A computationally efficient technique for state estimation of nonlinear systems

Abstract An estimation technique is presented for the class of nonlinear systems consisting of memoryless nonlinearities embedded in a dynamic linear system. The approach is based on a useful sampled-data nonlinear system simulation method, which involves the addition of an extra state variable for each nonlinear element. The nonlinear estimator is developed along the lines of the basic Kalman state estimation, using quasilinearization instead of the Taylor series linearization used in extended Kalman filters. It is demonstrated that this new method out performs the extended Kalman filter in terms of the mean-square error of the state estimate. This estimator was used effectively for state estimation in cases where the extended Kalman filter does not converge. Moreover the new method is directly applicable to feedback systems with multiple nonlinearities and stochastic disturbances.