Highway traffic flow prediction model with multi-component spatial-temporal graph convolution networks

Abstract Highway traffic flow forecasting is highly nonlinear and complex, so it is technically and operationally difficult to predict traffic flow accurately. In order to accurately predict the traffic flow of highway and alleviate its traffic congestion, a multi-component spatial-temporal convolution network model (MCSGCN) based on deep learning is proposed. The characteristics of short-term, daily cycle and weekly cycle of traffic data is modeled through three components and each component effectively captures the spatial-temporal correlation of traffic data by using spatial dimensional graph convolution and time dimensional convolution at the same time. Then input the sample data into the integrated model to train and extract the characteristics of traffic flow data. Finally, the model is tested on Highway England in England and PEMS-BAY, a public data set of Highway traffic in California. It is known through the experimental results that the integrated deep learning model is better than the single model in predicting the traffic flow of highway. The accuracy of traffic flow prediction on weekdays is higher than that on holidays. The prediction effect of MCSGCN is better than the existing model of traffic flow prediction.