A Multi-phase Intersection Traffic Signal Control Strategy with Deep Reinforcement Learning

Abstract

In this paper, a deep reinforcement learning (DNQ) algorithm for multi-phase intersection traffic control is proposed for improves the capacity of the urban road intersections. Here, deep learning is applied for extracting the features of traffic flow to learn the Q-function of reinforcement learning. The denoising stacked autoencoders are considered to reduce the effects of abnormal data generated during system operation. Considering the connection between the signal timing scheme and the phase sequence, the DNQ algorithm is used to adjust the sequence of the signal phase according to the dynamic traffic characteristics of the intersection while realtime self-adaptive adjustment of the signal timing. Simulations in platform consisting of VISSIM and Python are applied to test the algorithm. The performance of the proposed method is comprehensively compared with a traditional algorithm with fixed or free phase sequence under different traffic demand. Simulation results suggest that the proposed method signify-cantly reduces the delay in the intersection when compared to the alternative methods.