Neural-network-based adaptive secure control for nonstrict-feedback nonlinear interconnected systems under DoS attacks

This paper investigates the problems of decentralized adaptive neural network (NN) secure control for a class of nonstrict-feedback nonlinear interconnected large-scale systems (NILSs) under denial-of-service (DoS) attacks. The considered systems contain not only unknown interconnected terms but also general nonlinear functions that are not required to be globally Lipschitz, in contrast to most of the existing results in the area. When the system subject to the DoS attack, the sampled data can neither be sent by the sensors nor received by the actuators due to the appearance of DoS in the sensor-to-controller (S-C) and the controller-to-actuator (C-A) communication channels. The data packets sent from the sen-sors are blocked and the traditional backstepping technology cannot be adopted. In order to solve this difficulty, a novel dynamic gain observer is constructed, which can obtain the desired states from the NILSs under DoS attacks. Considering the energy limitation of attackers, two reasonable Assumptions of DoS attack intensity are given. The stability is proven based on the Lyapunov stability theory (LST) and the improved average dwell time (ADT) method. The proposed controller guarantees that all signals in closed-loop systems are bounded, while the tracking error signals converge to a small residual set. Finally, two simulation examples are provided to illustrate the effectiveness of the proposed control method. (c) 2021 Elsevier B.V. All rights reserved.

Automated summary

This paper investigates the problems of decentralized adaptive neural network secure control for a class of nonstrict-feedback nonlinear interconnected large-scale systems under denial-of-service attacks. A novel dynamic gain observer is constructed to solve the difficulty of sampling in the presence of DoS attacks. The stability is proven based on Lyapunov stability theory and the improved average dwell time method.