Multistep Wind Power Forecast Using Mean Trend Detector and Mathematical Morphology-Based Local Predictor

This paper proposes a novel forecasting model based on a mean trend detector (MTD) and a mathematical morphologybased local predictor (MMLP) to undertake short-term forecast of wind power generation. In the proposed MTD/MMLP model, the nonstationary time series describing wind power generation is first decomposed by the MTD, which employs some new notions and conventional morphological operators. The decomposition yields two components-the mean trend, which reveals the tendency of the time series, and the stochastic component, which depicts the fluctuations caused by high frequency of the variability. Subsequently, the p-step forecast is conducted for these two components separately. The mean trend is forecasted on the basis of the least-square support vector machine (LS-SVM) model, while the p-step forecast for the stochastic component is carried out by the MMLP, which involves performing morphological operations employing a novel structuring element (SE) in the phase space. Finally, the forecast of wind power generation is achieved by combining the separate forecasts of two components. In order to evaluate the accuracy and stability of the MTD/MMLP model, simulation studies are carried out using the data obtained from three widely used databases sampled in different periods. The results demonstrate that the MTD/MMLP model provides a more accurate and stable forecast compared to the traditional methods.