An Agent Based Modelling System for Activity-based Models



An Agent Based Modelling System for Activity-based Models

Authors

POLAK J and HUANG X, Centre for Transport Studies, Imperial College, UK

Description

This paper presents an agent-based and activity-based simulator (called MATS) for simulating travellers' behaviour in a transport network where Advanced Transport Telematics (ATT) are available and network and facility pricing policies are applied.

Abstract

This paper presents an agent-based and activity-based simulator (called MATS) for simulating travellers' behaviour in a transport network where Advanced Transport Telematics (ATT) are available and network and facility pricing policies are applied.

Multi-agent systems (MAS) provide a natural means of developing simulation models of complex interacting systems. MAS are concerned with coordinating intelligent behaviour among a collection of autonomous intelligent "agents", that may represent, for example, individual travellers, information and pricing systems, infrastructure elements and so on. The use of MAS technology allows simulations to be developed by concentrating on the behaviour of individual agents and their interactions, avoiding much of the procedural programming overhead associated with more conventional simulation modelling approaches. Agent based simulations have considerable appeal for the modelling of activity scheduling behaviour, especially the complex interactions that can arise in task and resource allocation and consequent travel behaviour of multiple person households.

MATS is being implemented in the SWARM multi-agent environment, developed at Santa Fe Institute. In MATS, there are three principal classes of agents; travellers, networks, and ATTs. Each agent has its own private data and corresponding methods for processing these data. This enables us to represent heterogeneous agents (e.g., utility maximisers and satisfiers) in a consistent manner within a single system. MATS has been developed so as to be independent of any particular travel behaviour model; rather it provides a framework in which a wide variety of substantive activity-based behavioural models may be programmed and simulated. This feature has important advantages including greater flexibility in the representation of population heterogeneity and ease of comparing new and established models.

In this paper, we outline the principal elements of MATS and describe its application to the development of models in two domains (i) network behaviour in the presence of variable message sign information and (ii) car allocation in multiple person households.

Publisher

Association for European Transport