Attention-Based Graph Convolution Reduces Traffic Flow Prediction Latency

A new spatial-temporal fusion graph convolution network addresses information propagation delays between adjacent nodes in traffic networks. The model replaces heavy, stacked structures with an attention-based mechanism to reduce computational training time and improve real-time prediction capabilities.