Leveraging graph convolution and multi-channel neural networks for accurate and robust spatio-temporal traffic flow prediction

The complexity and variability of traffic systems have made traffic prediction a highly sought-after research area. To enhance the accuracy of predicting traffic flow, we propose a Graph Convolution and Multi-Channel Neural Network (GCMNN) model that leverages spatio-temporal dynamic correlation in traffic data. Firstly, we use a graph convolutional network to extract the spatial structure of the traffic network and combine it with spatio-temporal correlation to derive dynamic correlation features of traffic conditions. Secondly, we employ an A-component comprising multiple onedimensional convolutional neural networks to learn how different combinations of past traffic conditions impact future traffic conditions. We then predict temporal embedding attention, extracting spatio-temporal dynamic correlation of the traffic road network by calculating weighted features. The outputs from the spatial and temporal modules are fused using a gated fusion mechanism, and final prediction results are obtained through linear layer mapping. We performed traffic prediction experiments using highway traffic data, the results showed that GCMNN was superior to other baseline. Furthermore, this model can be applied to short-term urban air quality prediction and other spatio-temporal predictions, providing valuable insights for government management and improving people's lives.