Machine Learning-Driven Curtailment Prediction in a Renewable School Microgrid System

Abstract

Reliable and sustainable electrification of educational institutions remains a major challenge in developing regions due to rising energy demand, grid instability, and environmental concerns. This study proposes a hybrid renewable microgrid for a primary school integrating solar photovoltaic (PV), wind turbine (WT), battery energy storage system (BESS), and utility grid support to achieve cost-effective and low-emission electricity generation. The optimized system achieved a minimum cost of energy (COE) of $0.0325/kWh and a net present cost (NPC) of 12,765, while reducing annual carbon emissions to 2,021 kg/yr, corresponding to only 20.79% of conventional fossil-fuel-grid emissions. To enhance operational intelligence, an XGBoost-based machine learning framework was developed for renewable energy curtailment prediction and curtailed power estimation. The proposed framework achieved 99.83% classification accuracy and 97.97% regression $\mathbf{R}^{\mathbf{2}}$, indicating excellent predictive reliability. Furthermore, SHAP-based interpretability analysis revealed Solar PV Output and Grid Sales as the dominant curtailment-driving parameters. The proposed framework demonstrates an effective pathway toward intelligent, low-carbon, and economically sustainable school microgrid operation.