Resilient neural network-based control of nonlinear heterogeneous multi-agent systems: a cyber-physical system approach

Resilient control and cyber security are of great importance for multiagent systems (MASs) due to the vulnerability during their coordination through information interaction among agents. Unexpected cyber-attacks on a single agent can spread quickly and can affect critically the safety and the performance of the system. This paper proposes a distributed adaptive resilient neural network (NN)-based control procedure to guarantee its stability and to make the agents to follow the leader's profile when there exist cyber-attacks and external disturbances in the system. Firstly, an adaptive neural network is designed to estimate the nonlinear part of the MAS. Then, a variable structure super twisting approach is proposed in which the adaptive weights of the NN, and a virtual disturbance are designed adaptively via updating laws. Moreover, stability criteria and control objectives are investigated through Lyapunov theorem, which leads to a novel scheme that can handle nonlinearity, cyber-attacks, and external disturbances without requirement of designing different controllers in an extra algorithm and considering any limiting predefined condition such as Lipschitz on the nonlinear function. As an application, an adaptive cruise control of a platoon of connected automated vehicles is considered to scrutinize the proposed procedure in the absence and the presence of cyber-attacks, which verify the theoretical results.