A stacking ensemble model based on nonlinear feature selection for photovoltaic power prediction

With the continuous expansion of photovoltaic (PV) power generation, the safe and intelligent operation of the grid system faces major challenges. Accurate PV power prediction can provide guarantee for grid-connected operation and dispatching plan of PV stations. But PV power is affected by a variety of factors and becomes difficult to predict. In this study, firstly, maximum information coefficient (MIC) and light gradient boosting machine (LightGBM) are used for nonlinear feature selection to obtain the optimal feature subset. Then extreme gradient boost (XGBoost), LightGBM and long short-term memory (LSTM) are used as base models to propose a stacking ensembel model (MIC-XLL) for PV power prediction. Experimental results on two PV datasets show that the proposed model is the best in several evaluation metrics compared with other models.