Machine Learning – Based Prediction of Daily Global Solar Radiation Using Local Meteorological Observations

This study aims to predict the daily total global solar irradiance (GHI) utilizing machine learning methods using meteorological indicators such as relative humidity (RH), wind speed (WS), temperature (TE), atmospheric pressure (AP), sunshine duration (SD) measured daily between 2015 and 2022 at the station within the Hakkari Provincial Directorate of Meteorology. Solar irradiance prediction is of great importance, especially in terms of determination of the solar energy potential, photovoltaic (PV) system performance analysis, and energy planning. In this context, Extreme Gradient Boosting (XGBoost), random forest (RF), support vector regression (SVR), linear regression (LR) algorithms were applied, and the prediction successes of the models were compared using coefficient of determination (R²), mean absolute error (MAE), root mean square error (RMSE), and correlation coefficient (R) performance criteria. The findings denoted that SD, TE and AP were the most effective determinants of daily solar radiation. Furthermore, among the four different algorithms used in the GHI estimation study, the SVR model was found to exhibit superior performance compared to the other models, with R²=0.703, RMSE=1.381, MAE=0.983, and R=0.838. In the LR model, R² was calculated as 0.685, RMSE as 1.422, MAE as 1.082, and R as 0.827. The RF and XGBoost algorithms, respectively, showed similar performance with R²=0.654, RMSE=1.491, MAE=1.061, R=0.808 and R²=0.646, RMSE=1.509, MAE=1.094, R=0.803. The findings demonstrate that, despite Hakkari's high altitude and variable climate conditions, reliable and cost-effective solar radiation prediction models can be developed using local meteorological data, making significant contributions to renewable energy applications.