A Regression-Based Method for Monthly Electric Load Forecasting in South Korea

Abstract

In this study, we propose a regression-based method for forecasting monthly electricity consumption in South Korea. The regression model incorporates key external variables such as weather conditions, calendar data, and industrial activity to capture the major factors influencing electricity demand. These predictor variables were identified through comprehensive data analysis. Comparative experiments were conducted with various existing methods, including univariate time series models and machine learning techniques like Holt–Winters, LightGBM, and Long Short-Term Memory (LSTM). Additionally, ensemble methods combining two or more of these existing methods were tested. In the empirical analysis, the proposed model was used to forecast monthly electricity demand for a 24-month period (2022–2023), achieving a mean absolute percentage error (MAPE) of approximately 2%. The results demonstrated that the proposed method consistently outperforms all benchmarks tested in this study.

Keywords

• mid-term load forecasting
• regression
• interaction effects
• machine learning