A novel hybridization of echo state networks and multiplicative seasonal ARIMA model for mobile communication traffic series forecasting

Abstract

For mobile communication traffic series, an accurate multistep prediction result plays an important role in network management, capacity planning, traffic congestion control, channel equalization, etc. A novel time series forecasting based on echo state networks and multiplicative seasonal ARIMA model are proposed for this multiperiodic, nonstationary, mobile communication traffic series. Motivated by the fact that the real traffic series exhibits periodicities at the cycle of 6, 12, and 24 h, as well as 1 week, we isolate most of mentioned above features for each cell and integrate all the wavelet multiresolution sublayers into two parts for consideration of alleviating the accumulated error. On seasonal characters, multiplicative seasonal ARIMA model is to predict the seasonal part, and echo state networks are to deal with the smooth part because of its prominent approximation capabilities and convenience. Experimental results on real traffic dataset show that proposed method performs well on the prediction accuracy.