Comparative study on three new hybrid models using Elman Neural Network and Empirical Mode Decomposition based technologies improved by Singular Spectrum Analysis for hour-ahead wind speed forecasting

High precision forecasting of wind speed is urgently needed for wind power utilization. In this paper, Empirical Mode Decomposition (EMD) based technologies, including EMD and its advanced versions ensemble EMD (EEMD) and complete EEMD with adaptive noise (CEEMDN) are applied for improving wind speed prediction accuracy. Three new hybrid models (EMD-SSA-ENN, EEMD-SSA-ENN and CEEMDN-SSA-ENN) are proposed in which EMD, EEMD and CEEMDN are combined with Singular Spectrum Analysis (SSA) and Elman neural network (ENN) respectively. SSA is exploited to re-handle components with the highest frequency disaggregated from the decomposition technologies, of which the procedure is systematically studied herein. The experimental prediction results show that: 1. through the retreatment of SSA, the performances of the new proposed hybrid models improve significantly; 2. compared to the persistence method, single ENN model, ARIMA, EMD-RARIMA and some methods in the references, all the proposed methods can give a much more accurate forecast; 3. among all the proposed methods, the performance of the hybrid model CEEMDN-SSA-ENN are the best.