Adaptive critic based neuro-observer

摘要

A new Neural Network (NN) based observer design method for nonlinear systems represented by nonlinear dynamics and linearonlinear measurement is proposed in this paper. In this new approach, as the first step, the observer design problem is changed into a 'controller' design problem by establishing the error dynamics, and then the Adaptive Critic (AC) based approach is applied on this error dynamics to design a 'controller', such that the errors are driven to zero. The resulting observer has inherent robustness from the AC based design approach. Some simulations are presented to illustrate the effectiveness of this approach.