Traffic assignment models : applicability and efficacy

This dissertation is concerned with the traffic assignment problem (TAP), an important tool in transportation planning. We first study a theoretical extension of TAP which incorporates symmetric interactions in the link travel costs. In particular, we prove the theoretical convergence of conventional solution methods for this problem, and analyze convergence behavior for these methods. We also show how a set of real world interactions such as merge models can be modeled using these type of symmetric interactions. Second, we apply these findings to a practical case study of railroad electrification, formulated as a rail network design problem. We solve this problem on a large network representing the North American railroad network, and analyze the solutions to provide policy recommendations. We model the interactions between diesel- and electric-goods flow as a symmetric congestion cost and a separable fuel/crew cost. Third, we study the empirical behavior of TAP under input uncertainty. Specifically, we analyze the effects of three types of input errors (uniform, origin- or destination- specific, spatially correlated) on network metrics, such as total system travel time or congestion, at equilibrium. Empirical bounds for these output metrics are identified for various levels of input error. We apply these findings to a case study to demonstrate potential usage for planning purposes. Lastly, we conduct a comprehensive empirical study on the convergence behavior of traffic assignment convergence metrics. We analyze five commonly used network metrics at various convergence levels, for different solution algorithms, and identify concrete thresholds for convergence. We also show the relationship between different convergence criteria metrics, allowing for transfer of these thresholds to different metrics.