MGCN: Dynamic Spatio-Temporal Multi-Graph Convolutional Neural Network

Abstract

Traffic prediction plays an important role in urban planning and smart city construction. Reasonable forecasting of future traffic conditions can effectively avoid traffic congestion and allow planning time for people to travel. However, complex traffic networks and non-linear time dependence make traffic prediction very challenging, and existing methods often lack the ability to model the dynamic spatio-temporal correlation of traffic data, making forecasting results unsatisfactory. We therefore propose a dynamic spatio-temporal multi graph convolutional neural network (MGCN) based on graph convolution network and attention mechanisms to perform traffic prediction tasks. Specifically, we design a spatio-temporal graph convolution module to simultaneously capture traffic network structure information, dynamic neighbour node information and traffic variation information in the temporal dimension, and effectively fuse them to represent comprehensive and dynamic spatio-temporal correlation. Further, we transform the historical time series into a future time series representation and resolve the future traffic conditions along the temporal and spatial dimensions on two Decoders respectively, aggregating the multidimensional information. Adequate experiments were conducted on two real large scale datasets, and the experimental results demonstrate the effectiveness and superiority of our approach and achieve a better level of performance than other baseline methods.