Interpretable Short-Term Electrical Load Forecasting Scheme Using Cubist.

Area of Science:

• Electrical Engineering
• Artificial Intelligence
• Data Science

Background:

• Accurate daily peak load forecasting (DPLF) and total daily load forecasting (TDLF) are critical for efficient power system operation.
• Existing forecasting models may lack interpretability or struggle to adapt to evolving load trends.

Purpose of the Study:

• To develop an accurate and interpretable Cubist-based incremental learning model for DPLF and TDLF.
• To identify key internal and external factors influencing electrical load patterns.

Main Methods:

• Utilized a Cubist-based incremental learning approach for load forecasting.
• Employed time-series cross-validation to capture recent electrical load trends.
• Conducted variable importance analysis to determine significant predictive factors.

Main Results:

• The proposed model achieved an average Mean Absolute Percentage Error (MAPE) of 7.77%.
• The model achieved an average Coefficient of Variation of the Root Mean Square Error (CV(RMSE)) of 10.06%.
• Temperature, holidays, and recent past loads (one day and one week prior) were identified as significant factors.

Conclusions:

• The Cubist-based incremental learning model provides accurate and interpretable DPLF and TDLF.
• The model's performance demonstrates its suitability for practical power system management.
• Understanding the influence of external (temperature, holidays) and internal (past loads) factors enhances forecasting reliability.