Optimization of Solar Photovoltaic System Performance Using Machine Learning

Abstract

Predicting photovoltaic (PV) power output accurately is necessary for managing energy, integrating solar into the grid, and pinpointing the exact conditions where systems run at peak efficiency. Focusing on solar irradiance, ambient temperature, module temperature, wind speed, humidity, and generated power as key metrics, this work forecasts PV output using three machine-learning methods: Random Forest, XGBoost, and an Artificial Neural Network (ANN). The models are evaluated against each other using Mean Absolute Error (MAE), Root Mean Square Error (RMSE), Mean Absolute Percentage Error (MAPE), and the coefficient of determination (R²). We then ran feature-importance rankings and an optimization routine on the top model to find the operating and weather parameters that yield the highest PV efficiency. The representative data show XGBoost outperforming the other two algorithms with the highest R² and lowest errors; output depends most heavily on solar irradiance and module temperature, with peak efficiency occurring under strong sunlight, low humidity, and moderate panel heat. This work outlines a turnkey, fully reproducible framework for forecasting and optimizing PV output that field engineers can adapt to any site with logged operating data.