Ollivier–Ricci Curvature Based Spatio-Temporal Graph Neural Networks for Traffic Flow Forecasting

Traffic flow forecasting is a basic function of intelligent transportation systems, and the accuracy prediction great significance for traffic management urban planning. The main difficulty predictions that there complex underlying spatiotemporal dependence in flow; thus, existing graph neural network (STGNN) models need to model both temporal spatial dependence. Graph networks (GNNs) are adopted capture flow, which can symmetric or asymmetric relations between nodes network. transmission process features GNNs guided by node-to-node relationship (e.g., adjacency distance) nodes, ignoring caused local topological constraints road To further consider influence topology on networks, this paper, we introduce Ollivier–Ricci curvature information connected edges network, based optimal transport theory makes comprehensive use neighborhood-to-neighborhood guide STGNNs. Experiments real-world datasets show with outperforms those only relationships ten percent average RMSE metric. This study indicates utilizing be captured more sufficiently, improving predictive ability models.