To improve the accuracy and reliability of solar irradiance prediction for photovoltaic power system, we propose a model to forecast short-term solar irradiance based on improved Stacking ensemble learning and error correction.First, the Gradient Boosting Decision Tree (GBDT) is used to perform feature selection and remove redundant characteristics of original data set, thus increase prediction accuracy and computing efficiency.Then, an improved Stacking irradiance prediction model is established.In accordance with the difference in prediction accuracy of prediction models in the primary layer under K-fold cross-validation, the prediction results are weighted, and the Box-Cox is employed to transform and process the training set data input from the first layer to the second layer of Stacking, so as to increase the normality and homoscedasticity of prediction.Finally, the historical prediction error data are extracted, and Random Forest is applied to construct an error model to further improve the prediction accuracy.The experimental results show that, compared with traditional models and classic Stacking models, the proposed method significantly improves the prediction performance on solar irradiance.