Hybrid machine learning and optimization method for solar irradiance forecasting

The objective of this study is to investigate a novel hybrid model for the accurate prediction of direct normal irradiance. For this purpose, a decomposition technique, a clustering technique, an optimizer, and the CatBoost algorithm have been employed. Data for all four seasons were gathered from China's Jiangsu province. After data collection, the most significant features were selected using feature selection methods. Then, these features were decomposed using the wavelet decomposition method and subsequently clustered by using the sample entropy method. Following that, these data were fed to the CatBoost model which was optimized by battle royale optimizer. The proposed hybrid model outperformed other algorithms used in this investigation by having the lowest error and highest efficiency. Especially, average values for the four-season coefficient of determination, root mean square error, mean absolute percentage error and mean absolute error were 0.99, 9.68, 1.95, and 3.85, respectively.