Guaranteed cost filtering for discrete-time multi-layer neural networks with time-varying delays and unideal measurements

This paper is concerned with the guaranteed cost filtering problem for discrete-time multi-layer neural networks with unideal measurements and time-varying delays. First, the innovative state space model of multi-layer neural networks can be described by the weighted-nonlinear function, which means that there have connections among neural layers. Then, the unideal measurements are made up by combination of random sensor nonlinearity and partial missing measurements, where partial missing measurements is the product of two mutually independent stochastic variables and normal measurements. Moreover, by using proportionate-additive filter and constructing a unified Lyapunov function, a novel criterion is proposed so that the augmented filtering error system achieves robust stability and has a guaranteed cost index. Finally, simulation results are presented to demonstrate the effectiveness of the derived method.