CONGO²: Scalable Online Anomaly Detection and Localization in Power Electronics Networks.

Area of Science:

• Power Electronics
• Network Security
• Data Science

Background:

• Accurate detection of electronic disturbances is vital for grid stability and damage prevention.
• Current anomaly detection methods suffer from low accuracy, high computational costs, and require expert personnel.
• A scalable, online algorithm for in-field analysis of large-scale power electronics networks is urgently needed.

Purpose of the Study:

• To propose a fast, accurate, and scalable algorithm for simultaneous anomaly detection and localization in power electronics networks.
• To address the limitations of existing methods in terms of accuracy, computational cost, and data requirements.
• To enable reliable in-field deployment for real-time monitoring and analysis.

Main Methods:

• Introduced the stratified colored-node graph (CONGO²) algorithm for hierarchical modeling of correlated waveforms and sensors.
• Utilized aggregation of sensor changes with neighbor inputs to identify anomalies undetectable by single sensors.
• Focused on short time-frame changes for computational efficiency and minimal data storage.

Main Results:

• The CONGO² algorithm demonstrated high accuracy and reliability in detecting and localizing a wide spectrum of anomalies.
• The method is data-driven, requiring no prior training data, making it broadly applicable.
• Successfully detected and localized cyber and physical attacks in a 37-node distributed energy resources power grid.

Conclusions:

• The proposed CONGO² algorithm offers an efficient and reliable solution for online anomaly detection and localization in large-scale power electronic networks.
• Its computational efficiency and minimal storage needs make it ideal for in-field deployment.
• The method provides a significant advancement over existing techniques, enabling robust grid security and management.