The growing global energy crisis has heightened the importance of renewable energy sources, making accurate short-term power demand forecasting crucial for maintaining supply-demand balance, minimizing curtailment, and optimizing renewable energy integration. However, empirical forecasting studies focusing on regions with a high share of renewable energy remain limited. To address this gap, this study conducts short-term power demand forecasting for the next 24 hours using Jeju Island’s electricity load data from January 2017 to September 2023. It examines (1) overall accuracy, (2) seasonal variations, (3) alternative data partitioning, and (4) a focused analysis of the most recent three years of intensified curtailment. This study compares three forecasting models?Multiple Linear Regression (MLR), Long Short-Term Memory (LSTM), and CatBoost. The results show that MLR demonstrates satisfactory prediction accuracy, fast training speed, and robustness against overfitting. The findings reveal that in regions where the share of renewable energy is rapidly expanding and output curtailment issues are intensifying, traditional forecasting techniques such as MLR can still be effectively utilized alongside advanced models. These results can contribute to the development of effective energy management strategies and support more stable and efficient power supply operations during the transition to sustainable energy systems.