Nonlinear Filtering Based on Model Prediction

Abstract

Nonlinear filtering is of great importance in many applied areas. As a typical nonlinear filtering algorithm, the unscented Kalman filter (UKF) has the merits such as simplicity in realization, high filtering precision, and good convergence. However, its filtering performance is very sensitive to system model error. To overcome this limitation, this paper presents a new UKF for state estimation in nonlinear systems. This algorithm integrates model prediction into the process of the traditional UKF to improve the filtering robustness. This algorithm incorporates system driving noise in system state by increasing the state space dimension to expand the input of system state information to the system. The system model error is constructed by model prediction to rectify the system estimation from the traditional UKF. Simulation and experimental analyses have been conducted, showing that the proposed filtering algorithm is superior to the existing nonlinear filtering algorithms such as the EKF and traditional UKF in terms of accuracy.