A Learning Algorithm for Adaptation to Traffic's Dynamical Variations

Abstract A hierarchically intelligent control system is employed to manage urban traffic. The system is composed of three levels: the Linguistic Organization, the Coordination, and the On-Line Controller; these are respectively responsible for the management of traffic in cities, neighborhoods, and single intersections. Traffic situations are divided into five categories: Sparse Traffic/ Computer System Failure, Light Traffic, Heavy Traffic, Oversaturated Traffic, or Immobile Traffic/Incident. Each coordinator determines the traffic situation in the neighborhood within its jurisdiction. For each traffic situation, then, a suitable control scheme is enforced. However, the selection of a control strategy solely on the basis of the traffic situation encountered does not thoroughly account for the traffic needs of the neighborhood. For example, in rush hour the coordination will declare Heavy Traffic regardless of the direction of the traffic trend (i.e., towards the business center area in the morning and out of that area in the afternoon). This paper, therefore, proposes a learning algorithm for adaptation to the coordination model so that management shortcomings are compensated, when the traffic trend changes its dynamical direction.