A Secure Control Learning Framework for Cyber-Physical Systems Under Sensor and Actuator Attacks.

Area of Science:

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

Background:

• Cyber-physical systems (CPS) are vulnerable to sophisticated sensor and actuator attacks.
• Ensuring the security and reliability of CPS is critical for their widespread adoption.

Purpose of the Study:

• To develop a learning-based secure control framework for CPS.
• To address the challenges posed by sensor and actuator attacks.
• To enhance the resilience and trustworthiness of CPS.

Main Methods:

• Utilized a bank of observer-based estimators for attack detection.
• Introduced a threat-detection level function for monitoring measurement reliance.
• Formulated a two-player, zero-sum differential game for attack mitigation.
• Employed a reinforcement-learning-based algorithm for learning the secure control policy.

Main Results:

• Successfully detected and mitigated sensor and actuator attacks in simulated environments.
• Demonstrated the efficacy of the learning-based secure control framework through numerical examples.
• Achieved robust system operation under nominal and adversarial conditions.

Conclusions:

• The proposed framework effectively enhances the security of cyber-physical systems against sophisticated attacks.
• Learning-based approaches, combined with differential game theory, offer a promising direction for secure CPS design.
• The framework provides a robust solution for maintaining system integrity and performance in the presence of cyber threats.