A traffic prediction algorithm based on Bayesian spatio-temporal model in cellular network

Abstract

5G communication will bring a surge traffic in cellular network. The traffic in cellular network not only has strong variability by time, but also has strong spatio-temporal correlation, which brings large difficulty to predict. In order to make reasonable use of communication network resources, it is important to describe and predict the spatio-temporal information of traffic in cellular network. In this paper, we propose a traffic prediction algorithm based on Bayesian spatio-temporal model to predict the spatial distribution of traffic in cellular network at different moments via realistic traffic data of base stations (BSs). Firstly, we select Gaussian Predictive Process (GPP) as the basic model of the Bayesian spatio-temporal model and set proper prior distribution of the parameters. Secondly, we train the basic model by Gibbs sampling and the realistic traffic data to obtain the posterior distribution of the parameters. Then, we predict the spatio-temporal information of traffic in cellular network by Markov chain Monte Carlo (MCMC) computational techniques. Finally, we make theoretical analysis of prediction accuracy for the prediction results. The Index of Agreement (IA) of the prediction results in three different areas can reach 0.9 above, which indicate good prediction performance. The traffic prediction algorithm can be used to predict the spatio-temporal information of traffic in cellular network in different areas.