Non-fragile [formula omitted] filtering for a class of switched neural networks

Abstract

This paper is devoted to non-fragile L2−L∞ filtering for switched neural networks with time-variant delay. The aim is to design a L2−L∞ filter subject to either additive or multiplicative gain perturbations, such that the filter-error system not only is asymptotically stable when there is no external disturbance but also has a predefined disturbance attenuation index under the zero initial condition. A criterion of the stability and L2−L∞ performance for the filter-error system is proposed by applying mode-dependent Lyapunov functionals, the Bessel–Legendre inequality, as well as the reciprocally convex combination technique. Then, a design method for the non-fragile L2−L∞ filter is developed by getting rid of some nonlinear coupling terms. The method is formulated as a problem of finding a feasible solution to a collection of linear matrix inequalities, which are computationally tractable. At last, two numerical examples are employed to illustrate the applicability of the L2−L∞ filtering design method.