Go to IEEE ICIST 2024 Conference homepageData-Driven Attack Compensate Control of Neural Networks
In this paper, we propose a novel data-driven approach to compensate for the effects of cyber attacks on neural network-based control systems. As neural networks become increasingly integral to critical control applications, these systems face heightened vulnerability to adversarial attacks. Our approach utilizes historical data to theoretically analyze and compensate for deviations in control performance caused by such attacks. The method integrates attack detection with a compensation mechanism designed to adjust the control input in real-time, aiming to mitigate the impact of the attack. Through rigorous theoretical analysis, we demonstrate the potential of this approach to enhance system stability and performance in the presence of cyber threats.