Multi-step wind speed forecasting model based on wavelet matching analysis and hybrid optimization framework

Abstract Accurate wind speed forecasting is beneficial to the management of the wind power system. A hybrid WPD-DA-NAR wind speed forecasting model under moving window framework is proposed in this study. The WPD (Wavelet Packet Decomposition) is utilized to process the original wind speed time series. Since the mother wavelet function is the core component of the WPD, the matching relationship between the predictor and 17 different mother wavelet functions is discussed. The optimal mother wavelet for the application of the proposed hybrid model is determined. The NAR (Nonlinear Autoregressive) network is employed to build the forecasting models for decomposed sub-layers. A hybrid optimization framework base on the DA (Dragonfly algorithm) is adopted to optimize the NAR network. In order to capture the characters of wind speed time series and update model accordingly, the optimized NAR is applied under a moving window framework. The proposed hybrid model is compared with 8 existing models. The experimental results indicated that: (a) the dmey wavelet provides the best results among all the included 17 mother wavelet functions; (b) the proposed hybrid WPD-DA-NAR model under moving window framework has the best performance in all steps and metrics, compared with 8 existing models.