A clustering fractional-order grey model in short-term electrical load forecasting

Area of Science:

• Electrical Engineering
• Data Science
• Applied Mathematics

Background:

• Accurate short-term electrical load forecasting is crucial for energy demand management.
• Power consumption data exhibit non-stationary, nonlinear, and multi-dimensional characteristics, posing prediction challenges.
• Fractional-order partial differential equations show promise in modeling complex power consumption behaviors.

Purpose of the Study:

• To introduce a novel Clustering Fractional-order Generalized Predictive Control (C-FGM) model for short-term electrical load forecasting.
• To leverage fractional-order calculus and clustering for improved prediction accuracy in electrical power systems.

Main Methods:

• Developed the C-FGM model, incorporating a parameter α to capture cumulative weather trends across clustered sub-series.
• Utilized fractional-order partial differential equations to model historical power series, with hyper-parameters optimized globally.
• Validated the model on two real-world electricity datasets.

Main Results:

• The C-FGM model efficiently learns hyper-parameters directly from datasets for accurate forecasting.
• Achieved superior accuracy compared to LSTM (MAPE 1.97-4.67% vs. 4.34%) and Transformer (5.42%).
• Demonstrated significant improvements in Mean Absolute Percentage Error (MAPE).

Conclusions:

• The C-FGM model offers a robust and accurate solution for short-term electrical load forecasting.
• This data-driven approach provides an effective tool for real-time energy demand management.
• Fractional-order modeling combined with clustering enhances predictive capabilities for complex power systems.