Akademik Veri Yönetim Sistemi
International Journal of Photoenergy, cilt.2026, sa.1, 2026 (SCI-Expanded, Scopus)
Accurate weather forecasting is a key requirement for the reliable operation and optimization of renewable energy systems, particularly solar photovoltaic (PV) installations. This study presents a comparative evaluation of three widely used machine learning algorithms—artificial neural networks (ANN), support vector regression (SVR), and random forest (RF)—for forecasting three critical meteorological parameters: ambient temperature, wind speed at 10 m, and global horizontal irradiance (GHI). Two climatically distinct cities were selected as case studies: Antalya (Turkey), representing a Mediterranean climate, and Cheboksary (Russia), representing a temperate continental climate. Using high-resolution weather datasets spanning from 2018 to 2023, the models were trained and tested for forecast horizons ranging from 1 to 10 days. The model performance was assessed using three standard evaluation metrics: root mean square error (RMSE), mean absolute error (MAE), and the coefficient of determination (R2). The results demonstrate that RF provided the lowest prediction error for temperature and wind speed across both locations, whereas ANN yielded the most accurate forecasts for GHI. Moreover, paired forecasts involving temperature and GHI showed the highest R2 values using ANN (0.968 and 0.948), and RF showed optimal accuracy for temperature and wind speed pairs (0.975 and 0.967). The findings underscore the importance of algorithm–climate interaction and offer insights for model selection in hybrid renewable energy system modeling.