Neural-network-based asynchronous security control for delayed Markov jump systems with imperfect transition probabilities via dynamic-memory event-triggered mechanism

In this article, we tackle the problem of realizing boundedness for delayed Markov jump systems (MJSs) with deception attacks and imperfect transition probabilities. To alleviate the burden of communication and conserve bandwidth in the communication network channel, a novel model-dependent dynamic-memory event-triggered mechanism (DMETM) is designed. Different from the general dynamic event-triggered mechanism (ETM), the designed DMETM includes historical trigger data in its triggering conditions, so it is able to dynamically adjust data transmission according to the long-term changes in the system state, thereby improving resource utilization. A hidden Markov model (HMM) with unmeasured probabilities is employed to comprehensively characterize plant-controller asynchrony. Moreover, deception attacks, considering the vulnerability of communication links, are taken into account. While existing studies typically assume bounded false data injection in MJSs, establishing accurate bounds in practical scenarios remains challenging. Considering this problem, a neural network-based asynchronous memory compensation feedback controller is developed, which effectively identifies and mitigates the impact of deception attacks. By leveraging a Lyapunov-Krasovskii functional (LKF), sufficient conditions for bounded in probability of the closed-loop system are derived. The effectiveness of the theoretical analysis is illustrated through an illustrative example.