Observer-Based Optimal Backstepping Security Control for Nonlinear Systems Using Reinforcement Learning Strategy.

Area of Science:

• Control Systems Engineering
• Artificial Intelligence
• Cybersecurity

Background:

• Nonlinear systems are susceptible to deception attacks, compromising control system integrity.
• Existing optimal control strategies often rely on strict assumptions about deception attacks, limiting their applicability.
• Reinforcement learning (RL) offers a promising approach for adaptive control but requires robust designs against adversarial conditions.

Purpose of the Study:

• To develop an observer-based optimal backstepping security control for nonlinear systems resilient to deception attacks.
• To introduce an improved security RL algorithm utilizing neural networks to enhance attack resistance and system optimization.
• To relax common assumptions regarding deception attacks by designing an observer that estimates states from attacked outputs.

Main Methods:

• An observer-based optimal backstepping security control framework is proposed.
• A critic-actor based reinforcement learning (RL) algorithm with neural network technology is employed.
• An advanced observer is designed to estimate system states using attacked outputs, mitigating deception impacts.

Main Results:

• The proposed RL algorithm effectively resists deception attacks in nonlinear systems.
• The developed observer successfully estimates unavailable states, eliminating the impact of attacks.
• The control strategy ensures semi-globally ultimately bounded signals within the closed-loop system.

Conclusions:

• The observer-based optimal backstepping security control strategy effectively enhances the security and performance of nonlinear systems under deception attacks.
• The integration of RL and neural networks provides a robust framework for resilient control.
• The study demonstrates the effectiveness of the proposed method through simulation experiments.