Intelligent Traffic Signal Control Strategy Based on Rainbow Algorithm

The reasonable control strategy of urban traffic signal can alleviate the congestion at intersections. To improve the learning efficiency and stability of intelligent signal control algorithms, this paper proposes an urban traffic signal control strategy based on the Rainbow algorithm. Based on the classical DQN algorithm, the Rainbow algorithm introduces the Double Q-Network, Noisy Network, Dueling-DQN, and Distribution-DQN into the network structure, and introduces the prioritized experience replay technology and multi-step learning into the algorithm training process. Taking the intersections of Niangre Road and Dangre Road in the Lhasa City as an example, three simulation scenarios are set based on the traffic flow intensity. Three traffic signal control strategies are conducted by using fixed-time signal control algorithm, classical DQN algorithm, and Rainbow algorithm, respectively. Results demonstrate that the Rainbow algorithm can converge quickly and reduce vehicle delay time, queue length effectively.