Generic Dynamic Graph Convolutional Network for traffic flow forecasting

Yi Xu, Liangzhe Han, Tongyu Zhu, Leilei Sun, Bowen Du, Weifeng Lv

Abstract: Highlights•A generic and dynamic graph convolutional network named GDGCN is proposed.•It is the first to explore the parameter-sharing mechanism in traffic forecasting.•A novel temporal graph convolutional block is designed.•A dynamic graph learning module is introduced to mine spatial-temporal relations.•Evaluations show that GDGCN outperforms other approaches in all instances.