East China Jiaotong University

Liyan Xiong

Yuan Jiang

Jiahao Ling

Yuming Ye

Xiaohui Huang

Disclosed is a traffic flow forecasting method based on a multi-mode dynamic residual graph convolution network, including following steps: constructing a relationship matrix and an adaptive matrix to learn the site dependence relationship for historical traffic data of traffic stations; using multi-mode dynamic graph convolution to extract traffic characteristics corresponding to different traffic modes; embedding the graph convolution into the gated cyclic neural network to realize the combination of space dependence and time dependence of traffic flow; connecting the network by using the dynamic residual, and combining the input traffic data with the decoding data to obtain the final forecasting value. The application utilizes two different methods to construct adjacency matrix, effectively captures traffic flow characteristics corresponding to different traffic modes, and dynamically fuses traffic flow characteristics of two different modes.

Growing community of inventors

Nanchang, China

Liyan Xiong