A Method for Traffic Flow Prediction Based on Spatiotemporal Graph Network in Internet of Vehicles

In the past, traffic flow prediction, whether using statistical methods or traditional machine learning, only started from the aspect of time or space, so it cannot make good use of the internal characteristics of the collected big data. This paper uses a brand-new prediction model, combines spatial network with temporal network, and selects the emerging technology graph convolution neural network GCN on the spatial side in recent years. On the time side, LSTM and GRU are selected for experiments to collect the data of three subnets. 70% of the data are selected for training and the rest for testing. The experimental results show that after using GCN instead of CNN, the prediction performance has been greatly improved compared with the DMVST-Net model from the same two aspects of time and space, and the MAPE can be controlled at about 3%. At the same time, the selection of LSTM and GRU has little impact on the experimental results. There is a small gap in the three evaluation indexes of MAE, MAPE, and RMSE. The accuracy of LSTM is relatively higher, while the training time of GRU is relatively shorter.