Computational methods for multi-agent simulations of travel behavior

Kai Nagel, ETH Zürich
Fabrice Marchal, ETH Zürich

Resource paper
Workshop on Computational Techniques

Moving through nets: The physical and social dimensions of travel 10 $^{\hbox{\small th}}$ International Conference on Travel Behaviour Research Lucerne, 10-14. August 2003

Computational methods for multi-agent simulations of travel behavior

Kai Nagel
Department of Computer Science
ETH Zürich
CH-8092 Zürich, Switzerland

$\parbox{0.5in}{Phone:}$ +41 (0)1 632 27 54
$\parbox{0.5in}{Fax:}$ +41 (0)1 632 13 74
$\parbox{0.5in}{eMail:}$ nagel@inf.ethz.ch

Fabrice Marchal
Computational Laboratory (CoLab)
ETH Zürich
CH-8092 Zürich, Switzerland

$\parbox{0.5in}{Phone:}$ +41 (0)1 632 56 79
$\parbox{0.5in}{eMail:}$ marchal@inf.ethz.ch

Abstract

Travel behavior research is, by definition, concerned with the behavior of travelers. In order to make those research results useful for policy decisions, often the behavior of many travelers needs to be considered in conjunction. This can either be achieved by models which look at groups of travelers, or by models which look at individual travelers. This paper is concerned with the latter, and how such models of individual travel behavior research can be combined into a model-based transportation forecasting tool. The method to achieve this is called multi-agent simulation.

This paper looks at multi-agent simulation for travel behavior research first from a modeling and then from a computational perspective. The modeling part discusses the main issues mobility simulation, strategy generation, learning/feedback, and initial/boundary conditions. The computational part then investigates how those modeling approaches can be implemented on existing computer architectures. Special emphasis is put on the interoperability between modules coming from different authors, and on parallel and distributed computing in order to achieve computability of large scenarios.

The strong emphasis on the computational aspects is related to the belief that modeling and implementation are not independent: Certain models are easier to implement than others; and most implementation decisions eventually favor some models over others.

Keywords

multi-agent simulation; agent-based modeling; traffic simulation; parallel computing; distributed computing; travel behavior International Conference on Travel Behaviour Research, IATBR