Short-Term Forecasting of Floating Photovoltaic Power Generation Using XGBoost Model

Accurate forecasting of energy output from floating photovoltaic (FPV) systems is crucial for optimizing energy management and ensuring grid stability. FPV technology offers advantages such as reduced land usage and enhanced efficiency due to water-based cooling effects. However, environmental variability-particularly fluctuations in temperature and solar irradiation-poses challenges to accurate power prediction. Traditional forecasting models often struggle to capture the nonlinear dynamics in FPV data. To address this methodological challenge, this study employs the Extreme Gradient Boosting (XGBoost) algorithm as a predictive framework for forecasting FPV power generation. The research utilizes empirical data systematically collected at Universiti Malaysia Pahang Al-Sultan Abdullah (UMPSA) over a threemonth observation period extending from December 2023 to February 2024. The predictive model incorporates three primary environmental and operational variables as input parameters: ambient temperature, transient horizontal irradiation, and photovoltaic module temperature. Model validation and performance assessment are conducted through rigorous statistical evaluation metrics, specifically Root Mean Square Error (RMSE) for quantifying prediction accuracy, Mean Absolute Error (MAE) for measuring average deviation magnitude, and the coefficient of determination (R<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup>) for assessing the model's explanatory capacity and statistical reliability. This methodological approach enables comprehensive evaluation of the XGBoost algorithm's efficacy in predicting FPV power output under varying environmental conditions. XGBoost is compared with Artificial Neural Networks (ANNs) and Random Forest, demonstrating superior accuracy with an RMSE of 61.61, MAE of 36.54, and R<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup> of 0.921. These findings highlight the effectiveness of machine learning models, particularly XGBoost, in improving short-term energy forecasting for FPV systems, underscoring the importance of high-quality data and advanced predictive modeling in renewable energy applications.