Regional Traffic Flow Prediction on multiple Spatial Distributed Toll Gate in a City Cycle

It is hard to predict the regional traffic flow including hundreds of predicted points in a city cycle. Highly non-linearity and complexity are the characteristics of these spatial distributed dynamic values of traffic flow. In this paper, the spatial-temporal correlation of traffic data is studied and combined with deep learning approaches. A novel and improved network structure taken the advantages of both temporal convolutional network (TCN) and graph convolutional network (GCN) is presented, termed as temporal convolutional network spatial-temporal graph convolutional networks (TCN-STGCN). Meanwhile, the original data of 186 toll stations in Shaanxi Province is obtained as a data set through the flow calculation method. Furthermore, the existing typical deep learning models are selected to compare with the improved models to predict traffic flow. The results show that the improved model can make accurate predictions in as fast as 16 minutes, and the effect of long-term prediction (45min) is improved by 17.922% compared with the model before the improvement, which provides the possibility for vehicle navigation systems and intelligent traffic control.