Wind speed forecasting method using wavelet, extreme learning machine and outlier correction algorithm

The wind speed forecasting is important for the security of the wind power systems. This paper proposes a new hybrid method for the multi-step wind speed forecasting based on Wavelet Domain Denoising, Wavelet Packet Decomposition, Empirical Mode Decomposition, Auto Regressive Moving Average, Extreme Learning Machine and Outlier Correction Method. In the proposed hybrid architecture, the Wavelet Domain Denoising is adopted to reduce the noise of the original wind speed series and a secondary decomposing algorithm is presented to reduce the intermittent of the original wind speed series. In the proposed secondary decomposing algorithm, the Wavelet Packet Decomposition is utilized to decompose the wind speed series into a number of non-stationary sub-layers and stationary sub-layers, and the Empirical Mode Decomposition is used to further decompose the obtained non-stationary data into a series of intrinsic mode functions. Finally, the Auto Regressive Moving Average and Extreme Learning Machine models are employed to complete the multi-step forecasting computation for the decomposed stationary sub-layers and intrinsic mode functions, respectively. In addition, the new Outlier Correction Method is proposed to guarantee the robustness of the built Auto Regressive Moving Average and Extreme Learning Machine models during their forecasting computation. To estimate the performance of the proposed new hybrid forecasting method, a series of performance comparing experiments is provided in the study. The involved comparing forecasting models consist of the Auto Regressive Moving Average model, the BP model, the Elman model, the Extreme Learning Machine and the proposed different hybrid models. The experimental results indicate that: in the comparisons of all the involved mainstream wind speed forecasting models, the proposed hybrid model has the best forecasting performance.