Cyber-resilient machine learning framework for accurate individual load forecasting and anomaly detection in smart...

Area of Science:

• Electrical Engineering
• Computer Science
• Data Science

Background:

• Smart grid evolution necessitates accurate and secure electricity load predictions for efficient energy management.
• Individual Load Forecasting (ILF) is critical for smart grid reliability and operational efficiency.
• Cyber-security threats, such as false data injection attacks, pose risks to smart grid integrity.

Purpose of the Study:

• To propose a scalable and cyber-resilient methodology for electricity consumption forecasting at the individual smart meter level.
• To enhance Individual Load Forecasting (ILF) accuracy and reduce computational complexity using machine learning.
• To implement an Anomaly Detection Scheme (ADS) for identifying cyber-attacks in smart meter data.

Main Methods:

• Utilized K-MEANS Clustering and Neural Networks (KMEANS-NN) for enhanced ILF.
• Employed Principal Component Analysis based One-Class Support Vector Machine (PCA-OCSVM) as an Anomaly Detection Scheme (ADS).
• Validated the integrated model using five months of real-world smart meter data from Egypt, including simulated cyber-attacks.

Main Results:

• KMEANS-NN significantly reduced Mean Absolute Adjusted Percentage Error (MAAPE) by up to 40% and computational time from days to minutes.
• The proposed Anomaly Detection Scheme (ADS) achieved high accuracy (99.9%), sensitivity (99.8%), precision (99.9%), specificity (99.9%), and F1-score (99.8%) in detecting cyber-attacks.
• The integrated model demonstrated improved forecasting accuracy across various benchmark models (ARIMA, CTREE, MLP, NNETAR).

Conclusions:

• The integrated KMEANS-NN and PCA-OCSVM model offers a scalable, accurate, and cyber-resilient solution for electricity load forecasting in smart grids.
• The methodology significantly enhances forecasting performance and provides robust protection against cyber-attacks.
• The proposed system shows strong potential for deployment in large-scale smart grid environments, improving energy management and reliability.