Pressure, Queue, and Average Speed - Based Multi-Agent DQN for Optimizing Traffic Signal Control

Traffic congestion poses a significant challenge in urban areas, and deep reinforcement learning (DRL) offers an encouraging method for traffic signal control. We evaluate the DQN algorithm, with PQAS (Pressure, Queue, Average Speed) using the SUMO traffic simulator. Our results demonstrate that DQN improves traffic flow and reduces congestion compared to traditional methods. Our findings highlight the potential of DRL in developing more effective traffic signal control systems, with DQN, combined with PQAS. PQAS represents three reward functions pressure-based, queue-based, and average speed-based reward functions. emerging as a promising candidate indicating that optimizing for average speed and minimizing queue lengths are key factors in achieving efficient traffic signal control.