Traffic Prediction Based on Spatio-temporal Graph Convolution Model of Gating Mechanism
ZHANG Jin
WANG Hui
KUANG Lidan
LIU Yuanyuan
LI Qiang
SUN Cheng
Abstract：The traditional traffic prediction method ignores the spatio-temporal dependence of traffic data and shows poor per-formance.The majority of the existing traffic prediction models are based on GNN and RNN.Although these models improve the pre-diction performance to a large extent,they still have some limitations,such as the gradient explosion problem caused by RNN when dealing with long time sequences.In this paper,a gating mechanism is proposed based on spatio-temporal graph convolution net-work(GSTGCN)for traffic prediction.The core of GSTGCN is the spatio-temporal convolution module(ST-Conv module),which mainly consists of three parts,which are standard gated temporal convolution layer,spatial graph convolution layer and gated diffu-sion causal temporal convolution layer.In addition,this paper also introduces the mask matrix to construct the adjacency matrix of the traffic graph and skip connection and residual connection are added to gated diffusion causal temporal convolutional layer.Final-ly,experiments are carried out on PeMSD7(M)and METR-LA datasets,and the experimental results show that the prediction per-formance of GSTGCN model is better than that of advanced baselines in medium-term and long-term prediction.
Keywords：mask matrixgating mechanismgated diffusion causal temporal convolutionskip connectionresidual connec-tion
Publication Date：2025-05-20
Online Publishing Date：2025-08-15(First online date of this platform, not the publication date of the document)
Pages：9( 1242-1250 )