ADAPTIVE LINEAR OBSERVER FOR NONLINEAR SYSTEMS

Abstract: In this paper, we describe the use of Luenberger state estimators for general nonlinear, time-varying systems. Since, in general, it is difficult to determine globally stable pre-specified observer gains for nonlinear systems, we propose using an adaptive observer gain vector that will allow learning of stable values throughout the state estimation process. To this end, we will derive a stochastic gradient adaptation algorithm for the observer gains based on the mean-square error of the estimated outputs. The performance of the adaptive observer scheme will be tested on linear and non-linear systems, including the chaotic Lorenz attractor. Copyright © Controlo 200