al., 1992; Ooishi and Taniguchi, 1999).
Actually most of these initiatives have not been realised, since they are innovative systems with
automation that requires a huge initial investment, although cost benefit analyses showed
positive results. (See Section 5.5)
Despite these results in the Netherlands, the Centre for Transport Technology provided large
subsidies to investigate opportunities for underground freight transport systems. The reason for
investing in this technology is based on the fact that other industries, i.e. the drilling industry,
can also benefit from these funds. For a specific project, called the Underground Logistic
system Schiphol, the cost-benefit calculations showed an economical feasible exploitation
under assumptions of certain growth scenarios (CTT, 1997a-d). For this project the conditions
seems to be (economically) favourable while two big private companies, flower auction
Aalsmeer and Airport Schiphol, are willing to invest. The knowledge and technology applied in
this project could be reused for underground city distribution projects.
Therefore, underground freight transport systems are included as a future development for City
Logistics. At the city level, some municipalities have already adopted this idea and have started
to work on a feasibility study for this new technology. At the regional level some companies
located at different industrial zones could become connected by an underground distribution
system. Especially companies with a strong supplier-customer relationship in terms of volumes
and pieces are potential users of these systems. Due to the flexible, continuous and reliable
underground transport system just-in-time deliveries become possible, thus reducing stock
positions at suppliers' as well as at customers' end. At a national level, huge economic centres
with underground distributions networks will be interconnected by high-speed rail connections.
The feasibility of these projects will be examined for logistical aspects, technical specifications
of infrastructure and means of transport, financial aspects, and environmental conditions. To
give an idea for some of the critical design issues:

(Future Perspectiws) 2 19

(a) Layout of the network

Which kind of layout can service most locations? What kind of traffic control system do we
need at the junctions in the network under all kinds of circumstances?
(b) Tubes
What are the influences of gradients on the performances of the system? Should the tube be
used for one-way traffic or two-way traffic? What should the diameter of the tube be?
(cj Energy supply
What is the influence of the power supply systems, like wire conduction, or loadable
batteries with power loading stations on the layout of the system?
(d) Terminals
How many terminals should be available in the system to meet customers requirements?
What should be the layout of terminal in terms of docking stations, driving lines and
parking facilities?
(ej Vehicles
What are the relationships between the physical characteristics (like speed. acceleration,
braking distance, sizes, and loading capacity) and the requirements derived from the
infrastructure (e.g. safety and guidance)?

(0Dock stations
What is the influence of top, rear, one-sided or two-sided loading?
(g) Freight flows
Can we define the maximum handling capacity of these systems? Can the system facilitate
loads?
(hj Disturbances and recovery
What are the effects of physical disturbances (for instance a defective vehicle in a tube)?
How can the system be reactivated again?

To answer these types of design questions it is very useful to use simulation modelling as a
technique. Most design issues are interrelated and therefore simulation can deal with these
dynamic characteristics. For setting dynamically -up several simulation experiments the
simulation approach is based on the development of library of components (objects). As a
consequence of this approach all k i d of configurations can be built and evaluated quickly. The

220 (Cig Logistics)

chosen language is SiMPLE ++, an object-oriented simulation language allowing hierarchical
building blocks (Aesop, 1997).
Of course, not all the questions can be answered by the use of simulation. For instance the
determination of the tube diameter size was identified with the application of Group Support
Systems. About a hundred potential users/experts brainstormed and voted on arguments pro
and contra 3% or 5-metre tube diameter. The final discussions evaluated after these sessions
showed a rich picture for making an appropriate, well-considered decision (De Vreede et al.,
2000)
9.3.2 Conceptual Modelling
To obtain a structured insight into all the dependencies among the logistic activities, the
technical aspects of vehicles and infrastructure, the transport demands and the interconnectivity
with other transport modes, three conceptual modelling levels of control are defined here:

at interconnection level (macro), in order to

achieve good connectivity with the transport
schedules of the external transport
modalities:
at transport order level (meso), in order to
control the transport orders, the assignment
of shipments and vehicles and the
management of empty vehicle stock;
at vehicle level (micro), in order to control
the traffic movements of vehicles in the
handling areas and tubes.
Figure 9.5 Systems Hierarchy (van der Heijden, 1997)
The construction of the hierarchy in conceptual models is important to provide transparency in
the models. Each controlling level can be considered a market, i.e. the dynamic interaction
between demand and supply. From each controlling perspective the models are specified to
such a level of detail being relevant for the market situation. Consequently, some processes or
activities are modelled as a black box. The black box is specified in more detailed in another,
usually lower level controlling perspective. The validation of this chosen hierarchy in models

(Future Perspectiixes) 22 1
can be evaluated by techniques such as meta-modelling and experimental design. Each
controlling level will be discussed in detail.
The micro-level: traffic management
The large-scale nature as the number of vehicles and transport demands, the great traffic
intensity caused by short throughput-times, and the high capacity loads require high reliability
from the transport system and great flexibility in the transport services. To meet these
engineering aspects, heavy demands are made upon the development of an adequate controlling
system for vehicle and traffic mana,oement.
We can base ourselves on the vast research experience relating to vehicle management which
was derived from the implementation of the automated guided vehicles at the terminals and the
implementation of a new transport technology CombiRoad (Evers, 1994). At present, most
Automated Guided Vehicles (AGV) systems working are controlled by a central system. For
example, two vehicles crossing each other, both slow down resulting in a loss of time. If the
intelligence will be decentralized to the level of vehicles, the vehicles can respond using their
own knowledge of traffic-rules. A new type of vehicle and traffic control has been developed
recently (Evers, 1997). The system design consists of four segments being called the Process
Infrastructure. The Process Infrastructure is specified in terms of primitives, such as (domain)
positions and semaphores, controlling the conflicting usage of resources by several demanding
actors. These primitives are categorized in the next four segments:
(i)

Process Co-ordination
The function of this segment is to control processes that try to allocate the competitive
process resources of the available process infrastructure. The way in which the processes
actually take place is described in so-called 'scripts'.

(ii)
Priorities & Timing
The modules of the segment Priorities & Timing provide additional information regarding
the intelligence of the semaphore-mechanisms in the segment Process Co-ordination. This
information is supplementary in terms of additional controlling with respect to timing rules
and priority rules.
(iii) Actor Mission
The segment Actor Mission registers information about the arrivals of the actors,
Information about the actor's position, its timing and priority is given as pre-information to

222 (City Logistics)

the segment Priorities & Timing. The actual arrival is finally given by the Process Coordination.
(ivj
Basic Processes
Basic Processes contain modules for the elementary physical processes that have to be
carried out by an actor. The input/output channels of these modules are somehow connected
with the domain positions of the process infrastructure.
The difference and its strength compared with the more conventional system architectures and
this system architecture are at the exclusive availability of the resources. In traditional systems
only one vehicle is allowed to be in a control segment. Even if the routes of the vehicles are not
conflicting, the system will ask for an exclusive usage of this segment. Generally, loss of
handling capacity and growth of the lead-times will be the result. In informatics the problem
with concurrent use of hardware seems to have identical characteristics. Dijkstra (1968) and
later Ben-Ari (1990) introduced the semaphore mechanism. The threshold of a semaphore, say
value S, is a non-negative integer that represents the capacity available. With signal (Sj-wait
(S)-instructions the permission of entering a resource can be controlled. The first tests for
different crossings seem to provide good results with respect to conflict handling. At this
moment at Delft University the logic of the simulation software is embedded into the hardware
of vehicles and semaphores and will be tested and evaluated in a 'real-life' laboratory. Six real
vehicles are driving around and will be evaluated while the other vehicles in the system are
being simulated.

Figure 9.6 Illustration of the simulation and the interaction testing-vehicles

(Photos: Intermediair. Holland)

(Future Persyecti1,es) 223

The meso-level: order management

To develop the order management, the logistic functions to be performed in the logistic concept
are identified. On the basis of underground urban freight distribution we demonstrate the
logistic concept. Goods with a destination in the inner city are received just outside the city at
an interchange referred to as a Logistics City Park (LCP). At this LCP a storage warehouse is
installed for rolling stock articles enabling frequent deliveries to customers. The LCP also
facilitates the direct distribution and the cross-docking flows. For these flows
grouping/degrouping and sorting activities are provided. The goods are transhipped on the
AGV's and transported to one of the district distribution shops (DDS) in the inner ciry.

incoming inter-city transport

outgoing in r city transport

9-

LCP

receipt hq cuqtomer

prepare
for tranrport

prepare
for tranrport

Figure 9.7 Logistics system design for an underground logistic network (meso-level)

From a modelling perspective the logistic functionality of DDS can be copied from LCP. The
logistics intensities are only on a small scale but the logistic functions are similar. From the
DDS goods can be temporarily stocked and finally distributed to the shops, consumers, offices
and businesses. The district interchanges also function as the starting point for collecting the

224 (CieLogistics)
return of goods, waste products and packaging. Several network configurations are set-up and
are optional in relation to the radius of the service area. Table 9.3 represents the implications of
several parameter values of the service radius for an average town of 200,000 inhabitants.

Service Area

# Houses/Shops

# DDS

Length of pipes

75 m

50

1800

250 km

250 m

500

180

100 km

500 m

2000

45

50 km

750 m

5000
25000

18
4

30 km

3000m

25 km

With this conceptual model we are able to perform sensitivity analyses for shortening leadtimes and changes in the network configuration in order to determine the total fleet-size of
vehicles in the system. As a result of the simulated transport demands and the resulting vehicles
moves, we will have a closer insight into the necessity of the underground spaces.

From the meso-level the arrival and departure processes of incoming and outgoing transport
loads are to be derived. In case of the underground urban transport system the transport-loads
arrive at the LCP and are distributed by the underground system. The return flows are collected
by the underground system and are temporarily stocked at the LCP. For the underground urban
transport system the LCP is the highest echelon. From this echelon the transport loads have to
be co-ordinated to the external transport processes. The LCP has accessibility for transport
modes such as rail transport, barge transport and road transport.

An important characteristic of these transport modes is the batch-production. Usually the

transport-loads of trucks, vessels, or trains contain several packages for various customers.
Therefore, the planning of empty vehicles should be geared to these arrival processes. For the
return flows the number and space of loading-docks have to be determined. The choices and
locations of loading-docks dependent on the external transport schedules. Due to the library
building some configurations of docks settings can be evaluated. For the ULS-project
calculations has been based on the busiest day estimated in the year 2020. In figure 9.7 we see
one of the best performing terminal designs (LCP). The middle of the terminal contains tracks
for quick passing and parking. Due to the large amount of available tracks the distribution of

(Future Perspectives) 225

vehicles is spread out over the terminal. At the dockside there is room for just one waiting
vehicle. The utilization of the docks can therefore become very high.

Figure 9.8 Terminal Layout

Most of the underground projects have just passed the definition phase and have continued
their project work with an exact specification of the elements. To support this work, similar
simulation studies have been started.
At the city level, towns in the Netherlands are examining the possibilities of underground
freight transport. A detailed plan has already been developed for Utrecht. This plan contains a
description of the network structure. The network consists of two underground circles with one
common central point, the LCP. The total number of DDSs in the network is 38.
The last project to be started at a regional level is the project in the Province of Limburg. DSM,
an international company for chemical products, NedCar, an assemblage facility of Volvo and
Mitsubishi, some sub-suppliers, a rail terminal and a barge terminal will be connected by an
underground distribution network. The first ideas of underground distribution arose two years
ago. So far, not all the actors are directly involved and many people have to be persuaded to
acknowledge the possible merits of an underground distribution network. Again, simulation
models can play a key role during this phase, because the internal processes of the system can
be animated. Due to this support, one can imagine how processes really work, and changes in
attitudes can occur.
Finally, it is to be concluded that the development of generic simulation models will provides
valuable insights into many technical engineering issues that are important for progressing the
development of underground distribution networks. However. there still a lot more questions

226 (Cig Logistics)

concerning underground distribution networks that cannot be resolved using simulation
modelling, including:

Is it public infrastructure? Should one company be responsible for the total freight
transportation or are other transport agencies also allowed to use the network? Should access to
the network be regulated by the municipality? Should an access place to the network be
accompanied by a city distribution centre? How about legal regulations - existing law are
mainly specified until 20 meters below the surface? Should the government provide subsidies
for underground technology?
So far, these questions have to be brought into the discussions about underground
transportation. Therefore, it will be probably around the year 2010 at the earliest that an
underground distribution system will be implemented.

9.4 VIRTUAL
FREIGHT
COMPANIES
An emerging new concept in the area of inter-organizational co-operation is the concept of
virtual organizations. It is difficult to find a common definition of a virtual organization. Many
authors (Davidow & Malone, 1992; Byme, 1993; Mowshowitz. 1994) have described a view
on the concept of virtual organizations and its value for business. Regarding City Logistics, the
definition of Byme seems to be most appropriate, where a virtual organisation is defined as, 'a
temporary of independent companies - suppliers, customers and even competitors - linked by
information technology to share skills, costs and access to one-another's markets. It will have
neither central office nor organizational chart. It will have no hierarchy, no vertical integration'.
For virtual freight companies one should add to the definition of virtual organizations, they
seek to realize more efficiency in the freight transport by collection principles. The main
markets for these companies can be found in the LTL-deliveries. At this moment there already
exits a number of virtual freight companies (in Europe):
(a) Teleroute and Freecargo are internet-based services that try to establish a freight demand
exchange between carriers in Europe. These services are reasonably successful in matching
(rest) load capacities and (rest) loads. For instance more than 10 percent of the Dutch
carriers are allied to these services.
(b) Paris (Planning And Routing Intermodal System) is an information system developed by
Cairo Systems that plans the hinterland transport of containers to and from ports. The main
objective of this system is to minimize empty mileage and to reduce transport costs within

(Future Perspectii5es) 227

the rules and preferences indicated by various shippers. This system could be called a
multi-actor planning system. Shippers send their transport orders to the system, while
transport companies provide information on the availability of their capacity and resources.
The assignment of loads to transport tasks is dynamic in real-time, ie. until the moment the
transport order is sent to a driver or transport operator, the planning can be adjusted.
(c) Dadira is a project in which producers, shippers, carriers and shopkeepers work together in
order to cooperate in such a way that the transport from producer to shopkeeper will
become more efficient in terms of reduction of empty mileage and higher truckloads. The
difference here is that this system is only accessible for their members. Recently two
website tests have proved that the profitability of the parties involved was improved for
truckload performance.
Still, investigation of the systems shows that these systems are set-up from a supply structure
instead of a demand-driven system. We can also conclude that most of these systems fail to
create more value for customers by actively matching, optimising and combining. Therefore,
we are probably at the start of new types of virtual transport services.
For successful implementing such a system [Bos 19991 the following conditions have to be met
for the application field of City Logistics:

.
..
.
m

The systems should be an integrated part of structure segments of current distribution

networks;
Participating receivers should have clear advantages to join the system;
Collection of some transport providers (from the start) in order to guarantee transport
quality and to give a transport guarantee;
Connection to the latest logistics trends of 24 hour delivery and JIT-delivery;
Attention to price-setting (seeking for matches):
Open-order-entry .

228 (Cify Logistics)

Demand side 4

Supply side

Interface A
respond card
telephone
telefax
PC
internet
other

Interface B
mail handling
call centers
voice response centers
application servers
internet (servers)
databases

Figure 9.9 ICT application for a demand-driven transport market

The pilot studies, which have been carried out, can be seen as a start to a new generation of
supply chain management software. The virtual freight centres can lead to greater opportunities
to embed modelling tools and decision support systems. The operational packages and
modelling support can be integrated into supply chain management software as illustrated in
Figure 9.10.
At this moment the APS (advanced planning & scheduling) have reached the stage of
integration with the most of the current ERP software-packages (Enterprise Resource
Packages). In the future, integration of logistics modelling and transportation modelling can be
embedded in these environments too. These developments will certainly stimulate
developments in the field of logistics modelling.

(Future Perspectiixes) 229

Operational Transactions

Logistm Modeling
Transportation scheduling

Data warehousing

Supply chain software integration:

Alliances between software suppliers
with standarised interfaces (EDI)

Figure 9.10 Integrated environment for supply chain management

(References) 23 1

REFERENCES
Abdewahab, W.H. and M.A. Sargious (1992). Modeling the demand for freight transport: new
approach. Journal of Transport Economics and Policy, 26(1), 49-70.
AESOP SiMPLE++ Tutorial (1 997). Optimization of Systems and Business Processes. AESOP
GmbH, Stuttgart.
Allen, W.B., M.M. Mahmoud and D. Mcneil (1985). The importance of time in transit and
reliability of transit time for shippers, receivers, and carriers. Transportation
Research, 19B(5), 447-456.
Arem B. van (1996). The development of Automated vehicle Guidance systems - A literature
survey. TNO-report INRO-VVG 1996- 15, TNO-Institute for Infrastructure, Transport
& Regional Development, Delft.
Ashtakala, B. and A.S.N. Murthy (1988). Optimised gravity models for commodity
transportation. Journal of Transportation Engineering, ASCE, 114 (4), 393-408.
Banister, D. (1994). Transportplanning in UK, USA and Europe. E&FN SPON, London.
Beckmann, M.J., C.B. Mcguire and C.B. Winstern (1956). Studies in the Economies of
Tramporatation. Yale University Press, New Heven, CT.
Bell, M.G.H. and Y. Iida (1997). Transportation Network Analysis. Wiley, UK.
Ben-Akiva, M. and S.R. Lerman (1985). Discrete Choice Analysis. MIT Press, Cambridge,
MA.
Ben-Ari, M. (1 990). Principles of concurrent and distributed programming. Prentice Hall.
Bergan, A.T., C.F. Berthelot and B. Taylor (1995). Effect of weigh in motion accuracy on
weight enforcement efficiency. Proceedings ITS World Congress, Vol. I, pp.504-508,
Yokohama, VERTIS.
Birge, J.R. and F. Louveaux (1997). Introduction to Stochastic Programming. Springer, New
York.
Bitzios, D. and L. Ferreira (1993). Factors affecting Route Choice of Commercial Vehicle
Drivers. Proceedings Australian Transport Research Forum (ATRF), Vol. 18, pp.2740.
Boardman, A.E., D.H. Greenberg, A.R. Vinning, D.L. Weimer (1996). Cost-benefit analysis.
Prentice Hall, Upper Saddle River, NJ.
Boerkamps, J. and A. Van Binsbergen (1999). GoodTrip - A new approach for modelling and
evaluating urban goods distribution. In: City Logistics I , 1 st International Conference
on City Logistics, Cairns (E. Taniguchi and R.G. Thompson, eds.), Institute of
Systems Science Research, Kyoto, pp. 175-186.
Bonomi, E. and J.L. Lutton (1986). The asymptotic behaviour of quadratic sum assignment
problems: A statistical mechanics approach. European Journal of Operations
Research, 26, 295-300.
Bos, A.C. (1 999). Vraaggestuurde Viutuele TransportMurkt. Stenfert Kroese, Rotterdam.

232 (City Logistics)

Bots, P.W.G., Twist, M.J.W. J.H.R. van, Duin (1999). Designing a Power Tool for Polic))
Analysts: Dynamic Actor Network Analysis. Paper submitted to HICSS-32 Modeling
Technologies and Intelligent Systems track.
Bowersox, D. J., D.J. Closs and O.K. Helferich, Omar (1986). Logistical Management,
Macmillan.
Bowyer, D. and M.A.P. Taylor (1985). Traffic and Transport Information Systems. In:
Microcomputers in Traffic and Transport (D.P. Bowyer, ed.), REAAA Workshop,
Singapore, Australian Road Research Board, pp. 151-60.
Bowyer, D.P. and K.W. Ogden (1988). Impacts of urban road investment on road freight
transport: Gateway Bridge Case Study. Proceeding ARRB Conference, Vol. 3, pp.224235.
Branston, D. (1976). Link capacity functions: A review. Transportation Research, 10, 223-236.
Bruijn, J.A. de and E.F.ten Heuvelhof (1995). Policy networks and governance. In:
Institutional design (D.L. Weimer, ed.), Kluwer Academic Publishers, Boston, pp,
161- 179.
Bruynooghe, M., A. Gibert and M. Sakarovitch (1968). Une methode d'affectation du traffic.
Proceedings of the 4th International Symposium on the Theory of Road Traffic Flow,
Karlsruhe.
Bureau of Public Roads (1964). Traffic Assignment Manual. U S . Department of Commerce,
Urban Planning Division, Washington D.C.
Byrne, J.A. (1993). The virtual corporation. Business Week, February, pp 36-41.
Calogero, V. (1994). GIS in logistics and freight planning. Proceedings of PTRC European
Transport Forum, Seminar N , P385, University of Warwick, pp. 123-146.
Campbell, J. (1990). Locating transportation terminals to serve an expanding demand.
Transportation Research, 24B(3), 173-193.
Chancong, V. and Y.Y. Haimes (1983). Multiobjective Decision Making: Theory and
Methodology. North-Holland.
Chiang, W.C. and R.A. Russell (1996). Simulated annealing metaheuristics for the vehicle
routing problem with time windows. Annals of Operations Research, 63, 3-27.
Cosmetatos, G.P. (1 976). Some approximate equilibrium results for the multi-server queue
(M/G/r), Operational Research Quarterly, 27(3), 615-620.
Cox, D.R. and W.L. Smith (1961). Queues. Methuen.
Crainic, T.G. and G. Laporte (1998). Fleet management and logistics. Kluwer, Boston.
CTT (19974. rapport 23: Hoofdrapport deel 1 , CTT.
CTT (1 997b). rapport 24: Hoofrapport deel 2: Internationale verkening; Logistieke
effecten;Concept-voertuig; Besturingssysteem; Laad/lospunten, CTT.
CTT (1997~). rapport 26: Deelrapportage deel 2: Tunnelconstructie; Tracering;
Juridischhestuurlijke aspecten; Realisatie, CTT.
CTT , (1997d). rapport 21: Deelrapportage deel3: Subsitutie en Milieu-effecten; Financiele en
ecomische aspecten; Exploitatie-structuur, CTT.

(References) 233
Dablanc, L. (1998). Urban freight regulation and the new French urban mobility master plan.
8"' World Conference on Transport Research, Antwerp.
Daganzo, C. F. (1987). Modelling distribution problems with time windows, Parts I and 11.
Transportation Science, 21, 35 1.
Daganzo, C. F. (1994). The cell transmission model: A dynamic representation of highway
traffic consistent with the hydrodynamic theory. Transportation Research, 28B, 269287.
Daskin. M.S. (1995). Network and DiscreteLocation ; Models, Algorithms, and Applications.
John Wiley & Sons.
Davidow, W.H. and M.S. Malone (1992). The virtual corporation. Harper-Collins, New York.
Davidson, K.B. (1966). A flow-travel time relationship for use in transportation planning.
Proceeding of 3rd Australian Road Research Board Conference, 3(1), 183-194.
Demetsky, M.J. (1 974). Measurement of urban commodity movements, Transportation
Research Record, 496, 57-57.
Desrosiers, J., M. Sauve, and F. Soumis (1988). Lagrangian relaxation methods for solving
minimum fleet size multiple traveling salesman problem with time windows.
Management Science, 34, 1005.
Desrosiers, Y., Y. Dumas, M.M. Solomon and F. Soumis (1995). Time constrained routing and
scheduling, In: Network routing (M.O. Ball, T.L. Magnanti, C.L. Monma and G.L.
Nemhauser, eds.), North-Holland, Amsterdam.
Dijkstra, E.W. ( 1968). Co-operating sequential processes, Programming Languages.
Academic Press, New York.
Drezner, Z. (ed.) (1995). Facility Location ;A Survey of Applications and Models. Springer.
Drissi-Kaitouni, 0. (1 990). Solution approaches for multimode, multiproduct assignment
problems. Transportation Research, 25 (5), 3 17-327.
Duin, J.H.R. van (1997). Evaluation and evolution of the city distribution concept. Urban
transport and the Environment for the 21" Century III, WIT Press, Southampton, pp.
327-337.
Duin, J.H.R. van (1998). Simulation of underground freight transport systems. Urban transport
and the Environmentfor the 21"' Century IV, WIT Press, Southampton, pp. 149-158.
Duin, J.H.R. van, P.W.G Bots and M.J.W. van Twist (1998). Decision Support for MultiStakeholder Logistics. Proceedings TRAIL conference, Scheveningen.
Eibl, P. (1996). Computerized vehicle routing and scheduling in road transport, Avebury.
Euler, G.W. (1990). Intelligent vehicle/highway systems: Definitions and application, ITE
Journal, 60(11), Institute of Transport Engineers, 17-22.
Eden, C. and Ackerman, F. (1998) Making strategy: the journey of strategic management.
SAGE Publications.
Evans, L. and R. Herman, (1976). A simplified approach to calculations of fuel consumption in
urban traffic systems, Traffic engineering and Control, 17(8 and 9).
Evans, S.P. (1976). Derivation and Analysis of some models for combining trip distribution

234 (City Logistics)

and assignment. Transportation Research 1 0 , 3 7 3 7 .
Everall, P.F. (1968). The effects of road and traffic conditions on fuel consumption. Ministry of
Transport, TRRL Report LR 226.
Evers, J.J.M. (1994). DiTrans: integrated robotized inter terminal container transport, TRAILstudies nr. 94/5, Delft, pp. 1-27.
Evers, J.J.M., J.A. Ottjes M.B. Duinkerken (1997). Voertuiggeleiding als systeem van
automaten, Document nr. S-TRNALL-0006, version 1 .O, TRAIL Research School,
Delft.
EZ (1999). Nota Ruimtelijk Economisch beleid. Den Haag.
Fischer, F. and J. Forester (eds.) (1993). The Argumentative Turn in Policy Analysis and
Planning. Duke University Press, London.
Freier, M. (1981). Twin cities metropolitan area heavy truck study. Transport Research Record
920.
Fujii, S., Y. Iida and T. Uchida (1994). Dynamic simulation to evaluate vehicle navigation.
Vehicle Navigation & Information Systems Conference Proceedings, pp.239-244.
Fujimori, Y., Y. Yamada and Y. Hirabayashi (1999). Outline of ETC system introduction plan
and the standards draft. Proceedings 6 ITS World Congress, Toronto, (CD ROM).
Gallo, G. and S. Pallottino (1986) Shortest path methods: A unifying approach. Mathematical
Programming Study, 26, 38-64.
Gendreau, M. and J.Y. Potvin (1998). Dynamic vehicle routing and dispatching, In: Fleet
Management and Logistics, (T.G. Crainic and G. Laporte, eds.), Kluwer.
Gendreau. M., G. Laporte and J.Y. Potvin, (1997). Vehicle routing: modern heuristics, Chapter
9, In: Local search in combinatorial optimisation (E. Aarts and J.K. Lenstra, eds.),
Wiley. pp.311-36.
Glassco, R. (1999). Simulation analysis of congestion-reduction strategies at an overloaded
weigh station. Proceedings 6th ITS World Congress, Toronto, (CD ROM).
Glover, F. and M. Laguna, (1993). Tabu search. In: Modern heuristic techniques for
combinatorial problems, (C. Reeves ed.), Halsted Press, pp.70-150.
Goldberg, D.E. (1 989). Genetic algorithms in search, optimization, and machine learning.
Addison Wesley.
Goor, A.R. van, Monhemius, W. and Wortmann, J.C. (1993). Poly-Logistiek zakboekje.
Koninklijke PBNA, Arnhem.
Govera, ( I 998). Container Terminal Alphen d d Rijn, Container Terminal Utrecht. Nieuwsbrief
Govera, nr. 3.
Gray, R. (1 982). Behavioural approaches to freight transport modal choice. Transport Reviews,
2 (2), 161-184.
Hall, R.W. (1987). Direct versus terminal freight routing on a network with concave costs.
Transportation Research, 21B(4), 287-298.

(References) 235

Ham, J.C. van, J.H.R. van Duin (1997). Logistics: freight transport networks, alliances. In:
Managing Enterprises - Stakeholders, Enginieering, Logistics and Achievement,
Mechanical Enginieering Publications Limited, London, UK, pp 33 1-335.
Hansen, P., M. Labee and D. Peeters (1987). Systems of Cities and Facility Location. Harwood.
Harker, P.T. (1985). The state of the art in the predictive analysis of freight transport systems.
Transport Reviews, 5 (2), 143-164.
Harker, P.T. and T.L. Friesz (1986). Prediction of intercity freight flows, I: theory.
Transportation Research, 20B (2), 139-153.
Heijden, R.E.C.M. van der (1997). Towards a system view on transport development and
transport policy. Working paper, School of Systems Engineering, Policy Analysis and
Management, Delft University of Technology, Delft.
Heijden, R.E.C.M. van der (1995). Het vierde luik; Over de fundering van vervoers- en
verkeersbeleid.
Inauguration lecture, Faculty of SEPA, Delft University of
Technology (in Dutch).
Hellaker, J. (1999). DYNAFLEET Info System - at service to commercial users. Proc. 3d ITS
World Congress, Orlando, (CD ROM).
Heskett, James L. (1977). Logistics - essentail to strategy. Harvard Business Review, 55(novdec) , 85-96.
HID0 (1996). ITS Handbook in Japan, Highway Industry Development Organisation, Tokyo.
Homaifar, A,, S. Guan and G.E. Liepins (1993). A new approach on the traveling salesman
problem by genetic algorithms. Proc. 5 International Conference on Genetic
Algorithms, 460-5.
Hoppe, R. (1 998). Bouwstenen voor argumentatieve beleidsanalyse. Elsevier
bedrijfsinformatie, s Gravenhage.
Hutchinson, B.G. (1974). Estimating urban goods movement demands. Transportation
Research Record, 496, 1- 15.
INRO (1996). EMME/2 Users manual, Software Release 8.
Janssen, B.J.P. and A.H. Oldenburger (1991). Product channel logistics and city distribution
centers: the case of the Netherlands. OECD Seminar on Future Road Transport
Systems and Infrastructures in Urban Areas, Chiba, pp.289-302.
Jara Diaz, S.R. (1982). The estimation of transport cost functions: a methodological review.
Transport Reviews, 2, 257-278.
Just, U. and Krug, S. (1999). From truck guidance to truck navigation - experiences in the city
of Bremen, Germany. Proceedings 6th ITS World Congress, Toronto (CD ROM).
Kakihara, M., Y. Furuta, H. Terada and Y. Aoki (1999). Toyoda VPS based ERP system,
Proceedings 6th ITS World Congress, Toronto (CD ROM).
Kanzaki, H., H. Yamada and M. Kawana (1986). Fuel consumption of automobiles. Technical
Memorandum of Public Works Research Institute, No.23 19. (In Japanese)
Karuo, S. and Y. Koyasu (1995). Automatic measuring system for heavy freight vehicles.
Proceedings 2Id World Congress on ITS, Vol. I, pp.497-503, VERTIS, Yokohama.

236 (City Logistics)

Keenan, P.B. (1998). Spatial decision support systems for vehicle routing. Decision support
systems, 22, 65-7 1.
Kendall, D.G. (1953). Stochastic processes occurring in the theory of queues and their analysis
by the method of the imbedded Markov chain. Annals ofMathematica1 Statistics, 24,
338-354.
Kim, J.H., J.H. Lee and B.S. Park (1998). Development of the risk assessment methodology for
NRHM on highway and application to CVO. Proceedings 4th ITS World Congress,
Seoul (CD ROM).
Kim, S . (1998). GIS based risk assessment of truck hazardous materials movements in Korea,
Proceedings 4th ITS World Congress, Seoul (CD ROM).
Klabbers, J.H.G., (199 1). Simulation: learning environments for (self-) steering in social systems.
In: Autopoiesis and Configuration Theory: New approaches to Societal Steering (R.J. in
t Veld, L. Schaap, C.J.A.M.Termeer and M.J.W. van Twist, eds.) Dordrecht.
Kloot, G. (1999). Melbournes arterial travel time system, Proceedings 6h ITS World
Congress, Toronto, (CD ROM).
Kohler, U. (1997). An innovating concept for city-logistics. Proceedings 4th World Congress
on Intelligent Transport Systems, Berlin (CD ROM).
Konings, J.W. (1993). Tarieven in het containewewoer. Delft.
Koshi, M., H. Yamada and E. Taniguchi (1992). New urban freight transport systems. Selected
Proceedings of 6th World Conference on Transport Research, pp.2117-2128.
Kraus, S. (1998). Estimating the length of trunk tours for environmental and cost evaluation of
distribution systems, In: Advances in distribution logistics, (B. Fleishmann et al. eds.)
pp.3 19-339, Springer, Berlin.
Kresge, D.T. and P.O. Roberts (1971). Systems analysis and simulation models, v.2. In:
Technologies of Transport Planning, (J.D. Meyer, ed.), The Brookings Institute,
Washington, D.C.
Kroon, M., J.van Ham and R.Smit (eds.) (1991). Freight transport and the environment
(Studies in Environmental Science 45), Elsevier Science Ltd.
Kurosaki, H., M. Yagi and H. Yokosuka (1993). Vehicle license number recognition system
for measuring travel time. Journal of Robotics and Mechatronics, 5(2), 192-7.
Laakso, J. and H. Rauhamaki (1999). The development of telematics for small and medium
sized trucking enterprises (KANTELE). Proceedings 6th ITS World Congress,
Toronto, (CD ROM).
Laporte, G., F. V. Louveaux and H. Mercure (1992). The vehicle routing problem with
stochastic travel times. Transportation Science, 26, 161-70.
Larraneta, J., J. Munuzuri and G. Montero (1999). Analysis of urban freight requirements and
alternative solutions - an application in Sevilla. In: City Logistics I (E. Taniguchi and
R.G. Thompson, eds.), Institute of Systems Science Research, Kyoto, pp.245-259.
Lawler, E.L., J. K. Lenstra, A. H. G. Rinnooy Kan and D. B. Shmoys (1992). The Traveling
Salesman Problem, Wiley.

(References) 237
Lawler, E.L., J.K. Lenstra, A.H.G. Rinnooy Kan and D.B. Shmoys (eds.) (1985). The
travelling salesman problem-A guided tour of combinatorial optimisation. John
Wiley & Sons, New York.
LeBlanc, L.J., E.K. Morlok and W. Pierskalla (1975). An efficient approach to solving the road
network equilibrium traffic assignment problem. Transportation Research, 9, 309-3 18.
Leonald, D.R., J.B. Tough and P.C. Baguley (1978). CONTRAM, a traffic assignment model
for predicting flows and queue during peek period. TRRL Laboratorj Report, 841, 112.
Li, C., D. Hunt, P. Pretorius and S. Capecci (1999). Deploying a CVO traveler information
system in Arizona. Proceedings 6th ITS World Congress, Toronto, (CD ROM).
Lindenberg, S., F.N. Stokman and I. Gadourek (1983). Modellen in de Sociologie. Van
Loghum Slaterus, Deventer.
Little, J.D.C. (1961). A proof for the queuing formula L=;L W . Operations Research, 9, 383387.
Ma, L. (1999). Modelling for traffic-generated environmental pollution analysis, In: C i 8
Logistics I , 1st International Conference on City Logistics, Cairns (E. Taniguchi and
R.G. Thompson, eds.), Institute of Systems Science Research, Kyoto, pp. 133-146.
Mahmassani, H.S., R. Jayakrishnan and R. Herman (1990). Network traffic flow theory:
microscopic simulation experiments on supercomputer. Transportation Research,
24A, 149-162.
Malandraki, C. and M. S. Daskin (1992). Time dependent vehicle routing problems:
formulations, properties and heuristic algorithms. Transportation Science, 26, 185200.
Maltby, D. ( 1970). Traffic at manufacturing plants. Traffic Engineering Control, 12(2), 72-77.
Marchau, V.A.W.J. (2000). Technology assessment of automated vehicle guidance, prospects
for automated driving implementation. Studies in Systems Engineering, Policy
Analysis and Management, No. 02, Delft, University of Technology, Delft.
Marin, B. and R. Mayntz (1991). Policy Networks: Empirical Evidence and Theoretical
Considerations. Westview, Boulder, CO.
Martin, B.V. and M.L. Manheim, (1965). A research program for comparison of traffic
assignment technique. HRB Record, 88, 69-84.
Matsoukis, E.C. (1986). Road traffic assignment - a review, Part I. non-equilibrium methods.
Transportation Planning and Technology, 11, 69-76.
McKeown, B. and D. Thomas (1988). Q methodology. Sage, Newbury Park.
Meadows, D.L. and J. Randers (1991). Beyond the Limits. Confronting Global Collapse;
Envisioning a Sustainable Future. Earthscan Publications Ltd., London.
Meyberg, A.H. and P.R. Stopher (1974). A framework for the analysis of demand of urban
goods movement. Transportation Research Record, 496,68-79.
Meyburg, A. (1979). The applicability of behavioural modelling to the analysis of goods
movements. In: Behavioural Travel Modelling (D.A. Hensher and P.R. Stopher, eds.),

238 (City Logistics)

Croom Helm, pp.624-635.
Ministry of Construction, Japan (1992). Study Report on Estimating the Effects of Road
Construction, Road Bureau, pp. 41- 45.
Mowshowitz, A. (1994). Virtual organisation: a vision of management in the information age.
The Information Society, 10, 267-288.
Muller, G., (1995). Interinodal Freight Transportation. Lansdowne VA.
Myerson, R.B. (1991). Game Theory: Analysis of Conflict. Harvard University Press,
Cambridge.
Nam, K. (1997). A study on the estimation and aggregation of disaggregate models of mode
choice for freight transport. Transportation Research E, 33(3), 223-23 1,
NEA Transport Research and Training, (1992). TEM 11, Het goederenvervoer in drievoud to
2015, NEA Rijswijk.
NENHaskoning, (1996). Haalbaarheid initiatieven in het gecombineerd weg-watervervoer,
Rijswijk, 1991.Publicatieblad van de Europese Gemeenschappen, Beschikking nr
1692/96/EG van het Europese Parlement en de Raad van 23 juli 1996 beti-effende
communautaire richtsnoeren voor de ontwikkeling van een transeuropees vervoersnet,
September.
Nissan (1998). Automotive Transportation in Perspective, 7(2), Environment and
Transportation Research Laboratory, Nissan Research Centre, Tokyo.
Noritake, M. and S. Kimura (1990). Optimum allocation and size of seaports. Journal of
Watenvay, Port, Coastal and Ocean Engineering ASCE, 116(2), 287-299.
OECD, (1992). Advanced Logistics c4nd Road Freight Transport. OECD Publications and
Information Centre, Paris.
Ogden, K.W. (1977). Modelling urban freight generation. Traffic Engineering and Control,
March, 106-109.
Ogden, K.W. (1978). The distribution of truck trips and commodity flows in urban areas: a
gravity model analysis. Transportation Research. 12, 13 1-1 37.
Ogden, K.W. (1991). Truck movement and access in urban areas. ASCE Transportation
Engineering Journal, 117 (l),71-89.
Ogden, K.W., S.G. Ritchie, W. Young and P.L. Dumble (1981). Analysis and appraisal of
freight generation characteristics. ARRB Research Report, No. 117.
Ooishi, R. (1996). Planning methodology for new freight transport systems in urban areas. PhD
dissertation, Kyoto University. (In Japanese)
Ooishi, R. and E. Taniguchi (1999). Effects and profitability of constructing the new
underground fright transport system. In: City Logistics I , 1st International Conference
on City Logistics, Cairns, (E. Taniguchi and R.G. Thompson, eds.), Institute of
Systems Science Research, Kyoto, pp.303-3 16.
O m a n , I. And J. Kelly (1996). Meta-heuristics: An overview. In: Meta-heuristics: T h e o q and
Applications, (1.Osman and J. Kelly, eds.) Kluwer, Boston, pp. 1-21.
Park, M. and R.L. Smith (1997). Development of a statewide truck-travel demand model with

(References) 239
limited origin-destination survey data. Transportation Research Record, 1602, 14-2 1.
Patriksson, M. (1994). The Traffic Assignment Problem: Models and Algorithms. VSP, Utrecht,
The Netherlands.
Pelensky, E.E., W.R. Blunden and R.D. Munro (1968). Operating costs of cars in urban areas.
Proceedings of the 4"'Corference, Australian Road Research Board, 4(1), 475-504.
Picard. G. and M. Gaudry, (1998). Exploration of a Box Cox logit model of intercity freight
mode choice. Transportation Research, 34E(1), 1- 12.
Pidd, M. (1996). Toolsfor thinking - nzodelling in management science. Wiley, Chichester.
Potvin, J.Y. (1996). Genetic algorithms for the traveling salesman problem. Annals of
Operations Research, 63, 339-70.
Psaraftis, H.N., (1995). Dynamic vehicle routing: Status and prospects, Annals of Operations
Research, 61, 143-64.
Ramathian, S . and C. Hiatt, (1989). Towards an Information-Based Theory of Irrational
Systems Behavior. Systems Research, 6, 7-16.
Reeves, C. (ed.) ( 1 993). Modern heuristic techniques for combinatorial problems. Halsted
Press, N.Y.
Research Institute of Innovative Technology for the Earth (1994). Report of the Study and
Research Committee for Technology for the Greening of the Desert as a Way to Deal
with COz, p. 98.
Rimmer, P.J. and J.A. Black (1981). Urban goods and commercial vehicle movements in
Sydney. Australian Road Research, 11(4), 15-29.
Roberts, P.O. and B.C. Kullman (1979). Urban goods movement: behavioural demandforecasting procedures. In: Behavioural Travel Modelling, (D.A. Hensher and P.R.
Stopher, eds.), Croom Helm, pp. 553-76.
Roumegoux, J.P. et al. (1979). Consommation d'energie par la circulation routiere. Institute de
Recherche des Transport-CERNE, Note d'Information, 14, Lyon.
Ruske, W. (1994). City Logistics - Solutions for urban commercial transport by cooperative
operation management. OECD Seminar on Advanced Road Transport Technologies,
Omiya. Japan.
Samuelson, P.A. (1952). Spatial price equilibrium and linear programming. American
Economic Review, 42, 283-303.
Sawaragi, Y., H. Nakayama, and T. Tanino (1985). Theory of Multiobjective Optimization.
Academic Press.
Schaffer, J.D. (1985). Multiple objective optimization with vector evaluated genetic
algorithms. Proceedings of the First International Conference on Genetic Algorithms
and Their Applications, pp.93-100.
Sheffi, Y. (1 985). Urban transportation nehvorks. Prentice Hall.
Sijbrands, M.J.C (1993). Strategische en Logistieke Besluitvorming. Tinbergen Research
Institute, nr. 53, Amsterdam.

240 (City Logistics)

Southworth, F, Y.J. Lee, C.S. Griffin And D. Zavattero (1983). Strategic motor freight
planning for Chicago in the year 2000. Transportation Research Record, 920,45-48.
Starkie, D.N.M. (1971). Review of industrial demand for transport. Journal of Transport
Economics and Policy, 5, 224.
Steenbrink, P.A. (1974). Optimization in transportation network. Wiley, New York.
Stern, E., J. Tzelgov and H. Avishai (1983). Driving efforts and urban route choice. The
Logistics and Transportation Review, 19 (l),67-77.
Sugden, R. and A. Williams (1978). The principles ofpractical cost-benefit analysis. Oxford.
Suzuki, K. (1999). Car navigation based car computer for commercial vehicles. Proceedings
6th ITS World Congress, Toronto, (CD ROM).
Taillard, E.D., P. Badeau, M. Gendreau, F. Geurtin and J.Y. Potvin (1997). A Tabu search
heuristic for the vehicle routing problem with soft time windows. Transportation
Science, 31, 170-86.
Takahashi, Y., K. Ikenoue, K. Yasui and Y. Kunikata, (1996). Travel-Time-Information
System (TTIS) in Mie Prefecture. Proceedings 3'" ITS World Congress, ITS America,
Orlando (CD ROM).
Takayama, T. and G, Judge, (1964). Equilibrium among separated markets; A reformulation.
Econometrica, 32, 5 10-524.
Takayama, T. And W.C. Labys, (1986). Spatial equilibrium analysis. In: Handbook of Regional
and Urban Economics I (P. Nijkamp, ed.), Elsevier, Amsterdam.
Taniguchi, E. and K. Sawada (1 979). Attenuation with distance of traffic-induced vibrations.
Soils and Foundations, 19(2), 15-28.
Taniguchi, E., T. Yamada and T. Yanagisawa (1995). Issues and views on co-operative freight
transportation systems. 7th World Conference on Transport Research, Sydney.
Taniguchi, E., T. Yamada, M. Tamaishi and M.Noritake (1998). Effects of designated time on
pickup/delivery truck routing and scheduling, In: Urban transport and the
environment for the 21" century IV (C. Borrego and L.J. Sucharov, eds.), WIT Press,
Southampton, pp. 127.136.
Taniguchi, E., R.G. Thompson and T. Yamada (1999a). Modelling city logistics. In: CiQ
Logistics I (E. Taniguchi and R.G. Thompson, eds.), Institute of Systems Science
Research, Kyoto, pp.3-37.
Taniguchi, E., M. Noritake, T. Yamada and T. Izumitani (1999b). Optimal size and location
planning of public logistics terminals. Transportation Research, 353(3), 207-222.
Taniguchi, E., T. Yamada and T. Hosokawa (1999~).Dynamic traffic simulation with optimal
truck routing and scheduling, Transportation and Traffic Theory, Abbreviated
Presentation Sessions of the 14"' International Symposium on Transportation and
Traffic Theory, Jerusalem, pp.419-439.
Taniguchi, E., T. Yamada and T. Takauchi (1999d). Stochastic vehicle routing and scheduling
using a heuristic approach. In: Urban transport and the environment for the 2 I y r
century V (L.J. Sucharov, eds.), WIT Press, Southampton, pp. 127-136.

(References) 241
Taniguchi. E., T. Yamada and D. Tamagawa (1999e). Modelling advanced routing and
scheduling of urban pick-up/delivery trucks, Proceedings 6th ITS World Congress,
Toronto (CD ROM).
Taniguchi, E. and R.E.C.M. van der Heijden (2000a). An evaluation methodology for city
logistics. Transport Reviews, 20(1), 65-90.
Taniguchi, E., T. Yamada and Y. Kakimoto (2000b). Probabilistic vehicle routing and
scheduling with variable travel times. 9"' ZFAC Symposium on Control in
Transportation systems 2000 (E. Schnieder and U. Becker, eds.), Braunschweig,
pp.45-50.
Taniguchi, E., S . Kawakatsu and H. Tsuji (2000~).New co-operative system using electric
vans for urban freight transport. In: Urban transport and the environmentfor the 21"
century VZ (L.J. Sucharov and C.A. Brebbia, eds.), WIT Press, Southampton, pp.201210.
Taylor, M.A.P., W. Young and P. W. Bonsall, (1996). Understanding Traffic Systems: Data,
Analysis and Presentation. Avebury, Aldershot.
Taylor, S . and K. Button (1999). Modelling urban freight: what works, what doesn't work? In
City Logistics I , 1st International Conference on City Logistics, Cairns, (E. Taniguchi
and R.G. Thompson, eds.), Institute of Systems Science Research, Kyoto, pp. 203218.
The Japan Synthetic Institute (1995). Research for Supplementing the Integrated Calculations
of the Environment and the Economy to the System of Calculating the National
Economy. pp. 93 - 99.
Thomas, W.I., (1966). d n social organization and social personalitv, Chicago, IL.
Thompson, R.G., Y. Wang, and I. Bishop (1999). Stochastic vehicle and routing using GIs.
Proc. International Conference on Computers in Planning, Venice.
Toth, P. and D. Vigo (1998). Exact solution of the vehicle routing problem, Chaprter I . In:
Fleet Management and Logistics (T.G. Crainic and G. Larporte, eds.). Kluwer, pp. 132.
Traffic Bureau, National Police Agency (1998). ITS developed by Japanese Police, Institute of
Urban Traffic Research, Osaka.
Traffic Planning Council for Kyoto, Osaka and Kobe Area (1997). Survey on comprehensive
urban traffic system in Kyoto, Osaka and Kobe area. (In Japanese)
Tsitsiklis, J. (1992). Special cases of the traveling salesman and repairman problem with time
windows. Networks, 22, 263.
Tsukaguchi, H. and U. Vandebona (1999). Analysis on dynamic scheduling of freight
collection vehicles in urban centres. In: City Logistics I , 1st International Conference
on City Logistics, Cairns, (E. Taniguchi and R.G. Thompson, eds.), Institute of
Systems Science Research, Kyoto, pp.101-116.

242 (City Logisticsj

Twist, M.J.W. van, P.W.G. Bots, J.H.R. van Duin (1998). Dynamic Actor Network Analysis on
the application of policy technology to institutional design. Paper presented at the
Netherlands Institute of Government Conference, Oisterwijk.
United States General Accounting Office, (1 992). Intermodal Freight Transportation:
Combined Rail-Truck Service Offers Public Benefits, but Challenges Remain.
(GAO/RCED-93-16), Washington DC.
Vincent, E. (1986). Vehicle classification and mass data collection systems, Proceedings 13"'
ARRB Confercnce, 13 (S), 96-108.
Visser, J.G.S.N. (1997). Underground networks for freight transport: a dedicated infrastructure
for intermodal short-distance freight transport. Proceedings of 3"d TRAIL Year
Congress on Transport, Infrastructure and Logistics, Delft, The Netherlands.
Vreede, G.J. de, R.O. Brigss, R. van, Duin and R. Enserink (2000). Atheletics in Electronic
Brainstorming: Asynchronous Electronic Brainstorming in Very Large Groups. Paper
submitted to HICSS-33, Maui.
Wang, R., (1998). A real time fleet management via GIS/GPS platform, Proceedings 5"' ITS
World Congress, Seoul, VERTIS. (CD ROM)
Wang, Y., R.G. Thompson and I. Bishop (1999). A GIS based information integration
framework for dynamic vehicle routing and scheduling. Proceedings IEEE
International Vehicle Electronics Conference (IVEC'99), IEEE Industrial Electronics
Society, Catelog Number 99EX257, Changchun, China, 474-9.
Watson, P.L. (1975). Urban goods movement: a disaggregate approach. C.D. Heath,
Lexington.
Wild, D. and K. Moller (1997). A set of freight telematics applications in City Logistics and
freight centres: SURFF. Proceedings 4"' ITS World Congress, Berlin. (CD ROM)
Wild, D. and K. Moller (1998). Experiences with the initiation, installation and integration of
telematic applications for city logistics and freight centres, Proceedings f h ITS World
Congress, Seoul. (CD ROM)
Wildavsky, A,, (1993). Speaking truth to power; the art and craft of policy analysis.
Transaction Books, New Brunswick.
Willemse, H., (1999). Manager logistics grocery Albert Heijn Zaandam, Interview at congress
The Hague www.psd-online.nl , 8 april.
Winston, C. (1983). The demand for freight transportation, models and applications.
Transportation Research, 17A, 236-3 11.
Yamada, H. (1980). Fuel consumption of automobiles: Experiment and analysis of results.
Civil Engineering Journal, 14(2). (In Japanese)
Yamada T., Taniguchi, E., Noritake, M. and Horie, A. (1999). Attitudes of companies towards
introducing co-operative freight transport systems. In: City Logistics I , 1st
International Conference on City Logistics, Cairns, (E. Taniguchi and R.G.
Thompson, eds.), Institute of Systems Science Research, Kyoto, pp.219-232,.

(References) 243

Yamamura, Y., S . Iwasaki, J. Matoba and F. Kogure (1998). Test operation of traffic
management in the ETC system of Japan. Proceedings 5 ITS World Congress, Seoul.
(CD ROM)
Yeomans, G.S. and Bake M.C. (1992). Towards a NSW urban freight transport model.
Proceedings 161hARRB Conference, Perth, 7,7144.
Yoo, T. (1998). Study on the real time logistics control system using wireless data
communication network. Proceedings SrhITS World Congress, Seoul. (CD ROM)
Young, W., M.A.P. Taylor and P. Gipps (1988). Microcomputers in Traffic Engineering,
Research Studies Press, Taunton.
Zlatoper, T.J. and Z. Austrian (1989). Freight transportation demand: a survey of recent
economic studies, Transportation, 16, 27-46.

SUBJECT INDEX (by chapter number)

Index Terms

Links

A
accessibility

activity-based-costs-model

actors

administrators

advanced information system

air pollution

all-or-nothing traffic assignment

alternative transport system

APS

arena

arrival rate

attenuation with distance by diffraction

attenuation with distance of vibration

attributes of the modes

automated vehicle guidance

auxiliary problem

average number of trucks

average travel speed

AVG

This page has been reformatted by Knovel to provide easier navigation.

Index Terms

Links

B
batch-production

benefit cost ratio

berth

berth cost

berth-truck cost ratio

BOX model

Box-Cox logit model

branch-and-bound method

building blocks

Bureau of Public Roads model

C
carriers

causal diagram

centroid

chance constrained model

characteristics of goods

chromosome

City Logistics initiatives

City Logistics measures

City Logistics schemes

CO2 emissions

cognitive mapping

combinatorial optimisation problem

combined choice model

This page has been reformatted by Knovel to provide easier

navigation.

Index Terms

Links

computer programming

conceptual modelling

configuration theory

congestion

constraint

construction effect

continuous location model

cooling parameter

co-operative freight transport system

correlation coefficient

cost-benefit analysis

costs of travel time

crash

crash-related benefit

criteria

cross-docking

crossover

cultural theory

customer

D
data

data collection

database

decision maker

decision making

decision support model

This page has been reformatted by Knovel to provide easier

navigation.

Index Terms

Links

decision variable

delay penalty

demand function for flow of trucks

demand model

departure distribution

depot

descriptive modelling technique

deterministic user equilibrium

diesel-engined vehicle

discrete choice model

discrete location model

district distribution shop

dominant frequency of the ground

double stacking

doubly constrained combined

distribution-assignment model

dual-mode truck

dynamic actor network analysis

dynamic behaviour

dynamic traffic assignment model

dynamic traffic information system

dynamic traffic simulation

E
early arrival penalty

e-commerce

econometric model

This page has been reformatted by Knovel to provide easier

navigation.

Index Terms

Links

economic agent

electronic data interchange

electronic toll collection

elite individual

EMME/2

endogenous variable

energy consumption

energy consumption model

engine map

environment

environmental model

environment-related benefit

equal travel time principle

Erlangian distribution

ERP

evaluation

evaluation criteria

exogenous variable

experimental design

exponent value

exponential distribution

F
facility cost

facility location model

facility location problem

feasibility

This page has been reformatted by Knovel to provide easier

navigation.

Index Terms

Links

financial model

fitness

fixed cost

forecasted model

four-stage model

Frank-Wolfe algorithm

free travel time

freight carrier

freight cost

freight generation model

freight network equilibrium model

fuel consumption

future logistics

G
game theory

gantry crane

gate

gene

generalised cost function

generation rate

genetic algorithms

genetic operator

genotype

Global Position System

golden section search method

GoodTrip model

8
8

This page has been reformatted by Knovel to provide easier

navigation.

Index Terms

Links

graphical interface

gravity model

greenhouse gas

group support system

growth-factor method

gut feelings

H
Hamiltonian circuit

heuristic technique

high-speed rail connection

hub-and-spokes-network

hub-terminal

I
impact model

impedance function

implementation

incremental assignment method

individual

information

information communication technology service

information exchange

initial investment

input-output analysis

input-output table

instability

This page has been reformatted by Knovel to provide easier

navigation.

Index Terms

Links

integer programming

interdependence

interest rate

intermodal agents

intermodal transportation system

internal rate of return

in-vehicle navigation system

ITS

J
Junctions

just-in-time

just-in-time distribution

just-in-time transport

K
Kendalls notation

L
land use

linear programming

linear programming model for terminal assignment

line-haul

link function

link performance function

Littles formula

This page has been reformatted by Knovel to provide easier

navigation.

Index Terms

Links

liveability

load factor

load factor controls

loading/unloading terminal

loading-dock

local optimal solution

local pick-up/delivery

logistical chain

logistics city park

logistics process

logistics terminal

logit model

lognormal distribution

long distance haulage

lower level problem

LTL-deliveries

M
MEWS () model

M/M/S () model

macroeconomic model

macroscopic model

macro-simulation model

mathematical programming

measures of regulation

meta-heuristic method

meta-modelling

This page has been reformatted by Knovel to provide easier

navigation.

Index Terms

Links

microscopic model

micro-simulation model

modal shift

mode selection

model

model validation

model verification

modelling actors perceptions

modelling framework

monitoring

multi-actor planning system

multi-criteria analysis

multi-disciplinary approach

multi-mode and multi-product assignment model

multi-objective optimisation model

multi-objective programming method

multiple regression

mutation

N
neighbourhood solution

net present value

NETSIM

network

network analysis

network characteristics

network location model

This page has been reformatted by Knovel to provide easier

navigation.

Index Terms

Links

node function

noise

non-linear programming problem

non-linear regression

NOx

O
objective

objective function

operation cost

optimal location model

optimum number of berths

P
Pareto optimal solution

p-centre problem

penetration rate

perception based modelling

performance indicator

performance measures

pickup/delivery truck

pluriformity

p-median problem

Poisson distribution

policy measures

present value of benefits

probabilistic model

This page has been reformatted by Knovel to provide easier

navigation.

Index Terms

Links

problem definition

production constrained gravity model

profitability

proportion of large vehicles

public assistance

public logistics terminal

Q
Q-method

queuing theory

R
rational planning

recourse model

regression model

reliability

reproduction

residents

resources

road structure

rolling resistance

route choice model

route guidance

This page has been reformatted by Knovel to provide easier

navigation.

Index Terms

Links

S
scripts

self-containment

semaphores

sensitivity analysis

service rate

shippers

shuttle-service

simple regression

simulated annealing

simulation model for terminal operations

slots

social discount rate

societal organisation

soft-OR

spatial price equilibrium model

SPM

stack

stakeholders

stochastic model

stochastic programming

stochastic user equilibrium

straddle carriers

supplier-customer relationship

supply model

surveys

sustainable

5
2

This page has been reformatted by Knovel to provide easier

navigation.

Index Terms

Links

system boundaries

systems analysis

systems synthesis

T
tabu list

tabu search

tabu tenure

tele-shopping

three-stage modelling approach

throughput

time window

time-related benefits

total cost

traffic assignment

traffic assignment model

traffic flow model

traffic intensity

traffic market

traffic simulation model

trajectory diagram

transhipment

Transport European Networks

transport market

transport need market

transportation cost

transportation network modelling

This page has been reformatted by Knovel to provide easier

navigation.

Index Terms

Links

transportation problem

transportation systems view

travelling salesman problem

travelling salesman problem with time windows

travel-related benefits

U
underground distribution system

underground freight transport system

underground transportation

upper level problem

urban traffic problem

user equilibrium traffic assignment

utilization

utilization factor

V
validation

value of stochastic solution (VSS)

vector evaluated genetic algorithms

VEGA-based optimal location model

vehicle routing and scheduling problem

vessel

vibration

virtual freight centres

virtual freight companies

virtual organisations

This page has been reformatted by Knovel to provide easier

navigation.