Road pricing has received increasing attention for traffic congestion mitigation, but travel demand uncertainties cause a profound impact on the effectiveness of road pricing. This study proposes a robust simulation-based optimization method (RSBO) to address a time-varying road pricing problem under travel demand uncertainties. In this method, the simulation is served as the evaluation tool for the road toll plan with a travel demand uncertainty parameter. For each road toll plan, a stochastic kriging meta-model (SKG) is constructed to predict (or estimate) the samples with untested uncertainty parameters, and then help search for the worst case of this toll plan. After that, based on these worst cases of sampled toll plans, another SKG is also built to predict (or estimate) untested toll plans and help search for global robust optimum. Then, in a two-stage sequential framework based on efficient global optimization technique (EGO), an optimal computational budget allocation (OCBA) method is incorporated to search the global optimum and improve the accuracy of identifying the current best solution with less computational time. The proposed method is validated with an M/M/1 queuing model, and results indicate its effectiveness of obtaining the worst-case optimum with less computation budget. In the field experiment, the Sioux Falls network is modeled as the simulation scenario by VISUM, and numerical results show that considerable travel time is saved for all network users with the robust optimal time-varying road toll plan under various levels of travel demand uncertainties. Keywords: road pricing, travel demand uncertainty, robust simulation-based optimization

CICTP