A Systematic Review of Statistical and Machine Learning Methods for Electrical Power Forecasting with Reported MAPE...

Area of Science:

• Electrical Engineering
• Computer Science
• Data Science

Background:

• Accurate electric power forecasting is essential for managing power systems, ensuring grid stability, and optimizing resource allocation.
• Reliable demand predictions are vital for efficient power generation, distribution, and mitigating risks associated with grid integration.

Purpose of the Study:

• To conduct a systematic literature review to identify the most precise electric power forecasting models.
• To compare the accuracy of classical statistical/mathematical (MSC) models against machine learning (ML) models.
• To evaluate the contribution and performance of hybrid models in electric power forecasting.

Main Methods:

• Systematic literature review of 257 accuracy tests across five geographic regions.
• Comparison of forecasting model performance: statistical/mathematical (MSC) vs. machine learning (ML).
• Analysis of hybrid model effectiveness and case study application.

Main Results:

• Forecasting errors are minimized by reducing the time horizon.
• Machine learning (ML) models integrating diverse exogenous variables demonstrate superior forecast accuracy.
• A significant increase in the accuracy of electric power forecasting models observed over the past five years.

Conclusions:

• Machine learning models, particularly those utilizing exogenous variables, are key to improving electric power forecast accuracy.
• Reducing the forecasting time horizon effectively minimizes prediction errors.
• The field of electric power forecasting has seen substantial accuracy improvements recently, driven by advanced modeling techniques.