Study on short-term photovoltaic power prediction model based on the Stacking ensemble learning

Abstract

As solar photovoltaic (PV) power generation is very sensitive to environmental changes, with the characteristics of randomness and intermittent, a new PV power prediction model based on Stacking ensemble learning method is proposed in this paper. Firstly, a PV power prediction model based on Stacking of multiple machine learning algorithms is established. It considers the difference of training principles and characteristic contribution analysis of different algorithms, and gives full play to the advantages of each model. Secondly, the single model and Stacking model are trained iteratively using the operating state parameters and meteorological parameters of PV panels provided by the data collection system (DCS) of the PV power station. The model can predict the future PV power generation. The eXtreme Gradient Boosting (XGBoost) with superior performance in single model is selected to compare with Stacking. The prediction results show that root mean square error (RMSE) of Stacking is 0.1007, 1.84% lower than that of XGBoost, and the score is 90.85, 1.49 higher than that of XGBoost. Besides, the PV power prediction model proposed is of great significance to distributed generation grid connection, power grid stability, unit optimization, economic dispatch and supervision.