A Traffic Flow Prediction Model Based on Time-Space Fusion Mechanism

The traffic flow prediction highly depends on space and time, and the solution method which improves utilization rate of spatiotemporal data. The paper proposes an algorithm that combines B-spline function method, Gating fusion mechanism, Attention mechanism (weak semantic supervision mechanism) with Space Time Embedding (STE) (BGASTE). B-spline function is used to fit the time data, and at the same time Gating monitoring mechanism is used to eliminate the abnormal data. For spatial data, the node2vec is used to digitize the space and connections between two spaces, and the attention mechanism fuse B-spline data for detail area prediction. BGASTE is a continuous iterative process and can continuously calibrate the prediction functions based on the blocked time data and therefore predict the future trend based on the previous outcome.