An Approach to Multi-agent Deep Q-Network Optimization of Signal Control in Multi-intersection Road Environments to Enhance Urban Traffic Flow

The escalating challenges of urban traffic congestion, resulting in amplified time and economic losses, profoundly impact daily life. Inappropriate signal switching on ordinary roads is identified as a significant contributor to this issue. Traditional approaches relying on human experiences for manipulating parameters in general signal control often yield suboptimal outcomes. To address this critical problem, this research proposes a dynamic traffic signal control system using a multi-agent approach with the Deep Q-Network method. In this urgent scenario, the proposed system aims to achieve precise parameter manipulation within a road environment featuring multiple intersections. A meticulous comparative analysis is conducted against static signal control and non-coordinated multi-agent systems, incorporating detailed numerical results to assess performance metrics. Results from a comprehensive 500,000-step experiment reveal the proposed method’s adeptness in balancing performance and computational efficiency, leveraging inter-agent cooperation. “Comparison Method 3,” inspired by Joo and Lim’s methodology (2021), consistently outperforms others, particularly in congestion reduction.