<![CDATA[Accurate photovoltaic (PV) power forecasting is critical for managing solar intermittency, especially within industrial power systems. The objective of this study is to develop a high-precision forecasting framework specifically tuned for the stochastic weather patterns at the Pertamina RU III Sungai Gerong facility. This study proposes a two-stage deep learning framework utilizing a full year of historical operational data (June 2024-June 2025), totaling 2,795 samples, to ensure it captures a wide range of seasonal variations. The system optimized RNN and LSTM models as base learners which were then integrated by an MLP meta-learner through a Stacking Ensemble. The results demonstrate that a Tapered MLP (64-32) configuration performs best, minimizing the Root Mean Squared Error (RMSE) to 130.26 kWh and maximizing the R² index to 0.816. The Storm Simulation scenario was constructed using an interactive environment to replicate stochastic solar radiation drops exceeding 50% of peak capacity caused by sudden cloud movement. The results reveal that the ensemble model maintains a low error rate of 5.3%, outperforming standalone models which failed with errors up to 19.7%. These findings has a potential to support proactive load-shifting and synchronization in industrial-scale PV systems.]]>
Copyrights © 2026
