Dynamic Spatiotemporal Graph Convolutional Networks for Short-Term Traffic Flow Prediction
Aiming at the problem that the existing traffic flow prediction models lack the ability to model the dynamic spatiotemporal correlation of traffic data,a new deep learning-based dynamic spatiotemporal graph convolutional network(DSTGCN)model is proposed.The model can model spatiotemporal correlations from traffic data without giving road network information.The dynamic spatiotemporal graph convolutional layer consists of two main parts,one is the dynamic adjacency matrix generation module,which uses temporal auto-correlation mechanism and spatial attention mechanism to capture dynamic spatiotemporal correlations in traffic data;the other is the spatiotemporal graph convolution,which uses graph convolution and standard 2D convolution to efficiently aggregate information.By stacking dynamic spatiotemporal graph convolutional layers,DSTGCN is able to capture spatiotemporal dependencies at different temporal levels.The method proposed in this paper is tested on a public California freeway datase.The results show that the proposed DSTGCN model outperforms the existing benchmark methods in every evaluation index.On the PeMSD04 dataset,compared with the current newer GeoMAN and ASTGCN models,the MAE is reduced by 4.00 and 2.16 respectively,verifying the effectiveness of the proposed model in traffic flow prediction.
NETL
NSTL
万方数据
维普
