A robust regressor model for estimating solar radiation using an ensemble stacking approach based on machine learning

ABSTRACT

Of late, with the rapid advancement in solar power generation, some difficulties have cropped up due to solar intermittency, necessitating an accurate forecast of Global Horizontal Irradiance. To address this challenge, this work proposes a feature selection method based on the variance inflation factor combined with mutual information (VIF-MI). In addition, this article also proposes an improved stack ensemble with an extra tree (SE-ET) regressor in which the machine learning models such as Decision Tree, Random Forest, Light Gradient Boost Machine, Cat Boost, and Extreme Gradient Boost are taken as base learners and the ridge regressor along with the extra tree regressor are taken as meta learners. The prediction performance is evaluated by comparing the outcome of each considered model and bagging with the proposed method. For the VIF-MI method, the proposed SE-ET exhibits the best estimation performance and good reduction of error, such as in mean absolute error by 16–60%, in root mean squared error by 7–64%, and in mean absolute percentage error by 20–75% as compared to the other models. Conclusively, it may be inferred that the ensemble technique minimizes the prediction error and can help plan for using sporadic solar resources more effectively.

KEYWORDS:

• Ensemble learning
• bagging method
• stacking
• machine learning
• global horizontal irradiance

Disclosure statement

No potential conflict of interest was reported by the author(s).