Development of new solar radiation prediction models using different machine learning techniques

• Three new model for global SR estimation was developed by ML approaches. • The newly developed model was compared with different models. • The different statistical error tests were used. • it was observed that the results were acceptable. • it is expected that this study will make very important contributions to the literature. In this study, three new prediction models have been developed in global solar radiation (GSR) prediction using different machine learning (ML) approaches. The aim of the study is to present three new ML models that will make GSR predictions using GSR data and to compare the performance of these new models with different GSR prediction models widely used in the literature. While developing the prediction models; Support Vector Regression (SVR), Robust Linear Regression (RLR) and Gaussian Process Regression (GPR) algorithms were used. In this study, the new models were developed using ML in the MATLAB program. The obtained results were compared using different statistical error analysis tests such as mean absolute percentage error (MAPE), root mean square error (RMSE), mean square error (MSE) and the analysis of variance (R 2 ). The results on the all used statistical indictors revealed that the GPR method (MAPE: 0.018979; RMSE: 0.026872; MSE: 0.0007221; R 2 : 0.85) outperformed the all tested models in terms of accuracy. The most important original aspect of this study is that GSR estimation models were developed by determining Angstorm coefficients for the first time with ML techniques that have not been used before in the literature.