Adaptive Variational Bayesian Extended Kalman Filtering for Nonlinear Systems

Abstract

Definite modeling and known invariant parameters (including system parameters and noise statistics) are prerequisites of the well-known extended Kalman filtering (EKF). Naturally the performance of EKF may be degraded due to the fact that the statistics of measurement noise might change in practical situations. For nonlinear systems, an adaptive variational Bayesian extended Kalman filtering (AVBEKF) algorithm is developed in this paper. This algorithm regards both the system state and time-variant measurement noise as random variables to estimate. It provides a scheme that variances of measurement noises are approximated by variational Bayes, and thereafter system states are estimated at standard update step. Simulation results demonstrate that, in the context of a nonlinear model, the performance of the proposed filter is unaffected by the time-variant noise and AVBEKF is capable of tracking measurement noise as well.