Representation-Learning-Based CNN for Intelligent Attack Localization and Recovery of Cyber-Physical Power Systems.

Area of Science:

• Electrical Engineering
• Computer Science
• Cybersecurity

Background:

• Cyber-physical power systems (CPPSs) integrate communication, computation, and control, leading to increased complexity and new security vulnerabilities.
• Malicious cyber-attacks like false data injection, jamming, and denial of service pose significant threats to CPPSs, necessitating robust detection and recovery mechanisms.

Purpose of the Study:

• To develop an intelligent method for accurately localizing cyber-attacks in CPPSs.
• To enable automatic system recovery by effectively filtering out measurements affected by cyber-attacks.

Main Methods:

• A representation-learning-based convolutional neural network (RL-CNN) is proposed for multilabel classification to detect the location of cyber-attacks.
• The RL-CNN leverages implicit measurement information for improved attack localization in complex CPPSs.
• A mean-squared estimator is employed for automatic attack filtering and system state estimation, utilizing prior system state knowledge.

Main Results:

• The proposed RL-CNN method demonstrates high accuracy in localizing cyber-attacks within CPPSs.
• The integrated approach effectively performs automatic attack filtering, facilitating system recovery.
• Simulations on IEEE bus systems validate the method's performance under diverse cyber-attack scenarios.

Conclusions:

• The RL-CNN offers a powerful tool for enhancing the security and resilience of CPPSs against sophisticated cyber-attacks.
• Accurate attack localization and automatic filtering are crucial for maintaining the stable operation of smart power grids.