Distributed learning control of traffic signals

Abstract

This paper presents a distributed learning control strategy for traffic signals. The strategy uses a fully distributed architecture in which there is effectively only one (low) level of control. Such strategy is aimed at incorporating computational intelligence techniques into the control system to increase the response time of the controller. The idea is implemented by employing learning classifier systems and TCP/IP based communication server, which supports the communication service in the control system. Simulation results in a simplified traffic network show that the control strategy can determine useful control rules within the dynamic traffic environment, and thus improve the traffic conditions.