Multiresolution-based Echo State Network and its Application to the Long-Term Prediction of Network Traffic

A multiresolution-based echo state network (MESN) based on echo state network (ESN) is proposed in this paper. ESN proves to be very efficient for modeling and time series prediction. The learning process of MESN was further improved by using a multiresolution-based learning algorithm. The proposed MESN was applied to the long-term prediction of real network traffic and its performance was compared with the traditional ESN. The results show that the prediction of MESN gives a 27.32% reduction in terms of the normalized mean square error (NMSE) over traditional ESN, which indicates that MESN is very appropriate for network traffic prediction.