Author(s)

Peng Zheng, Yuansong Li, Maoyan Lin, Youxin Hu

School of Computer Science and Engineering, Sichuan University of Science and Engineering, Yibin, China.

Urban traffic flow prediction plays an important role in traffic flow control and urban safety risk prevention and control. Timely and accurate traffic flow prediction can provide guidance for traffic, relieve urban traffic travel pressure and reduce the frequency of accidents. Due to the randomness and fast changing speed of urban dynamic traffic data flow, most of the existing prediction methods lack the ability to model the dynamic temporal and spatial correlation of traffic data, so they cannot produce satisfactory prediction results. A spatio-temporal convolution network (ST-CNN) is proposed to solve the traffic flow prediction problem. The model consists of two parts: 1) a convolution block used to extract spatial features; 2) a block of time used to characterize time. Data has been fully mined through two modules to output the prediction results of spatio-temporal characteristics, and at the same time, skip connection (direct connection) has been made between the two modules to avoid the problem of gradient explosion. The experimental results on two data sets show that ST-CNN is better than the baseline model.

Traffic Flow, Deep Learning, RNN, CNN

Zheng, P. , Li, Y. , Lin, M. and Hu, Y. (2023) Urban Traffic Flow Prediction Based on Spatio-Temporal Convolution Networks. Journal of Computer and Communications, 11, 15-23. doi: 10.4236/jcc.2023.113002.