Document Type : Original Article

Authors

• mohammad hajian ¹
• Tayyeb Nazghelichi ²

¹ Gorgan University of Agricultural Sciences and Natural Resources

² Department of Biosystems Engineering, Gorgan University of Agricultural Science and Natural Resources, Gorgan, Iran

Abstract

This study investigates the prediction of photovoltaic (PV) energy production using advanced machine learning algorithms, leveraging meteorological data and production capacity from 300 residential PV plants in Sydney, Australia. The dataset was processed into daily values to account for weather variability, and three machine learning models (Random Forest Regression (RFR), Support Vector Regression (SVR), and Light Gradient Boosting Regression (LightGBR)) were implemented. Following rigorous preprocessing and hyperparameter optimization, LightGBR exhibited superior predictive performance, achieving a coefficient of determination (R²) of 0.9020, a mean absolute error (MAE) of 3.1621, and a mean squared error (MSE) of 0.1005. Compared to previous studies, the optimized LightGBR model demonstrated enhanced accuracy in PV energy forecasting, underscoring its potential for improving predictive modeling in this domain. These findings have significant implications for optimizing energy distribution, enhancing smart grid integration, and supporting decision-making in energy management systems. Accurate forecasting of PV energy output is essential for improving operational efficiency, minimizing energy waste, and advancing sustainability objectives in renewable energy management.

Keywords

• Forecasting
• Photovoltaic Systems
• Machine learning
• Random Forest algorithm
• Support Vector Regression (SVR)
• Light Gradient Boosting Regression (LightGBR)