A Short-Term Regional Wind Power Prediction Method Based on XGBoost and Multi-stage Features Selection

When wind power is integrated to the grid on a large scale, the power system is greatly disturbed by wind power because of the randomness and volatility of wind power. Regional wind power prediction (WPP) is an effective measure to improve the ability of grid to absorb wind energy. A short-term regional WPP method based on XGBoost and multi-stage features selection is presented in the paper to improve the accuracy and efficiency of regional WPP. Firstly, the original features extraction of regional wind farm is performed, and high-dimensional features of regional wind farm are constructed. Then, the filtering features selection method and the embedded features selection method are applied to perform multi-stage features selection for the high-dimensional features of the constructed regional wind farm. The XGBoost short-term regional WPP method is constructed based on the results of multi-stage features selection. And the key parameters of XGBoost short-term regional WPP model are optimized. Finally, the results of the XGBoost short-term regional WPP method is compared with the four-hour regional WPP results of three models, which are regression tree, random forest and Support Vector Machine (SVM). The result is shown that compared with regression tree, random forest and SVM, Root Mean Squared Error (RMSE) of the XGBoost model is reduced about 1% in the 4-hour regional WPP.