Now on TAP : accelerated solutions to the traffic assignment problem

The traffic assignment problem (TAP) represents the final and most computationally difficult step in the Urban Transportation Modeling System (UTMS). Prior work has studied several means of solving this problem quickly. This thesis furthers our understanding of this problem by investigating two methods designed to speed up TAP solution scenarios. A free and open-source implementation of the UTMS was implemented from scratch, including an implementation of Dial’s Algorithm B to solve the TAP. Chapter II of this thesis investigates the effects on computation time of a single Algorithm B parameter: the number of equilibrations performed on solution bushes prior to improvement. The results show that determining the best-performing value for this parameter is network-dependent, and that a point exists beyond which additional equilibrations do not provide improvements in runtime, potentially slowing down some networks’ computation. Chapter III investigates methods of approximating TAP solutions using artificial neural networks (ANNs) in the context of the network design problem, in which multiple different network designs may be tested. To remove the need to re-solve the TAP under each scenario, an ANN is trained to predict link flows given changes in network capacity. The results show that this method provides close approximations to the analytical solution in substantially less time than evaluating all possible scenarios from scratch.