In view of the non-linear and dynamic characteristics of traffic flow at urban intersections, a non-linear traffic flow queuing model is proposed. Based on the inherent periodic characteristics of macro traffic flow, an asynchronous Q-reinforcement learning algorithm with two hidden layers of artificial neural networks is designed, so that the vehicle queue length at each entrance lane can be balanced as far as possible. The convergence of the algorithm is analyzed theoretically. Compared with traditional actuated control and linear Q-learning traffic signal control, the effectiveness of the proposed algorithm is verified by taking average delay as performance index.

CICTP