<![CDATA[Accurate solar irradiation prediction is crucial for optimizing solar energy generation and supporting energy management systems. This study addresses the challenges of advancing the accuracy of solar irradiance forecasting by developing a Hybrid XGBoost-LSTM residual learning model, using historical solar radiation data from Lampung Selatan. The results highlight the capability of the Hybrid XGBoost-LSTM model as a powerful tool for forecasting solar irradiation, providing a more precise and reliable solution compared to standalone methods. This is demonstrated by the improvement in R² values from the standalone XGBoost predictions, increasing from 0.62 to 0.87 for the training data and from 0.50 to 0.77 on the testing datasets. In addition, the hybrid model demonstrates a decrease in RMSE and MAPE values when compared to standalone XGBoost model, with RMSE and MAPE dropping from 0.26 and 4.19% to 0.14 and 2.39% for the training dataset, and 0.32 and 5.28% to 0.23 and 3.70% for the testing dataset. These findings indicate that the inclusion of LSTM improves the model's ability to refine the residuals from XGBoost, capturing intricate temporal dynamics and fluctuations in solar irradiance data, making it a more reliable and effective tool for predicting solar energy potential in complex environments.]]>
Copyrights © 2026
