Optimal Stealthy Deception Attack Against Cyber-Physical Systems.

Area of Science:

• Control Theory
• Cybersecurity
• Optimization

Background:

• Deception attacks pose a significant threat in control systems.
• Detecting and mitigating these attacks is crucial for system security.
• Quantifying attack stealthiness and impact is an ongoing challenge.

Purpose of the Study:

• To design an optimal deception attack that maximizes a defined utility function.
• To incorporate the Kullback-Leibler divergence as a constraint for attack detection.
• To analyze the stealthiness of the deception attack and its properties.

Main Methods:

• Decomposing the attack signal into stealthy and non-stealthy components.
• Deriving necessary and sufficient conditions for bounded attack benefits.
• Proving the optimal attack's linear transformation follows a Gaussian distribution.
• Transforming the problem using Gaussian distribution parameters and proposing a suboptimal policy.

Main Results:

• A necessary and sufficient condition for non-unbounded benefits of strict stealthy attacks was established.
• The optimal attack's linear transformation was shown to be a Gaussian distribution.
• A suboptimal attack policy was developed, with an upper bound on benefit loss provided.
• The effectiveness was validated through a numerical example involving an unmanned ground vehicle.

Conclusions:

• The proposed suboptimal attack policy offers a practical approach to deception attacks.
• The study provides a framework for analyzing stealthy attacks in control systems.
• The findings are applicable to securing unmanned ground vehicles and similar systems.