탄소중립이행을 위한 기계학습 기반의 전력수요예측 방법론 고도화 연구

Research on Advancing Electricity Demand Forecasting Method Using Machine Learning to Implement Carbon Neutrality

As concerns over climate change and greenhouse gas emissions intensify, carbon neutrality has emerged as a critical global goal. Achieving this goal requires accurate electricity demand forecasting, which plays a pivotal role in maintaining grid stability and managing the variability of renewable energy. However, traditional models that focus on the mean often fail to capture the complexity of modern electricity demand patterns. To address this issue, this study proposes the Composite Variation Method(CVM) based on quantile regression. This approach reduces the bias in mean-based electricity demand forecasts and provides higher accuracy, particularly in scenarios with outliers or complex conditions. The CVM is applied to Random Forest and XGBoost models using hourly electricity demand data from Ontario, Canada. The results demonstrate a 9.75% performance improvement for Random Forest and a 1.31% improvement for XGBoost. These findings suggest that CVM enhances the precision of electricity demand forecasting, contributing to more effective energy management strategies essential for carbon neutrality.