<![CDATA[This study evaluates one-hour resolution Global Horizontal Irradiance (GHI) forecasting in tropical Indonesia using NASA POWER satellite reanalysis data. Three approaches are comparatively evaluated: Persistence, a temperature-corrected radiative physics model, and an Extreme Learning Machine (ELM) utilizing six selected features alongside a physical capping mechanism. The evaluation was conducted on 9,691 identical test samples using a 60:20:20 chronological validation protocol. The results indicate that the ELM achieved the highest accuracy (RMSE: 10.70 W/m², MAE: 5.56 W/m², sMAPE: 4.03%) with a Skill Score of 0.885, representing an 88.5% error reduction compared to the baseline. The physical capping mechanism proved effective in stabilizing the sMAPE by constraining predictions within atmospheric physical limits. This exploratory study demonstrates that a straightforward integration of data-driven approaches and physical constraints holds potential as a foundational alternative in developing irradiance forecasting frameworks for renewable energy integration in tropical regions.]]>
Copyrights © 2026
