Enhancing solar irradiance prediction precision: A stacked ensemble learning-based correction paradigm

Abstract

Accurate solar irradiance prediction is critical for ensuring reliable control of solar energy systems. This study proposes a stacked ensemble learning model to correct daily solar irradiance forecasts derived from numerical weather prediction (NWP). The ensemble framework integrates 5 base models—multiple linear regression (MLR), artificial neural network (ANN), K-nearest neighbors (KNNs), random forest (RF), and support vector regression (SVR)—using stacking technology, with a meta-model applied for final prediction refinement. Experimental results demonstrate significant improvements over the original NWP forecasts: the corrected model reduces the mean absolute error (MAE) and root mean square error (RMSE) by 47% and 41%, respectively, while increasing the R² determination coefficient by 11%. The proposed approach effectively enhances the accuracy and reliability of traditional solar irradiance prediction models, offering a novel and practical solution for solar energy forecasting. This work holds substantial value for optimizing solar power system operations and advancing renewable energy utilization.