An Anti-Attack Neural Sliding Mode Framework Based on a Novel Non-Fragile Observer

This article investigates anti-attack stabilization with passivity problem of uncertain singular semi-Markov jump systems (singular S-MJSs) with exogenous disturbance and delay. An ingenious non-fragile observer-based neural sliding mode control (SMC) scheme is put forward to solve the problem. First, considering unmeasured states, a distinctive non-fragile and decoupled observer, which does not contain the control input or any auxiliary sliding mode compensator design as in existing observer-based SMC approaches, is established such that the disadvantages of sliding mode switching in observers in existing literature can be avoided. Then, “only one sliding surface” design and a new system analysis route are presented, and the derived sliding surface is accessibly designed. Next, a new version of stochastic admissibility and passivity sufficient condition is given, and a related algorithm via an optimization problem is proposed to determine the controller gain and the observer gain by linear matrix inequalities (LMIs). Further, a novel observer-based anti-attack neural SMC law, which utilizes a neural network-based approach to approximate actuator attack, is proposed to stabilize the singular S-MJSs against actuator attack. Finally, simulation and comparison results are presented, which demonstrate the effectiveness and superiority of our method.