Origin-based algorithms for transportation network modeling

The transportation planning process relies on travel forecasts that result from various transportation models. Some of the well-known models are formulated as non-linear convex optimization problems. Solving these problems is quite challenging due to their non-linear nature and their combinatorial structure. Large scale networks of practical interest increase the need for computationally efficient algorithms.; At the heart of most transportation models stands the traffic assignment problem, which is to predict the route choice of travelers given the origin and destination of each traveler, under the assumption that each traveler seeks to minimize the time/cost associated with their chosen route. Almost all known algorithms for the traffic assignment problem are either link-based or route-based. Link-based algorithms fail to achieve highly accurate solutions, while route-based algorithms are regarded as impractical for large-scale networks due to their large memory requirements.; This research presents a different approach in which the solution is represented by origin-based link flows aggregated over all destinations. This approach provides highly accurate detailed solutions, its memory requirements are reasonable even for large-scale networks, while computation times are substantially lower than for link-based algorithms.; Formal mathematical definition of the proposed algorithm for the traffic assignment problem is presented, followed by a proof of its convergence. Computational results on several networks, including large-scale one, demonstrate the superior performance of the proposed algorithm.; A direct procedure to extract route flows from origin-based solutions is presented, based on a behavioral assumption named bypass proportionality. The close relationship between this assumption and entropy maximization is examined. The ability to disaggregate origin-based link flows to route flows demonstrates that the detail provided by origin-based solutions is practically equivalent to the detail provided by route-based solutions, while the memory requirements of origin-based solutions are substantially lower.