Reinforcement Learning-based Signal Control Strategies to Improve Travel Efficiency at Urban Intersection

Aiming at reducing urban traffic congestion and overcoming the defects of traditional timing control methods, two real-time signal control strategies based on Q-learning (QL) and Deep Q-learning network (DQN) algorithms were proposed and compared respectively. An algorithm framework was constructed with radar and video detector data as input and optimal intersection control strategy as output. Based on a traffic simulation platform, a typical urban intersection in Nanjing was simulated and the control effect of the methods were tested. The results show that the proposed two intelligent control strategies can actively respond to various traffic states, converge in a short training time and find the optimal control strategy. The two control strategies can effectively reduce the travel time by more than 20% and the stop delay by more than 30%. DQN-based control strategy is more effective than QL-based control strategy.