A modelling of route choice behaviour in transportation networks: an approach from reinforcement learning

Abstract

This paper presents a simple yet powerful traffic equilibrium calculation method. The basic concept of the method is motivated from reinforcement learning with profit sharing. In the model, individual drivers are regarded as heterogeneous entities, being assumed to form their own values for each route through driving experiences and communications to the environment. The method realizes a disaggregate user equilibrium on a congested network so that it is useful to analyze the interrelationships between each driver's characteristics and the resultant network equilibrium. Further, this method not only covers from stochastic user equilibrium to deterministic user equilibrium, it also is applicable to a network with asymmetric cost functions or with discontinuous cost functions.