State Estimation of Memristor Neural Networks with Model Uncertainties

Abstract

This paper is concerned with the problem of state estimation of memristor neural networks with model uncertainties. Considering the model uncertainties are composed of time-varying delays, floating parameters and unknown functions, an improved method based on long short term memory neural networks (LSTMs) is used to deal with the model uncertainties. It is proved that the improved LSTMs can approximate any nonlinear model with any error. On this basis, adaptive updating laws of the weights of improved LSTMs are proposed by using Lyapunov method. Furthermore, for the problem of state estimation of memristor neural networks, a new full-order state observer is proposed to achieve the reconstruction of states based on the measurement output of the system. The error of state estimation is proved to be asymptotically stable by using Lyapunov method and linear matrix inequalities. Finally, two numerical examples are given, and simulation results demonstrate the effectiveness of the scheme, especially when the memristor neural networks with model uncertainties.