Short-term interval prediction of PV power based on quantile regression-stacking model and tree-structured parzen estimator optimization algorithm

Abstract

In recent years, the photovoltaic (PV) industry has grown rapidly and the scale of grid-connected PV continues to increase. The random and fluctuating nature of PV power output is beginning to threaten the safe and stable operation of the power system. PV power interval forecasting can provide more comprehensive information to power system decision makers and help to achieve risk control and risk decision. PV power interval forecasting is of great importance to power systems. Therefore, in this study, a Quantile Regression-Stacking (QR-Stacking) model is proposed to implement PV power interval prediction. This integrated model uses three models, extreme gradient boosting (Xgboost), light gradient boosting machine (LightGBM) and categorical boosting (CatBoost), as the base learners and Quantile Regression-Long and Short Term Memory (QR-LSTM) model as the meta-learner. It is worth noting that in order to determine the hyperparameters of the three base learners and one meta-learner, the optimal hyperparameters of the model are searched using a Tree-structured Parzen Estimator (TPE) optimization algorithm based on Bayesian ideas. Meanwhile, the correlation coefficient is applied to determine the input characteristics of the model. Finally, the validity of the proposed model is verified using the actual data of a PV plant in China.