Experiences from clp applications in Microscopic Transport Modeling

Comparison of both Conceptual Approaches

Download 105.3 Kb.
Size105.3 Kb.
1   2   3   4   5   6

5Comparison of both Conceptual Approaches

Though both models under consideration rely on constraint solving engines, they follow a different philosophy, mainly due to their different model objects – activity scheduling and transport planning. In a compact way, the following tables compare the two modeling approaches with respect to the underlying systems’ properties.

Table 2: Comparison of the passenger and freight model – Part 1


CLP Travel supply-demand expert system

INTERLOG Freight simulation model

Fundamental Properties of the mapped system

Linear relationship between individual behavior patterns and the aggregate demand system behavior.

Network and scaling effects and resulting non-additivity. // Demand of commodity supply deduced from production statistics.

Function of CLP

  1. Mapping of restricted choice sets of actors

  2. Assurance of consistency between micro behavior and system transitions

  3. Questioning for hypothetical service configurations

  1. Mapping of restricted choice sets of actors,

  2. mapping of combinatorial decision making


Household groups (public sector / transport companies so far exogenously determined)

Shippers, recipients, forwarders, hauliers

(physical infrastructure first exogenous)

Primary Objective

Understanding and prediction of short-/ medium-term travel demand behavior

Understanding and prediction of short/ medium-term behavior of freight transport and logistics systems.



Inter-city / supra-regional trip-making

Inter-regional road freight transports

Table 3: Comparison of the passenger and freight model – Part 2


CLP Travel supply-demand expert system

INTERLOG Freight simulation model

System’s representation



Description of the Time-space problem

Households’ activity scheduling problem with:

  1. Budget / Capability Limits

  2. Agenda of required Trips

  3. Medium-/Long-Term decisions

  4. Need maximize to utility

En-tour pickup and delivery problem with time windows. “Hard” constraints (time-windows, frequency etc.) can change during simulation.



Short-term: activity scheduling // Medium-term: utility maximization of the expected program

Short-term: transportation cost minimization // Medium-term: total logistics cost minimization

Model dimensions

Time, money, trip purpose, time slice, trip party size, mode and route choice options, trip frequency

Time, length, load meters, weight, packaging platform type, packaging type, relationship network configuration

Model’s time steps

Week and year and five-year period

Days, weeks and years

„Agreement Constraints“ between model-led variables and observations

Security “pegs”, taking into account some 15 to 20 per cent statistical errors of the data samples.

No used at an aggregate level. Only considered at a microscopic operational level, problem is considered as secondary, especially because of meso-structures in freight transport prevailing from simple micro-macro aggregation.

Sequence implementation

Location patterns supposed as exogenously fixed at short- and medium-term planning

Location patterns and freight flows supposed as exogenously fixed. Logistics planning interactive.

Actor-network interaction

Evaluation of price and service levels through optimal path search and computation of expected personal utility; Link load pattern determine transfer times

Individual paths between loading and unloading points are stored at the network level. Paths and relating constraints can be updated from time to time.

Model states after heuristic domain reduction

According to the model design, long-term decisions are fixed for medium-term decision-making; and medium-term decisions are fixed for activity scheduling on a short-term basis;

Pricing and taxation policy applied and

infeasible mobility pattern are ruled out

Temporally fixed states:

Warehouse policy by producer-recipient relationship, awarded forwarder by producer-recipient relationship

States changing on a daily basis:

Assignment from orders to tours and variables describing tours (time windows, sequences, weight, volume ...).

Dealing with the underdetermination

in the problem formulation

Additional heuristics restricting the domains of the constraint variables resulting in system fluctuations. Some variables are not necessarily completely determined to one value. The microscopic underdetermination and the fluctuations are accepted as long as certain statistical reliability is reached at an aggregate level.



In the late 1990ies, two branches of transport modeling have crystallized which are now used in parallel: The more macroscopic, sequential approach using results from regression analysis and statistical simulation techniques mapping the behavior of synthetic individuals.

We demonstrated that Constraint Logic Programming (CLP) is a promising concept to re-unify these two approaches: A CLP approach allows for an increase of the models’ explanatory power by adding and coupling further resolution levels – i.e. to provide a meso- and micro-foundation of traditional models – while guaranteeing consistency – both internally and externally, i.e. the agreement with observations and factors of influence.

In the passenger demand modeling sphere, the aggregation of individual trips of household (types) to O&D flows is rather straightforward, whereas in freight demand modeling a consolidation process at the level of logistics centers, forwarding agencies and carriers is needed in order to derive a resulting vehicle flow.

This fundamental difference between these two types of transports is reflected in the design of the passenger and freight models: Whilst the passenger model focuses on the micro-macro consistency, the freight model also addresses detailed requirements of freight logistics through the employment of constraint rules and maps the solution combinatorial planning problems at a micro-scale.

In effect, constraints become an attempt at capturing and explaining choice behavior phenomena in both passenger and freight transport. The study demonstrates that several well-known modeling approaches can be translated into the Constraint Logic Programming language. It can be stated that a declarative re-writing as CLP-based models generalizes existing transport models. Even probabilistic choice behavior can be incorporated for instance by approximating Logit curves. Both models use CLP-based decision engines to map individual behavior. Until the first publication (HEINITZ and LOCK 1998), the employment of constraint solving techniques for travel behavior modeling could not be encountered in the literature so far.


Having itemized the advantages of CLP, this approach has still a number of downsides which need to be addressed:

a) Speed: Solving of CLP problems is often a time-consuming task. In the absence of dedicated pruning algorithms, a near exhaustive search is required. Depending on the formulation of search goals, the solution process may be “misdirected” to “disfavorable” regions of the search space. Furthermore, the management of the domains of the constraint variables excessively consumes memory and CPU resources. On the other hand, CLP is most suited for fast feasibility checks and ad-hoc solutions, which can be improved and refined by meta-heuristics such as, for instance, genetic or learning algorithms. State-of-the-art optimization applications using CLP therefore do not only rely on declarative programming and propagation algorithms, but also integrate elements of imperative, object-oriented and linear programming. Through these meta-heuristics, behavior of individuals can be mapped in the most “reality-like” fashion.

b) Under- / Overdetermination: The constraint-solving process narrows the domains of decision variables inasmuch as the current set constraints is satisfied, resulting in an “uncertainty principle”: An ambiguous set of solutions may remain. As a consequence of this, aggregated indicator variables may become fuzzy intervals instead of being pinpointed to a specific value. With just one inconsistency, the constraint solving fails – but the detection of the inconsistency in the model is a problem for itself. There is no a-priori solution path, which avoids overdetermination. As a way out, the restrictions are stepwise imposed to the variables and intermediate results are stored. This procedure is slowing down the modeling progress considerably. The employment of a meta-framework of automated testing and backtracking and / or partial constraint satisfaction would be useful.

c) Data-hungriness: In contrast to the well-established functional models, the declarative model is knowledge-intensive – and therefore data-hungry. The typical lack of empirical data is even more dramatic when using CLP. Domain knowledge for new constraint relations has to be acquired e.g. from cross-sectional and longitudinal surveys such as the INVERMO project (Chlond 2000). In passenger transport this has proved to be far easier than in freight.

6.3Summary and Outlook

CLP is a promising way to overcome the counter-productive conventional forecasting paradigm and to offer new possibilities to formulate transport models in a more consistent way. CLP can thereby help to solve some classical problems in transportation modeling. CLP as a modeling technique allows the expression of meaningful constraints assuring micro and macro consistency as well as restricted choice sets of individuals. The search for solutions is then conducted through decision engines that are disassociated from the problem statement giving a behavioral foundation of the computed microscopic and mesoscopic system states.

Though the spatial-temporal freight transport problem has a different nature than its passenger travel counterpart, these two specimen of successful model implementation for both cases have been provided in this overview. These models shall not be regarded as a new third alternative to sequential macroscopic models or to simulation approaches. Attempting to generalize the sequential and “compartmentalized” methods, this approach rather allows for the pragmatic incorporation and combination of methodological elements of both of them –while assuring a certain degree of consistency between the macroscopic, disaggregate and microscopic levels. CLP thereby answers a current challenge in transportation modeling and bridges a widening gap between the more and more ambitious scientific models and rather pragmatic needs in decision support to both transport companies and authorities. In fact, it is also a proposition to overcome the noticeable division into a more “econometrics-oriented” and a “simulation-oriented” research stream – as practical CLP model applications have been freely combining these methods for decades.


Arentze and Timmermans 2000 Arentze, T.; Timmermans, H.: ALBATROSS: A Learning-Based Transportation Oriented Simulation System. Eindhoven: European Institute of Retailing and Services Studies Eindhoven, 2000.

Axhausen 1992 Axhausen, K.W. and Gaerling, T.: Activity-based Approaches to Travel Analysis. Conceptual Frameworks, Models and Research Problems. In: Transport Reviews 12, pp.323-341, 1992

Axhausen 1998 Axhausen, K.W.: Some trends in travel behaviour research and modelling (Editorial), The Elsevier Science Transportation Newsletter, p.5, Amsterdam, 1992

Bayliss 1970 Bayliss, B.L., Methodische Probleme von Verkehrsprognosen. Vortraege und Studien aus dem Institut für Verkehrswissenschaft an der Universität Muenster (Westf.), 10, Vandenhoek & Ruprecht, Goettingen, 1970

Ben-Akiva 2004 BEN-AKIVA, M.: Choice analysis and market research for freight. – Lecture at the colloquium “Recent Developments in Transport Modelling” at the University of Antwerp, March 2004

BVU et al. 2001 BVU ; IFO ; ITP ; PLANCO: Verkehrsprognose 2015 für die Bundes-verkehrswegeplanung. 2001. – Project on behalf of the German Ministry for Housing and Transport Berlin/Bonn

Chlond 2006 Chlond, B., Last, J., Manz, W.,  Zumkeller, D.: Long-Distance Travel in a Longitudinal Perspective: The INVERMO Approach in Germany- 2006 - Conference paper presented at TRB Annual Meeting January , Washington DC

Chlond 2000 Chlond, B., Manz, W.: INVERMO - Das Mobilitatspanel fur den Fernverkehr . In: "Dynamische und statische Elemente des Verkehrsverhaltens – Das Deutsche Mobilitätspanel“, Series B 234, pp. 203-227, ISBN 3-933392-34-9

Fruehwirth 1997 Fruehwirth, T. and Abdennadher, S.: Constraint Programmierung. Springer Verlag Berlin, 1997

Gaudry 1998 Gaudry, M., Heinitz, F., Last, J., Mandel, B. Methodological Developments within the Quasi-Direct Format Demand Structure: the Multicountry Application for Passengers MAP-1, CRT Publication #98-46, Université de Montréal, 1998

Haegerstrand 1970 Haegerstrand, T. What About People in Regional Science? In. Regional Science Association Papers 24, pp.7-21, 1970

Himanen 1997 Himanen, V., Jaervi-Nykaenen, T., Raitio, J. Daily Travelling Viewed by Self-Organizing Maps. In: Himanen,V., Nijkamp, P. et al.: Neural Network Applications in Transport. Aldershot, Avebury, 1997

Ilog 1998 ILOG SA Ilog Solver, User Manual., Gentilly, France, 1998

Infratest 1996 Infratest Burke Verkehrsforschung. Ergebnisbericht Mobility ’95. Eine Analyse des deutschen Fernreisemarktes. Munich, 1996

Jones 2003 JONES, P.: The Formation of an Activity-Based Approach to Travel Choice. 2003. – URL http://www.tsu.ox.ac.uk/seminars/festschrift/index.html . – Lecture at the 30th Anniversary Festschrift -Wednesday 15th October 2003, St. Anne’s College, Oxford

de Jong et al. 2002 JONG, G. de ; GUNN, H. ; WALKER, W. ; WIDELL, J.: Study on Ideas on a New National Freight Model System for Sweden. RAND Corporation; Santa Monica, 2002. – Prepared by RAND Europe and TRANSEK AB for the SAMGODS Group. – ISBN 0-8330-3327-1

Lock 1997 Lock, H.C.R. (1997) Documentation for the PlaCo Solver Library. Institut für Programmstrukturen und Datenorganisation, Universität Karlsruhe, 1997

Mandel 1997 Mandel, B., Gaudry, M.J., Rothengatter, W. A Disaggregate Box-Cox Logit Mode Choice Model of Intercity Passenger Travel in Germany and its Implications for High Speed Rail Demand Forecasts. The Annals of Regional Science 31, 2, pp.99-120, 1997

Naeger 1996 Naeger, G. (1996) Constraint-Netze als Grundlage für Bottom-up Planungsansaetze in der Unternehmensmodellierung Infix Verlag, Sankt Augustin, 1997

Ortùzar and Willumsen 1990 ORTÙZAR, J. de D. ; WILLUMSEN, L.G.: Modelling Transport. John Wiley & Sons; Chichester, 1990

Recker 1986 Recker, W.W., McNally, M.G. and Root, G.S., A Model of Complex Travel Behavior. In: Transportation Research A 20A (4), pp.307-318 / 319-330, 1986

Sano and Wisetjndawat 2004 SANO, K. ; WISETJNDAWAT,W.: A Behavioral, Micro -Simulation Model to Evaluate Policies on Urban Freight Transportation. In: Proceedings of the 10th World Conference on Transport Research in Istanbul (2004)

Stefik 1981 Stefik, M. Planning with Constraints (MOLGEN: Part I), In: AI Journal 16, pp. 111-139, 1981

Williams 2005 WILLIAMS, I.: EUNET2.0 – A spatial I-O and freight logistics model – Presentation at the “MODEM” workshop organized by NESTEAR on behalf of the French Transport Ministry in Paris, November 2005

1 The concept of meso-structures describes the fact, that transport logistics networks combine shipments from different shipper-recipient relationships on one tour. The emergence of hub&spoke systems could be considered a extension of this concept. The concept of meso-structures originated from Prof. Sjöstedt.

Download 105.3 Kb.

Share with your friends:
1   2   3   4   5   6

The database is protected by copyright ©ininet.org 2020
send message

    Main page