<![CDATA[Accurate weekly load forecasting is essential for improving the operational efficiency of thermal power plants, particularly in regions with dynamic demand patterns such as the Java–Madura–Bali (Jamali) power system. This study applies the Prophet machine learning model to forecast weekly electricity load in a 350 MW unit of the Pelabuhan Ratu coal-fired power plant and investigates the effect of different historical data lengths on forecasting accuracy using standard statistical metrics. The results show that a one-month dataset achieved the highest accuracy, with a MAPE of 14.72% and an RMSE of 47.32 MW, while longer datasets introduced additional noise and reduced sensitivity to recent load variations. The findings reveal that Prophet’s forecasting performance depends on a trade-off between generalization and responsiveness, with shorter and more recent datasets providing the optimal balance between accuracy, stability, and computational efficiency. The study confirms Prophet’s suitability as a lightweight and interpretable forecasting tool for operational planning, fuel management, and supporting biomass co-firing initiatives in coal-fired generation systems. while also offering new insights into the sensitivity of its performance to historical input length in operational-scale forecasting of thermal generation units.]]>
Copyrights © 2026
