Mixed-frequency fractional derivative grey model for forecasting electricity consumption

Abstract

Accurate medium-term and long-term electricity demand forecasts are essential for energy management. To this end, this paper proposes a mixed-frequency fractional derivative grey model, and systematically applies the mixed-frequency modeling method to electricity consumption forecasting across 31 regions in China for the first time. The model integrates mixed-frequency sampling mechanisms with fractional grey modeling techniques to uniformly process data features of different temporal frequencies, including monthly, quarterly, and annual indicators, thereby enhancing the ability to capture both long-term memory effects and high-frequency information in time series data. Additionally, a weight function based on the Weibull distribution is designed to improve the flexibility of feature modeling. The arithmetic optimization algorithm is introduced to adaptively optimize model parameters, thereby improving overall model stability and adaptability. Monte Carlo simulations verify the robustness of the proposed model under different frequency combinations, noise levels, and sample sizes. In the empirical forecasting of electricity consumption across 31 regions, the model demonstrates outstanding performance. Ablation experiments also confirm the critical role of the mixed-frequency mechanism in enhancing model performance. According to the forecasting results, by 2030, total electricity consumption in eastern, central, western, and northeastern China is projected to reach approximately 5843, 2170, 3752, and 607 billion kWh, respectively. This study not only fills the application gap of mixed-frequency grey modeling in large-scale regional electricity forecasting but also provides solid data support and methodological innovation for optimizing regional energy allocation.