Nonparametric Probabilistic Wind Power Prediction Method Based on Ensemble Learning and Kernel Density Estimation

Under the global energy transition, wind power generation, as a clean alternative, has garnered significant attention. However, the uncertainties of wind power are becoming increasingly prominent. Current research on wind power prediction primarily employs parametric prediction methods, which are limited in scope. To address these limitations, a nonparametric probabilistic prediction method for wind power based on multi-model stacking quantile regression (Multi-Model Stacking Quantile Regression, MMSQR) is proposed. Firstly, the Pearson correlation coefficient is used to reveal the correlation between wind speed and the output power of wind farms. Subsequently, the DBSCAN clustering algorithm is employed for preprocessing and feature mining of the wind farm data. Secondly, based on the historical wind power data set, an ensemble learning framework called stacking is utilized to integrate multiple prediction models. A wind power prediction error distribution model is then established using the nonparametric kernel density estimation (NKDE) method. Finally, the proposed method is validated through simulation. The results demonstrate that compared with other single prediction methods, the proposed model shows closer peaks to the actual results and exhibits higher accuracy with more concentrated prediction outcomes.