Ultra Short Term Wind Power Prediction Model Based on WRF Wind Speed Prediction and CatBoost

Abstract

Due to the urgent requirements of clean energy development, the installed capacity of wind turbine is gradually increasing. In order to achieve effective wind turbine grid control, accurate wind power prediction is very important. In this paper, based on the US Atmospheric Research Center (NCAR) Weather Research and Forecast (WRF) model, the predicted wind field is obtained. Then, a super short-term fast wind power prediction model is proposed, which is composed of Spearman correlation coefficient method for feature extraction and catboost algorithm. With historical power, historical wind speed and predicted wind speed as inputs, the predicted power in the next 4 hours (15 minutes interval) is output. In order to evaluate the prediction ability of the proposed model, four algorithms, LSTM, CatBoost, XGBoost and LightBGM, are adopted. Taking root mean square error (RMSE), mean absolute error (MAE) and training time as evaluation indexes, the results show that CatBoost has the best comprehensive performance.