A fast convergent extended Kalman observer for nonlinear discrete-time systems

A new extended Kalman observer for nonlinear discrete-time systems is derived. As always, stability and convergence rate depend directly on the matrices Q k and R k and the initial state estimation error, the new observer has made improvement on these two aspects. A new criterion for the design of matrices Q k and R k enables the observer to enlarge the convergent domain, an off-line-trained neural network can significantly reduce the initial state estimation error without increasing on-line computation burden. The integration of the techniques results in a stable and fast convergent observer. The observer performance is demonstrated with the estimation of flux and angular speed of an induction motor.