STHGCN: A spatiotemporal prediction framework based on higher-order graph convolution networks

Abstract

Spatiotemporal prediction based on deep learning is one an important content in spatiotemporal knowledge mining with the development of various sensor technologies such as global position system(GPS), mobile devices, remote sensing and traffic detection. Different from the simple calculation of multivariate time series data, spatiotemporal data have obvious time dependence and space dependence. Therefore, the key to spatiotemporal prediction lies in the modelling of time and space dependence. In this paper, we propose a new spatiotemporal prediction framework named STHGCN, whose core idea is to model spatiotemporal dependencies with higher-order dependencies. Specifically, for the space–time dimension, we propose and implement a high-order temporal different network and high-order spatial semantic GCN network. At the same time, in the experimental part, we have conducted extensive experiments on several actual databases. The experimental results show that STHGCN is superior to the existing state-of-the-art spatiotemporal prediction models.