Dynamic Spatiotemporal Graph Convolution Network for Cellular Communication Traffic Prediction

With the increasing complexity of the spatial topology of cellular communication networks and dynamic temporal characteristics of mobile network services, accurate prediction of cellular services is challenging. This article proposes the use of Dynamic Spatiotemporal Graph Convolutional Network (DSGCN) for mobile traffic prediction. This network modelling the dynamic characteristics of nodes in the cellular network traffic graph, and that capturing the dynamic spatiotemporal characteristics of border nodes by converting the cellular network traffic graph into hypergraph. In addition, for all nodes on different timestamps, complex spatiotemporal correlations are gathered through dynamic graph convolutional networks. By performing collaborative convolution on traffic flow maps and their hypergraphs, mobile traffic prediction is enhanced. Compared with conventional methods, the results of experiments show that this model has better predictive performance and higher training efficiency.