Optimizing short-term energy demand forecasting: a comprehensive analysis using autoregressive integrated moving average method

Abstrak

This study addresses the critical gap in short-term electricity demand forecasting in South Sulawesi, where inconsistencies between projected and actual peak loads hinder daily operational planning, system stability, and investment efficiency. While previous studies have applied approaches such as fuzzy logic, ARIMA-ANN, and hybrid models, few have focused on simple, robust ARIMA-based models validated across different time spans for daily operational use. To address this, the autoregressive integrated moving average (ARIMA) model is implemented within the Box-Jenkins framework, using automated model selection through the pmdarima library and Akaike’s information criterion (AIC) to identify optimal parameter configurations. The study analyzes daily peak load data from 2018 to 2023, producing realistic forecasts with high accuracy. The selected ARIMA model achieves a mean absolute percentage error (MAPE) of 1.91% and a root mean square error (RMSE) of 38.123, demonstrating its effectiveness in capturing short-term load trends. These results confirm the suitability of ARIMA for short-term forecasting in energy systems and its potential to enhance operational decision-making, reduce forecasting errors, and improve investment planning. The study also establishes a methodological foundation for future development, including the integration of ARIMA with machine learning and the use of extended datasets to support strategic energy management.