Solar energy systems play a vital role in alleviating the potential environmental risks that arise from using conventional energy sources. Since the performance of these systems relies heavily on solar radiation, it is crucial to develop reliable tools for accurate solar radiation forecasting. This study investigates the utilization of supervised machine learning models for predicting solar radiation in the Southern Anatolian Region in Türkiye. Nine different models were used to predict both instantaneous and daily solar radiation in the study area, based on 18 years (2005–2022) of weather data obtained from the NSRDB database. The results showed that the tree-based models had better performance than other models evaluated. Moreover, the extra trees model was found to have the best performance, with R2 scores above 0.999 for daily global horizontal irradiation, 0.975 for daily direct normal irradiation, 0.955 for instantaneous global horizontal irradiation, and 0.945 for instantaneous direct normal irradiation. Moreover, the extra trees model achieved its highest accuracy when predicting the daily global horizontal irradiation, with a station-wise average R2 score of 0.9999, root mean squared error of 0.0244, mean absolute error of 0.0142, and mean absolute scaled error of 0.0047.
Bashir et al. (Thu,) studied this question.