Robustification of Finite Impulse Response Filter for Nonlinear Systems With Model Uncertainties

This article investigates an iterative finite impulse response (FIR) filter for a nonlinear system with model uncertainties. We begin by designing a simple initialization method for the iterative FIR structure and use the unscented transformation (UT) to realize iterative state estimation in each finite horizon. Additionally, a compensation variable is introduced to compensate for the inaccuracies in the nonlinear dynamic model, thereby reducing uncertainties and improving the estimation accuracy. To estimate the compensation variable, we employ the variational Bayesian (VB) method. Compared with other filters, our design can not only avoid calculating the Jacobian matrix and batch processing to reduce computational load but also enhance robustness. Simulation results demonstrate the validity of the proposed filter.