CLSTGCN: Closed Loop Based Spatial-Temporal Convolution Networks for Traffic Flow Prediction

Traffic flow prediction plays a crucial role in assisting operation of road network and road planning. However, due to the dynamic correlations of road network nodes, the physical connectivity may not reflect the relationship of roads nodes. In this paper, a closed loop based spatial-temporal graph convolution neural networks (CLSTGCN) is proposed by constructing the closed loop with spatial correlation information of road network nodes. The designed model consists of multiple spatial-temporal blocks, which combines the attention mechanism with closed loop correlation information to promote the aggregation in spatial dimensions. Meanwhile, in order to improve the accuracy of long-term prediction, longterm road network trend is supplied into the model, which can capture the temporal features accurately. The experiments on two real world datasets demonstrate that the proposed model outperforms the state of art baselines.