Dynamic System Optimal Offline Assignment For Arrival Time Based OD: Model And Algorithm

Dynamic traffic assignment (DTA) is the fundamental research of ITS. Although in the last two decades, DTA models have been studied extensively, there are still numerous obstacles in using DTA in practice. The objective of this thesis is to develop a new dynamic system optimal assignment model for arrival time based OD, which can be put into use in large scale actual network to estimate and evaluate various measures and policies of dynamic traffic management precisely and efficiently. In this thesis, existing DTA literature is first reviewed by classifying the various approaches into four broad methodological groups: mathematical programming, optimal control, variational inequality, and simulation-based. Classic models and developments of each group are discussed in-depth. The key obstacles of DTA offline models are then summarized.Regarding the estimation of time-dependent demand, a new estimation procedure is proposed. In the proposed procedure, information contained in travel survey data that is currently ignored by the demand forecasting process is utilized to generate time-dependent demand of each category of trips. Entropy maximizing theory is then explored to develop formulations for distribution of time-dependent trips. The impact of time of day on modal split is modeled in two ways.After the time-dependent OD of each category of trips is gained, the thesis focuses on the problem of DTA for arrival time based demand, which has not been studied extensively. A new set of linear programming (LP) formulations with cell transmission model embedded as the underlying traffic flow theory is presented to study the system behavior. Insights into the problem are obtained by constructing the dual of the LP and using complementary slackness theory. Computational results on test networks are presented to test the impact of the change of parameters. Although solution algorithms are very well established for LP problems, the size of the problem that can be solved using the LP formulation is still limited by the power of mathematical programming solvers. To relax such limitations, the network structure embedded in the LP formulations is explored to develop decomposition schemes with the Dantzig-Wolfe decomposition technique. Such decomposition schemes of the LP formulation can potentially allow us to solve DTA analytically on large scale actual networks. Computational results on test networks are presented to test the applicability of the decomposition schemes.Possible means by which the DTA model proposed in this thesis can be used to help analyze existing problems are finally discussed. The whole dynamic network analysis procedure including estimation of time-dependent OD is implemented on the case study of Changchun City to obtain issues necessary for this implementation and demonstrate the applicability of this procedure on large scale actual networks.