Agent-Based Modeling for Carpooling

Introduction

Modeling activities and travel for individuals in order to estimate traffic demand, leads to large scale simulations. Most current models simulate individuals acting in a mutually independent way except for the use of the shared transportation infrastructure (road and transit capacity use). In many cases those tools scale (i) linearly with respect to daily agenda generation for mutually independent agents and (ii) quadratically with the number of travel analysis zones (TAZ) due to the use of origin-destination (OD) pair based flow and impedance matrices.

As soon as cooperation between individuals is accounted for, the need for agent-based simulation arises because (i) on one hand individuals have their own goals and plans and (ii) on the other hand they need to communicate, coordinate and negotiate to achieve their goals. This results in rapidly increasing problem complexity (combinatorial explosion). Agent-based simulations are known not to scale well. However, some problems raise questions that can be answered using this technique only.

This chapter focuses on the specific transportation related problem of carpooling for commuters. This case has been chosen because the study of scalability questions requires a well defined problem (and application context) in order to apply the results of mathematical and computer science research. The concept of shared trip execution is diverse. First the context is sketched both by analysing the essential phenomena relevant to cooperation and by showing a selection of existing advisory services available to candidate carpoolers today. Then, the principle of operation for a commuter trip matching advisory service is described. In order to analyse the behaviour of such advisor, it can be exercised by virtual community implemented by an agent-based simulator. The objectives are (i) analysing the advisor behaviour in order to tune the built-in machine learning mechanism and (ii) evaluating the transient phenomenon between initial deployment and the state in which the required critical mass of users has been attained. The last section presents initial results calculated from schedules generated by the FEATHERS activity-based model for the six million population in Flanders.

Carpooling is a form of cooperation while executing a daily agenda hence requiring coordination between people. In actual practice, carpooling and similar concepts can be supported by intelligent advisory systems for trip matching. Evaluating the operational fitness of such systems in the testing phase requires an active community of users. This chapter focuses on the use of an agent-based model to exercise the advisory system under test.