Analysis of Intersection Traffic Light Management System in Mandalay City

Abstract

Traditional traffic signal controls at intersections are ineffective that causes traffic congestion, time wasting, and environmental problems. This paper proposes an intersection traffic light management system that incorporates machine learning technique with the traditional traffic light management system to solve the above challenges. To tackle traffic congestion and waiting time problems, Q learning algorithm is used as the reinforcement learning to choose new action. Action states include various traffic signal phases that are important in generating in realistic control mechanism. In this work, SUMO open source traffic simulator is used to construct realistic traffic intersection settings and the intersection Discretized Representation method is applied to get environment state and to calculate reward. Then, experience replay technique is used to enable reinforcement learning agent to memorize and reuse past experience. The results show that the proposed traffic system outperforms traditional traffic system in Mandalay city. The waiting time reduced 3 times compared with the traditional traffic control.