Traffic Signal Control for Large-Scale Scenario: A Deep Reinforcement Learning-Based Cooperative Approach

Road travelers worldwide encounter a pervasive issue of traffic congestion. Because of its ability to manage intricate large-scale urban settings, deep reinforcement learning (DRL) has found extensive use in Adaptive Traffic Signal Control (ATSC). However, the majority of current algorithms are tailored to particular road networks or traffic scenarios, posing difficulties in their adaptation to large complex road environments. To tackle the mentioned issues, this paper suggests novel cooperative approach-based Multi-Agent Reinforcement Learning (MARL) for controlling large-scale networks. Traffic congestions are forecasted using the intensity of pheromone considering its neighboring agents to create fluid green waves. Double Deep Q-Network (DDQN) is designed to enhance execution efficiency and traffic signal stability in complex environments for each agent, where each agent controls a region. Based on Batna traffic data, results showcase the efficacy of the proposed approach outperforms baseline methods in mitigating traffic congestion.