Dynamic Congestion Pricing in Urban Networks with the Network Fundamental Diagram and Simulation-Based Dynamic Traffic Assignment

Abstract

This thesis focuses on modeling and optimization of two-region urban pricing systems and analyzing and understanding the effects of pricing on the network traffic flow. The motivation of this work is the fact that traffic congestion is growing fast in cities around the world especially in city centers, and hence the need for an effective and efficient travel demand management (TDM) policy. With the aim of advancing the current congestion pricing theory, this thesis proposes and integrates different advanced pricing regimes with the Network Fundamental Diagram (NFD) and simulation-based dynamic traffic assignment (DTA), studies and compares different computationally efficient simulation-based optimization (SO or SBO) methods, and analyzes and under-stands the effects of different pricing regimes on the network traffic flow. This thesis demonstrates through computer simulations the effectiveness of a well-designed pricing system on improving the network performance. The major finding is that the distance only toll, which represents the state of the practice, naturally drives travelers into the shortest paths within the pricing zone (PZ) resulting in a more uneven distribution of congestion and hence, a larger hysteresis loop in the NFD and lower network flows especially during network recovery. This limitation is overcome by two more advanced pricing regimes, namely the joint distance and time toll (JDTT) and the joint distance and delay toll (JDDT), through the introduction of a time and a delay toll component, respectively. Moreover, this thesis explicitly models and minimizes the heterogeneity of congestion distribution as part of the toll level problem (TLP). The toll area problem (TAP) is also investigated by means of network partitioning. To optimize different pricing regimes through computer simulations, this thesis develops two computationally efficient SO frameworks. The first framework employs a proportional-integral (PI) controller from control theory to solve a simple TLP featuring a low-dimensional decision vector, a set-point objective and only bound constraints. The second framework employs regressing kriging (RK) from machine learning to solve a complex TLP that has either a high-dimensional decision vector, a complex objective, or a set of complex constraints. A comprehensive comparison between the two methods and two other widely used methods, namely simultaneous perturbation stochastic approximation (SPSA) and DIviding RECTangles (DIRECT), are performed. Overall, this thesis provides valuable insights into the study, design, and implementation of urban pricing systems and the effects of pricing on the network traffic flow. Results of this work not only help in developing effective pricing systems to mitigate urban traffic congestion, but also provide competitive solutions to other types of network design problems (NDPs).