FORECASTING METHOD OF PHOTOVOLTAIC POWER GENERATION BASED ON FEATURE SELECTION AND XGBOOST ALGORITHM CONSIDERING INFLUENCE OF EXTREME ASTRONOMICAL AND METEOROLOGICAL EVENTS
After the parameter correlation was judged by Pearson's r feature selection method,the analysis decision tree index model was built to improve the prediction accuracy of the total horizontal radiation intensity,and the dimensionality reduction of the training set was realized through the principal component extraction of meteorological parameters.XGBoost algorithm is used to construct the prediction model,adding regular terms to control the complexity of the model,reducing the overfitting rate and improving the adaptability of the model to unknown data.By means of Taylor expansion,the selection of loss function and algorithm optimization process are decoupted,and the performance prediction and model evaluation of photovoltaic power stations under extreme astronomical and meteorological conditions are realized.The comparison between the predicted results and the measured values shows that the proposed method can automatically learn the missing value processing strategy,support various types of base classifiers,and have a wide range of optimization space.This model has good prediction accuracy and stability,as the average absolute percentage error in predicting photovoltaic power Pw,Performance ratio(PR),and capacity utilization factor(CF)is within 15%.
NETL
NSTL
万方数据
