Surrogate modeling for real-time traffic management
Real Time Traffic Management aims to reduce congestion by dynamically affecting vehicles’ movements through advanced sensing and communication technologies. Simulation-based optimization can be employed to address this control problem, but it might become unsuitable for real-time applications given the large number of simulations and, ultimately, the significant computational effort to produce high-quality answers. Surrogate modeling creates a statistical model (or surrogate model) that is used to accurately represent the outcome of the simulation. Subsequently, this trained surrogate model can be integrated with optimization in place of the original simulation for a wide range of cases. In this project we develop and investigate the performance of alternative surrogate models for managing curbside deliveries occurring on an arterial road.
