Traffic Flow Prediction Model based on Multi-view Spatio-temporal Convolution
Traffic flow forecast is always the primary task of intelligent transportation system.Due to the complex spatio-temporal dependencies of traffic flow sequence,it is extremely challenging to predict it accurately.Many existing works are mainly based on cyclic neural networks,graph networks and Transformer models to build traffic flow forecast models.Considering that the convolutional network has the advantages of high computational efficiency and strong feature extraction ability,in the paper a traffic flow forecast model is proposed based on multi-view spatio-temporal convolution.The model performs representation learning on the sequence data at the input encoding layer,and fuses it with position information and time information.In the spatio-temporal feature representation learning layer,multiple representation learning modules are designed considering that sequences have different periodic patterns.Each spatio-temporal representation learning module completes local spatio-temporal feature mining based on one-dimensional convolution,and then realizes global spatio-temporal feature mining based on causal convolution.In the prediction layer,the channel attention mechanism is introduced to improve the effectiveness of the model in utilizing spatio-temporal features.Experimental results on two real traffic datasets verify the effectiveness of the MSTC model on traffic flow prediction tasks.
NETL
NSTL
万方数据
