An ensemble machine learning-based solar power prediction of meteorological variability conditions to improve accuracy in forecasting

ABSTRACT The increasing energy demand has significantly improved solar photovoltaic (SPV) systems as a distributed energy source. Real-time control of SPV performance is vital for accurate solar power (SP) prediction. The article proposes an ensemble Machine Learning Approach (MLA) called Random Forest Algorithm-Based Regression Model (RFARM) for hourly forecasting of SP. The approach selectively analyzes meteorological and solar irradiance data (SI) to enhance short-term solar panel prediction. It focuses on employing a correlation-based approach using an RFA with regression to achieve improved SP prediction accuracy. The study compares the PV power generated at Thiagarajar College of Engineering (TCE), Madurai, using four prediction techniques: Artificial Neural Networks (ANN), Random Forest (RF), K-Nearest Neighbours (KNN), and Support Vector Machine (SVM) along with a proposed RFARM for different meteorological weather conditions over a 24-hour time horizon. The proposed RFARM method achieves high prediction accuracy by selecting significant parameters, avoiding artificial filtering, and minimizing errors, particularly in predicting solar output during cloud shading. The RFARM model outperforms conventional methods in predicting the daily curve of solar power performance. It achieves an RMSE of 1.52, MAE of 14, and R-squared of 98%. Feature selection further improves accuracy, reducing RMSE by 12.5% and MAE by 17.2% respectively.