SOLAR RADIATION PREDICTION FOR RENEWABLE ENERGY: A MACHINE LEARNING PERSPECTIVE

Abstract

Accurate solar radiation forecasting plays a pivotal role in enhancing the efficiency, reliability, and integration of large-scale renewable energy systems. The performance of photovoltaic (PV) systems strongly depends on atmospheric and seasonal variability, necessitating precise short-term predictions to support optimal energy management and maintain grid stability. This study applies advanced machine learning (ML) techniques within a time-series forecasting framework to improve the accuracy of solar radiation prediction. Rigorous data preprocessing—encompassing cleaning, segmentation, and validation—ensures dataset integrity and prevents data leakage. A range of regression models, including Ridge, Lasso, XGBoost, Decision Tree, Random Forest, and Linear Regression, undergo evaluation using Root Mean Squared Error (RMSE) as the primary metric. K-fold cross-validation identifies Random Forest as the most effective model, demonstrating its superior performance in enhancing predictive accuracy and enabling more reliable integration of solar energy into modern power grids.