A novel wind speed prediction method: Hybrid of correlation-aided DWT, LSSVM and GARCH

This paper addresses the difference between one time and real time decompositions in the wind speed prediction and results show the real time decomposition-based method may be ineffective in practice. Then the comprehensive analysis about challenges on applying the real time decomposition-based method is conducted, which is less addressed in literature. Such challenges mainly include: (i) the subseries decomposed from the training part are constantly changing with newly obtained data; (ii) the illusive components introduced by the decomposition reduce decomposition effectiveness; (iii) the end effect increases subseries volatility. Furthermore, to reduce these difficulties in prediction, a new hybrid method of correlation-aided discrete wavelet transform (DWT), least squares support vector machine (LSSVM) and generalized autoregressive conditionally heteroscedastic (GARCH) model is proposed. In this method: (i) if the correlation coefficient between each subseries and original data is smaller than the selected threshold, the corresponding subseries will be eliminated as illusive component; (ii) GARCH model is used to characterize the error for the remaining subseries and better capture the volatility in these subseries; (iii) model parameters are adjusted in real time to better reflect the wind speed change. Finally, case studies show that the proposed method has satisfactory performance in both accuracy and stability.