FasterSTS: A Faster Spatio-Temporal Synchronous Graph Convolutional Networks for Traffic flow Forecasting

Abstract--Accurate traffic flow prediction heavily relies on the spatio-temporal correlation of traffic flow data. Yet, CNNs have limited capability in modeling non-Euclidean I. In [16], Yu feature extraction components such as multi-head et al. proposed a spatio-temporal graph convolutional network self-attention mechanism, RNN, and TCN, but only relies (STGCN) to complete the task of traffic flow forecasting. The specific contributions of FasterSTS CNN to capture comprehensive spatial correlation of different are as follows: traffic nodes in traffic networks. In [17], the task of traffic 1) In order to describe the current node, the conventional flow forecasting has been modeled as a diffusion process graph computation approach first collects the information on a directed graph with a diffusion convolutional recurrent of all other nodes for each node.