<![CDATA[The increasing reliance on clean energy has accelerated the development of solar energy infrastructure. However, its intermittent nature—especially in tropical urban climates—poses significant challenges to maintaining grid stability. This study compares the performance of two machine learning algorithms, Long Short-Term Memory (LSTM) and Extreme Gradient Boosting (XGBoost), for hourly solar irradiance forecasting in two climatically distinct tropical cities: Jakarta and Bogor. Using a 10-year historical dataset from NASA POWER that includes solar irradiance and relevant meteorological variables, this research addresses the gap in comparative analysis of deep learning versus ensemble models within high-granularity tropical data settings. The methodology involves data acquisition, preprocessing, feature engineering, model development, hyperparameter tuning, and evaluation using RMSE, MAE, and R² metrics. The results show that LSTM consistently outperforms XGBoost in both cities. In East Jakarta, LSTM achieved a RMSE of 29.24, MAE of 15.63, and R² of 0.9875, compared to XGBoost with RMSE of 38.65, MAE of 18.92, and R² of 0.9782. Similarly, in Bogor Regency, LSTM achieved RMSE of 30.73, MAE of 16.89, and R² of 0.9862, outperforming XGBoost which recorded RMSE of 38.41, MAE of 18.68, and R² of 0.9785. These findings highlight LSTM's superior ability to capture complex temporal dependencies and nonlinear trends in solar irradiance time-series data, especially under the fluctuating weather patterns characteristic of tropical urban environments. The results provide strong empirical support for implementing LSTM-based forecasting in solar energy management systems across similar geographic regions.]]>
Copyrights © 2025
