FedAvg-P: Performance-Based Hierarchical Federated Learning-Based Anomaly Detection System Aggregation Strategy for...

Area of Science:

• Cybersecurity
• Electrical Engineering
• Artificial Intelligence

Background:

• Advanced Metering Infrastructures (AMIs) are crucial for efficient electrical systems but collect vast data, increasing vulnerability to cyberattacks.
• Centralized data storage in AMIs poses privacy risks and creates a Single Point of Failure (SPoF).
• Federated Learning (FL) offers a decentralized approach but faces challenges with client performance and global model reliability.

Purpose of the Study:

• To develop a performance-based hierarchical federated learning (HFL) anomaly detection system for AMI networks.
• To enhance the security and reliability of critical electrical infrastructure against sophisticated cyber threats.
• To address data privacy concerns and mitigate the risk of system failure in decentralized learning models.

Main Methods:

• Developed a deep learning model for detecting attacks targeting AMI critical infrastructure.
• Introduced a novel aggregation strategy, FedAvg-P, to improve the global performance of federated learning models.
• Proposed a peer-to-peer architecture to eliminate Single Points of Failure (SPoF) in the HFL system.

Main Results:

• The proposed HFL system effectively detects anomalies and potential attacks within AMI networks.
• The FedAvg-P aggregation strategy demonstrated enhanced global performance compared to standard methods.
• The peer-to-peer architecture successfully guarded against system failures, ensuring continuous operation.

Conclusions:

• The developed hierarchical federated learning anomaly detection system provides a reliable solution for securing AMI networks.
• The study confirms the effectiveness of the proposed deep learning model, aggregation strategy, and decentralized architecture.
• This research contributes to the resilience and security of smart grid infrastructure against evolving cyber threats.