Evolution of adaptive route choice behaviour in drivers

摘要

Traffic assignment, the process by which vehicle origin-destination flows are loaded on to discrete paths traversing a road network, has been traditionally approached as a non-linear optimisation problem where it is expected that travellers will each minimise their own travel time. While such models are suitable for obtaining an `average’ expected network state, traffic conditions on a day to day basis are inherently uncertain due to variations in travel patterns and incidents such as vehicle breakdowns, roadworks or bad weather resulting in fluctuations in realised traffic flows. Further, such models do not consider the transition from one `average’ state to another when an aspect of infrastructure is changed such as a new road opening or the introduction of long term roadworks. This paper therefore examines the evolution of driver route choice over time in stochastic time-dependent networks, specifically focusing on how individual experience of network conditions guides future decisions and its relationship with en-route switching opportunities. Existing algebraic and empirical models of route choice evolution are assessed (particularly using discrete whole path choices to assess benefits of information provision) and it is proposed that incorporating adaptive path routing based on expected correlations in traffic flow behaviour is more suitable than fixed path models for capturing the extent of observed uncertainty in network conditions. We present this issue and explore through simulation a model where drivers adapt expected road link travel times for a given trip based on a combination of previous experience and discovered link travel times on that trip. We show how adaptive behaviour produces travel times which are on average faster than non-adaptive behaviour, confirming the potential of this modelling approach.