Optimal Time-Varying Pricing for Toll Roads Under Multiple Objectives: A Simulation-Based Optimization Approach
本文构建了基于仿真的优化框架，利用代理模型近似仿真函数，优化马里兰州五段收费公路的费率，使网络平均出行时间降低20%，通行费收入增加50%。
The determination of the pricing for tolled facilities always involves consideration of multiple objectives, e.g., efficiency, safety, pollution, reliability, and economy. Simulations are widely used to evaluate the performance of transportation systems as to the various objectives in response to different travel demand management policies. However, transportation simulation is usually associated with high computation costs and non-closed-form objective functions. This paper builds a simulation based optimization framework, and uses surrogate models to approximate the true simulation function. A significant amount of computation time can be saved with this method. To solve real world application problems, we develop infill strategies for multiobjective and constrained optimization problems. Using DynusT as the simulator, we optimize the toll rates for a five-segment toll road in Maryland, and successfully update the Pareto front based on initial samples for the multiobjective optimization problem. The method works even more efficiently for the constrained optimization problem. By adjusting the toll rates for the five segments, the network-wide average travel time can be reduced by 20% compared to the currently implemented toll scheme: A total of 22,250 hours can be saved in travel time for all network users in the three-hour morning peak period. Also, the toll revenue is increased by 50% compared to the baseline case.