Research On Application Of Nonlinear Filtering In Navigation System

Currently, extended Kalman filtering (EKF) has been one of the main methods for studying the two problems of initial alignment and INS/GPS integrated navigation that have nonlinear characteristics. Because EKF necessitate the linearization of the original nonlinear models, the performances achieved through using EKF are limited to some extent. In recent years, unscented Kalman filtering (UKF) and particle filtering (PF) are becoming the focus of nonlinear estimation.The dissertation investigates the two nonlinear filtering methods, namely UKF and PF, and then applies them to the two problems of initial alignment and INS/GPS integrated navigation, concentrating on the impacts of various nonlinear characteristics. The researches on nonlinear filtering are organized as following:1. UKF is analyzed and extended from the aspect of representative points of random distribution;2. UKF for two classes of hybrid system are investigated and a specific simplified UKF (SUKF) is raised.3. The representative points of random distribution are applied to PF, resulting in a deterministic PF using the representative points rather than random samples which has better performance than the original PF.Consequently, the above results are applied to the two problems of initial alignment and INS/GPS integrated navigation. The main work consists of the following aspects:1. The nonlinear models of SINS are investigated. The relationship among them is completed and a unified description is presented.2. UKF is applied to nonlinear initial alignment of SINS on the stationary base. Simulation and experiment results on the three-axis table show that the nonlinear initial alignment not only release the constraint on the azimuth error (within 10 degree), but result in a good estimate.3. UKF and PF are applied to INS/GPS integrated navigation. The influences of nonlinear parts on the performance are investigated by means of simulation, rotating table, and filed tests. The results show that using nonlinear process equation can further improve the performance, but using nonlinear pseudo-range equation has no significant improvement compared to the linearized pseudo-range equation.