1265

Traffic flow prediction incorporating urban road network structure is a highly nonlinear and complex time-series data prediction problem with spatio-temporal dynamic correlation. The accurate and real-time prediction of future traffic status using spatio-temporal data in the region can effectively monitor and analyze the regional characteristics, which in turn can provide convenient and effective reference for traffic travel, road planning and management. Considering the relative closure of the expressway and the topological structure of the road network, a new prediction model incorporating a temporal convolutional network (TCN) with improved graph convolution (GCN) is proposed to predict regional road network traffic flows in parallel and accurately, termed as temporal convolutional network spatial-temporal graph convolutional networks (TCN-STGCN). The model uses GCN to learn the complex road network topology to capture spatial correlation, and then captures temporal correlation with the features of TCN such as large expansion field and short training period. Finally, the improved model is validated by selecting 186 real data of expressway toll stations in Shaanxi Province as the dataset in this paper. The experimental results show that the proposed model improves the long-time prediction performance by 13.04% (30 min) and 17.92% (45 min), respectively, and reduces the running time by 22.7% compared with the regional prediction model (STGCN), and improves the performance by 27.96% (30 min) and 50.91% (45 min), respectively, compared with the single-point prediction optimal model (TCN). The runtime is reduced by 73.40%, respectively. The proposed model can be more suitable for real-time traffic flow prediction at different time periods.