An improved unscented Kalman filter for nonlinear systems with one-step randomly delayed measurement and unknown latency probability

Abstract

In this paper, an improved unscented Kalman filter is presented to achieve state estimation for nonlinear systems with one-step randomly delayed measurement and unknown latency probability. Firstly, a Bernoulli random variable is introduced to describe the situation of randomly delayed measurement. Then, a joint prior probability density function is obtained. Finally, in the improved unscented Kalman filter algorithm, the state vector and the latency probability are estimated by utilizing a variational Bayesian technique, and the covariance matrices are computed by using the unscented transform. A univariate nonstationary growth model and a Lorenz system are included to verify that the proposed algorithm can not only implement enhanced estimation accuracy compared with the existing methods but also provide the latency probability accurately.