Day-to-day travel choice dynamics in transportation networks: Models and computational issues

A Bayesian updating model is developed to capture the mechanism by which travelers update their travel time perceptions from one day to the next in light of information provided by Advanced Traveler Information Systems (ATIS) and their previous experience. The availability and perceived quality of traffic information are explicitly modeled within the proposed framework. The development of the framework is motivated by the need to predict travelers' response to different information supply strategies. The perception updating model and the choice model are integrated with a dynamic traffic simulator (DYNASMART). A simulation framework for studying perception updating and day-to-day travel choice dynamics in presence of ATIS is also developed. For an efficient implementation of a detailed driver behavior model and other Dynamic Traffic Assignment (DTA) models, an agent based framework for solving ATIS/ATMS operations is developed. Within the overall distributed framework, two models are proposed for distributed traffic simulation. The proposed distributed models do not depend on hardware capability for speed-up and hence they are useful from the practical standpoint. Simulation experiments were carried out on the basis of the proposed framework. The results from these experiments are in conformity with some of the previous studies. In addition to modeling the mechanism by which drivers revise their perceptions from one day to the next in light of travel time information and experienced travel time, two important issues that were addressed in this thesis were: (a) the impact of the utility function; and (b) the existence of a steady-state in the network. It was found that the parameters of the utility function can have substantial impact on the network traffic pattern. Another significant result from the empirical study of the model was that overall departure rates as well as time dependent traffic patterns on individual links reached a steady state after some time. A preliminary implementation of the distributed model was also performed with DYNASMART. A large number of communication calls was needed to synchronize the simulation. Due to a very high communication latency of the Ethernet network, the communication overhead negated the advantage of the distributed system.