Neural Network-Based Attack-Compensation Control of T–S Fuzzy Systems Against Actuator Attacks With Improved Dynamic Memory-Event-Triggered Scheme

This paper investigates the neural network-based event-triggered attack-compensation control problem of T-S fuzzy systems in the presence of actuator attacks. In order to enhance the triggering performance, a novel improved dynamic memory-event-triggered mechanism is developed. Three significant features of this mechanism are: 1) the average of some memory outputs is used rather than the current output to decrease the triggering events raised from frequent data jitter; 2) a dynamic variable related to the memory-based triggering condition is introduced to enlarge the triggering interval; 3) Jensen inequality is applied to relax the conventional triggering condition expressed as the quadratic form of the average of memory signals to further improve the triggering efficiency. To execute the state- feedback control and mitigate the impact of unknown malicious attacks injected in the actuator, a T-S fuzzy observer and a radial basis function neural network-based estimator are designed to estimate the system state and actuator attack, respectively. Based on the combination of them, an attack-compensation controller is constructed. Then, with the aid of Simpson's second rule to tackle the variable limit integral functions resulting from the memory signal and communication delay, several sufficient criteria are derived for assuring that the system is uniformly ultimately bounded with $H_{\infty }$ performance. Lastly, the benefits of the proposed dynamic memory-event-triggered mechanism and attack-compensation controller are confirmed through some simulation results.