Research on an Enhanced Exponential Smoothing Method in Campus Electricity Demand Forecasting

Abstract

Campus electricity consumption exhibits distinct periodicity, high volatility, and susceptibility to external disturbances. Therefore, accurate forecasting is crucial for optimizing power allocation and reducing operational costs. However, traditional exponential smoothing methods often fail to effectively capture the differences in electricity consumption between weekdays and weekends, resulting in limited prediction accuracy. To address this issue, this study improves the traditional exponential smoothing method by introducing a binary seasonal factor. This approach successfully distinguishes between weekday and weekend electricity consumption patterns without increasing model complexity. Validation using simulated campus hourly electricity consumption data demonstrates that the improved model significantly outperforms the traditional model. Experiments reveal substantial reductions in mean squared error, root mean squared error, and mean absolute percentage error for the new method. These findings demonstrate that this low-cost, interpretable, and efficient tool will support campus logistics in electricity procurement and peak-valley shifting strategies, contributing to low-carbon campus development. Future work will explore adaptive seasonal factors and their integration with machine learning models.