RBFNN-estimation-based data-driven iterative learning consensus control for a class of heterogeneous nonlinear MASs under DoS attacks

This paper proposes a data-driven iterative learning control (DDILC) scheme based on the radial basis function neural network (RBFNN) to achieve consensus control for nonlinear heterogeneous multi-agent systems under DoS attacks. First, a linear data model (LDM) is presented for the heterogeneous agents by the dynamic linearisation method. The data loss caused by DoS attacks can be effectively resolved through an output compensation mechanism. Subsequently, an RBFNN estimation algorithm is used to approximate the fast varying pseudo partial derivative of LDM. With linearisation on the ideal iterative learning controller, the need to pre-determining the controller structure is eliminated. Then, the learning gain is optimised using Newton-type optimisation and a local ILC protocol relying on neighbouring agents' information is designed. Finally, the convergence of the RBFNN-based DDILC scheme is rigorously analysed and a simulation of four permanent magnet motors is used to demonstrate the validity of the proposed scheme.