A Multiple-Model State Estimator based on the Unscented Kalman Filter

Abstract

We consider the state estimation of nonlinear dynamic systems based on the framework of the Kalman filter. There have been various nonlinear Kalman filters in literatures for dealing with the estimation problem, including the well-known extended Kalman filter (EKF) and the unscented Kalman filter (UKF). In this paper, we propose a novel algorithm based on the UKF to promote estimation performance. With the application of multiple-model estimation, our proposed filter is formulated by using the fusion of multiple UKFs. Overall estimation rules and corresponding posterior model probabilities are provided as well. Furthermore, we present a convenient method to design the model set, which is based on the unscented transformation (UT). Numerical examples show that the new filter gains competitive performance compared with the EKF and the UKF in both accuracy and robustness.