Adaptive Dynamic Programming for Multi-intersections Traffic Signal Intelligent Control

This paper aims at developing near optimal traffic signal control for multi-intersections in city. As a new optimization technique, adaptive dynamic programming (ADP) combines concepts of reinforcement learning and dynamic programming. ADP could learn continually from experience to achieve a near optimal control policy under varying conditions. However, without the cooperation among adjacent intersections, the near optimal control for each individual intersection can not guarantee a larger traffic area composing several intersections to be near optimal. This paper presents a new signal control method based on a model-free action-dependent ADP (ADHDP). This method can be used for cooperative control of multiple intersections. In every intersection, an ADHDP signal controller is adopted to adjust signal time according to an integrated unity parameter. The unity parameter is designed to consider not only the control performance in local intersection but also those in the neighbor intersections. Thus the designed controllers could achieve a set of near optimal control police for multi-intersections in a long run. Simulation results show that the trained controller achieves shorter average vehicular delay.