Adaptive Neural Network-Based Resilient Output Feedback Control of Cyber-Physical Systems Under Multi-Channel Stochastic False Data Injection

Cyber-physical systems (CPSs) are prone to cyber-attacks, which can cause them to malfunction in the presence of a malicious attacker. To resolve this dilemma, an observer-based controller based on neural networks (NNs) has been employed in this study to provide a resilient output feedback approach for CPSs exposed to unknown stochastic cyber-attacks. We have examined the impact of multi-channel stochastic false data injection (FDI) attacks on the network, specifically targeting the system's output and the controller's connection to the actuator channel. Adaptive NN-based compensation signals are designed as a method for estimating attacks, compensating for their effects, and controlling system performance for control applications. The system states are reconstructed using a modified Luenberger observer, which is fed with corrupted information from the output of a system. An estimation of the true system states and true input to the actuator are achieved with the aid of novel compensation signals based on intelligent adaptive NNs. Using Lyapunov theory, it has been demonstrated that the proposed output feedback approach guarantees stability under stochastic attacks along with bounded control and adaptation parameters. A simulation of an industrial process is performed to showcase the effectiveness of the proposed approach.