Existing transport planning methodologies which have been applied to hundreds of transport studies throughout the world for the past 40 years involve a sequential process for predicting short-run transport equilibria, often with four stages: trip generation, trip distribution, modal split and traffic assignment. Unfortunately, the sequential approach has an inherent weakness; its predictions need not be internally consistent. This deficiency has motivated attempts to predict all four stages simultaneously. Research intended to develop integrated models and related computational procedures for predicting short-run transport equilibria has proceeded in three directions. One line of investigation, the Equivalent Optimization approach, has significant computational advantages; the others, the Variational Inequality and Stochastic Equilibrium approaches, permit richer modeling of user behavior. A critical review of previous studies in transportation network equilibrium models illustrates the trade-offs between behavioral and computational aspects of the transport equilibrium problem.;In this dissertation, we address these trade-offs by performing a formal comparison between the variational inequality, equivalent optimization, and traditional sequential approaches to the problem.;To have a consistent comparison between the equivalent optimization approach and the variational inequality approach, a Generalized Simultaneous Transportation Equilibrium Model (GSTEM) has been developed. This model explicitly combines trip generation, trip distribution, modal split and traffic assignment for a general class of behaviorally sound demand models, and general asymmetric cost functions. The GSTEM is a generalization of the Simultaneous Transportation Equilibrium Model (STEM) which was developed by Safwat and Magnanti (1988) and which can be cast as an Equivalent Convex Program (ECP). The GSTEM can not be cast as an ECP, but as a Variational Inequality (VI). A relaxation algorithm has been developed to solve this VI.;Implementation programs for comparative analysis of computational and behavioral issues have been developed for the Tyler, Texas urban transportation network.;The main findings of this dissertation are: (1) The simultaneous approach to travel demand forecasting can consistently produce better traffic flow predictions compared with the existing conventional sequential approach currently used in practice, at essentially no additional computational cost. In the case of Tyler, Texas the relative improvements were between 9-43% with an average of 25%. (2) The equivalent optimization approach, in addition to offering significant computational advantages compared with the variational inequality approach, is behaviorally not as restrictive as was previously thought. That is, the additional efforts to include link interaction may not be strongly justified in practice, and hence the equivalent optimization approach (e.g., STEM) represent a reasonable practical compromise between computational and behavioral considerations of the problem.;Based on these findings, further use and application of the simultaneous approach (particularly the equivalent optimization models) to other urban transport studies throughout the world is strongly recommended.
展开▼