Network Performance Optimization Using a Queueing Network Model

Abstract In this paper, we present and analyze a new aggregate model of urban traffic. The main challenge in the formulation of an aggregate model is to analytically grasp the correlation between the different components of the network while maintaining a tractable model. This correlation explains congestion effects and quantifies their overall network impact. Our analytic queueing network model captures this correlation and can therefore explicitly model complex phenomena such as spillbacks. This is useful to identify the sources of congestion (e.g. bottlenecks) and quantify their effects (Osorio and Bierlaire, forthcoming). This analytic model fits well within an optimization framework. We illustrate this in the context of the optimization of traffic signal timing in the network of the city of Lausanne. The aggregate model is calibrated using the outputs of a disaggregate model, that is a microscopic traffic simulation model. The analytic formulation of this optimization problem is presented. The optimal solution is then evaluated with the microscopic traffic simulation model. Its performance is compared with that of several other methods. We show how these methods differ in their ability to cope with increasing congestion; and emphasize the importance of taking into account the correlation between consecutive roads.