Sensitivity reduction of unscented Kalman filter about parameter uncertainties

Abstract

This study aims to reduce the sensitivity of the unscented Kalman filter (UKF) estimates with respect to uncertain model parameters, leading to a more robust UKF. The standard minimum-variance cost index is augmented to include a penalty on the sensitivities of the state estimates about parameter uncertainties in the form of a weighted norm. A new filter gain is thus obtained, and the desensitised UKF (DUKF) provides better estimation of the true states than the standard UKF with an imperfect plant model. Numerical examples are shown to demonstrate the efficacy of the DUKF.