A simulation optimization framework for the management of congested urban road networks

Deriving optimal traffic management schemes for urban road networks typically relies on the use of complex simulation tools, that capture in detail the behavior of drivers as well as their interaction with the network infrastructure. The integration of these traffic simulators within an optimization framework is an intricate task. Indeed, these simulators can be seen as stochastic nonlinear functions that are expensive to evaluate. Simulation-based network optimization should therefore start with an important modeling effort, in order to exploit the structure of the problem at hand. In particular, we believe that in order to perform both fast and reliable simulation-based optimization for congested networks, information from the simulation tool should be combined with information from a metamodel (surrogate) that captures at a lower degree of detail the structure of the underlying problem. In this paper, we propose a surrogate that combines information from a calibrated microscopic traffic simulation model with an analytical queueing network model. We integrate this surrogate within a derivative-free trust region optimization framework. We apply the framework to solve a fixed-time traffic signal control problem for a subnetwork of the Lausanne city center. We compare the performance of the derived signal plans with that of an existing signal plan for the city of Lausanne.