Title: Intelligent traffic assignment method of urban traffic network based on deep reinforcement learning

Authors: Zhiyong Jing; Huanlong Zhang

Addresses: College of Computer and Communication Engineering, Zhengzhou University of Light Industry, Zhengzhou, 450001, China ' College of Electric and Information Engineering, Zhengzhou University of Light Industry, Zhengzhou, 450001, China

Abstract: In order to overcome the problems of high relative error and long response time of traditional methods, a new intelligent traffic flow allocation method based on deep reinforcement learning is proposed. In this method, deep reinforcement learning is introduced, and experience pool technology is used to obtain and retain samples in a certain stage to train urban traffic network. The complete track is divided into several independent state action pairs, and the sample database is established. In a certain range, the vehicle congestion density is simplified to the degree of congestion. When the starting point and the end point are known, all traffic demands between the two points are calculated, allocation and intelligent traffic network traffic assignment is then realised. Experimental results show that the average relative error of passenger travel time is 12.34%, the traffic flow prediction indexes are the lowest, and the allocation time is the highest, which is 0.878 s.

Keywords: deep reinforcement learning; DNQ; urban traffic; network traffic; intelligent distribution.

DOI: 10.1504/IJICT.2023.127683

International Journal of Information and Communication Technology, 2023 Vol.22 No.1, pp.60 - 72

Received: 16 Nov 2020
Accepted: 06 Jan 2021
Published online: 14 Dec 2022 *