The increasing integration of renewable energy into smart grids requires accurate forecasting to maintain grid stability and optimize energy management. This study compares the performance of three forecasting models—XGBoost, LightGBM, and Long Short-Term Memory (LSTM)—for predicting solar photovoltaic (PV) output, wind power output, and total renewable energy generation. To improve forecasting capability, the study applied feature engineering techniques, including time-based variables, rolling averages, and lag features to capture temporal dependencies within the dataset. An 80/20 chronological train-test split was used to maintain time-series integrity. Model performance was evaluated using Root Mean Squared Error (RMSE), Mean Absolute Error (MAE), and coefficient of determination (R²). Results show that gradient boosting models significantly outperformed the LSTM model across all forecasting targets. LightGBM achieved the best performance for solar PV forecasting (RMSE: 3.27, R²: 0.987) and total renewable energy prediction (RMSE: 4.66, R²: 0.988), while XGBoost produced the highest accuracy for wind power forecasting (RMSE: 3.26, R²: 0.987). In contrast, the LSTM model generated negative R² values, indicating poor predictive performance. These findings demonstrate that gradient boosting methods are highly effective for structured renewable energy forecasting datasets and offer strong potential for intelligent smart grid applications. The forecasting framework is designed to support IoT-enabled smart grid systems, where renewable energy data are continuously collected through distributed sensors and transmitted via intelligent communication networks. The proposed models can assist real-time energy monitoring, edge-based prediction, and intelligent network management for modern smart grid infrastructures.
Kinasih et al. (Thu,) studied this question.