Nonlinear optimal observer design using neural networks and state dependant Riccati equation

The difficulty in measuring all the required states of a system necessitates the use of observers/estimators. The use of Extended Kalman Observer/Filter (EKO/EKF) for providing optimal state estimates for nonlinear system is well documented in literature. However the linearization assumption inherent in EKO/EKF implementation hinders its estimation characteristics. In this work a new optimal method for estimating states of a nonlinear system is presented. The observer design is posed as an optimal output tracking problem by defining an appropriate cost function. The correction to the observer states is shown to be function of the solution of the Two Point Boundary Value Problem (TPBVP) resulting from minimizing the cost function --Abstract, page iii