Y.R. Guo, X.M. Wang, H. Zhang, G.J. Jim

Pages: 13-23

Abstract
Aiming at the problems of poor prediction effect, low accuracy and long prediction time, the traffic flow prediction method based on double flow graph convolution network is proposed. This paper analyzes the composition and basic principle of the dual flow graph convolution network, and establishes the traffic flow prediction model of the diversion area according to the basic characteristics of the traffic flow; uses the double flow graph convolution network to process the high-dimensional data of the traffic flow, trains the diverging area in the peak hours, obtains the weight of the network, and obtains the classification results of the characteristics; extracts the spatial characteristics of the traffic flow through the convolution spectrum of the dual flow graph The basic structure of time dimension modeling is established by attention coding model, and the time characteristics of traffic flow are extracted, and the traffic flow prediction value of diversion area is obtained, and the traffic flow prediction of diversion area is realized. The experimental results show that the prediction accuracy of the traffic flow prediction method is high, and the traffic flow prediction time is about 23 MS, which can effectively shorten the traffic flow prediction time.
Keywords: double flow graph convolution network; traffic flow prediction; graph convolution spectrum method; graph convolution neural network; Laplace matrix