Deep Q-Network Based Intelligent Traffic Signal Control: Case Study of a Large Single Intersection Using SUMO

As urbanization continues to escalate, traffic congestion has emerged as a significant challenge in metropolitan areas. Existing traffic signal systems mainly rely on manually designed signal plans, which struggle to adapt to the dynamic and complex nature of modern traffic environments. This paper presents a theoretical introduction to an intelligent traffic signal control system based on Deep Q-networks (DQN) through a case study of a large single intersection. The DQN-TSC model is developed using the SUMO (Simulation of Urban MObility) platform. An agent for deep reinforcement learning (DRL) within the proposed model is designed by properly defining the state, action, and reward function. Through the appropriate parameter settings of the traffic model and reinforcement learning, simulation experiments are conducted in the SUMO environment. The comparative experimental results demonstrate that the proposed DQN-TSC model can effectively improve traffic efficiency at intersections, with significant improvements in evaluation metrics including total waiting time, average waiting time, and total number of stopped vehicles.