An Intelligent Signal Control Method Based on Optimized DDQN Algorithm with Connected and Automated Vehicles

At present, most traffic signal control systems rely on fixed signal phase sequence and timing. The signal control method based on deep reinforcement learning technology can adapt to various traffic environments and make optimal decisions. However, most research studies in related fields do not consider the state of signal lights in the state space design, and the algorithm structure is also relatively backward. In this paper, an Optimized Double Deep Q Network (ODDQN) algorithm is proposed, which includes signal states in the state matrix and adds a priority experience playback mechanism to further improve the performance of DDQN algorithm. Finally, the simulation experiment is carried out using SUMO and real traffic pattern. The experimental results indicate that the intelligent traffic signal control system designed in this paper successfully reduces the average waiting time of vehicles and the average queue length at intersections, showing superior control efficiency.