Online Traffic Light Control Through Gradient Estimation Using Stochastic Fluid Models

Authors:	Panayiotou Christos, University of Cyprus, Cyprus Howell William C., University of Maryland, United States Fu Michael, University of Maryland, United States
Topic:	7.4 Transportation Systems
Session:	Traffic Flow Modelling and Control
Keywords:	Traffic control, perturbation analysis, stochastic approximation, stochastic systems.

Abstract

In this paper, we consider the problem of dynamically regulating the timing of traffic light controllers in busy cities. We use a Stochastic Fluid Model (SFM) to model the dynamics of the queues formed at an intersection. Based on this model, we derive gradients of the queue lengths with respect to the green/red light lengths within a signal cycle. We derive both a simple and a periodic model and report preliminary numerical results comparing the performance of the estimates with finite-difference and smoothed perturbation analysis estimates. Then all estimators are used to optimize the traffic system via Stochastic Approximation.