Short-term probabilistic forecasting for regional wind power using distance-weighted kernel density estimation

As the integration of wind power into the power grid increases rapidly, the total output of the regional wind farms has become the concern of the power system operators and market traders. This study proposes a short-term probabilistic forecast model for this regional application. The uncertainty information provided by the proposed model can help the users make better decisions in the power system. A new distance-weighted kernel density estimation (DWKDE) method is proposed to forecast the full distribution function of the wind power. Its distance kernel is able to assign different weights to the samples similar to the target point. The beta kernels are introduced to adapt to the double-bounded characteristic of the wind power density. To further improve the performance of the DWKDE model, a regime-switching strategy is applied based on the regional wind direction clustering, while a feature selection method of minimal-redundancy-maximal-relevance is provided to determine the proper feature set. A case study of 28 wind farms in the East China is provided to evaluate the performance with the quality measures of reliability, sharpness, and the pinball score. The proposed method is easy to use and performs well according to the results of the evaluation.