Solar Forecasting for Future Floating PV Deployment at Jatiluhur Reservoir: A Comparative Study of Statistical and Deep Learning Approaches

Accurate solar irradiance forecasting is critical for the integration of floating photovoltaic (FPV) systems into power grids, particularly in tropical regions characterized by high convective variability. This study evaluates four forecasting approaches persistence, multiple linear regression (MLR), random forest (RF) and long short-term memory (LSTM) for predicting global horizontal irradiance (GHI) at Jatiluhur Reservoir, Indonesia’s largest reservoir and a prospective site for large-scale FPV development. Using only open-access NASA POWER satellite data (2021–2025), models were trained and tested for 1-hour and 3-hour ahead forecasts with lagged GHI features incorporated to enhance temporal representation. Results show that all models benefit significantly from lag features, with LSTM achieving the highest accuracy (RMSE = 91.87 Wh/m², forecast skill = 83.7%) at the 3-hour horizon. The study demonstrates that high-accuracy forecasting is achievable without ground-based measurements, provided appropriate feature engineering and model selection are applied. These findings offer a practical framework for energy planners and policymakers to support FPV feasibility studies and grid integration strategies in data-scarce tropical environments.