Resilient Output Regulation for Cyber-Physical Systems Against False Data Injection Attacks: A Hybrid Iteration...

Area of Science:

• Cyber-physical systems engineering
• Control theory
• Machine learning for security

Background:

• Cyber-physical systems (CPSs) are vulnerable to false data injection attacks (FDIAs).
• The linear optimal output regulation problem (LOORP) is crucial for CPS stability and performance.
• Existing solutions often require exact system dynamics or stabilizing gains.

Purpose of the Study:

• To investigate the LOORP for discrete-time CPSs under FDIAs.
• To develop an online, adaptive control scheme resilient to data attacks.
• To reduce the reliance on precise system models and initial control parameters.

Main Methods:

• Decomposition of the LOORP under FDIAs into static and dynamic minimax problems.
• Development of model-based schemes for the decomposed problems.
• Proposal of a hybrid iteration (HI)-based Q-learning algorithm for online solution.
• Demonstration using a discretized LCL-coupled inverter-based distributed generation system.

Main Results:

• The proposed HI-Q-learning scheme effectively solves the LOORP under FDIAs.
• The scheme operates online without requiring exact system dynamics or initial stabilizing gains.
• Fast iteration speed and robust performance were achieved.
• Successful validation on a practical distributed generation system.

Conclusions:

• The HI-Q-learning approach provides an efficient and robust solution for secure CPS control.
• This method enhances the resilience of discrete-time CPSs against false data injection attacks.
• The findings contribute to the advancement of secure and adaptive control strategies for modern power systems.