Ultra-short-term forecast of wind speed and wind power based on morphological high frequency filter and double similarity search algorithm

This paper proposes a forecast model for ultra-short-term prediction of wind speed and wind power, which is based on a morphological high-frequency filter (MHF) and a double similarity search (DSS) algorithm. The MHF is proposed to decompose the time series into two components: the mean trend, which reveals the non-stationary tendency of the time series, and the high frequency component, which depicts the fluctuations. The same strategy is employed to forecast the mean trend and the high frequency component, respectively. The two components are reconstructed in the phase space, respectively, where a non-uniform embedding strategy is proposed to better reveal their information. To select similar segments to be used for local forecast, the novel DSS algorithm is proposed for high frequency component, while the Euclidean distance is used for the mean trend. Finally, the least squares-support vector machine (LS-SVM) model is applied to forecast each component, respectively, and their sum composes the final prediction. Simulation studies are carried out using wind speed and wind power data obtained from four databases, and the results demonstrate that the MHF/DSS model provides more accurate and stable forecast compared to the other methods.