SINS/GPS Integrated Navigation Based On Variational Bayesian Filtering Method

Considering that the measurement noise of SINS/GPS integrated navigation system may be heavy-tailed and time-variant,which may cause the decline of the filtering precision,we applied a Student-t distribution to model the measurement noise.Combined variational Bayesian learning and Kalman filtering method,this paper proposed variantional Bayesian filtering methods based on Student-t distribution.The main contents and achievements of this paper are summarized as follows:(1)Aiming at the problem that the accuracy of traditional Kalman filter is reduced due to the presence of heavy-tailedness in the measurement noise,this paper proposed a variation Bayesian Gaussian filtering algorithm based on Student-t distribution for SINS / GPS integrated navigation system.The proposed algorithm considered the heavy-tailed characteristics of measurement noise and modeled the noise as the Student-t distribution.With the help of the variational Bayesian learning and the Gaussian filtering method,we could estimate the joint posterior distribution of the system state variable and the latent variable,the algorithm is robust to outliers and outperforms than those traditional methods.(2)Considering that the filtering precision of the proposed variational Bayesian Gaussian filter will degrade under time-variant noise,the paper proposed a variational Bayesian Expectation Maximizaiton Algorithm based on the Student-t distribution for the SINS / GPS integrated navigation system.This algorithm model the measurement noise as a multi-dimensional Student-t distribution,and combine the Expectation Maximizaiton method with variational Bayesian learning,which iteratively update the three hyper-parameters of the Student-t distribution.This algorithm is more adaptive,and can achieve better filtering performance under time-variant noise than variational Bayesian Gaussian filter.(3)The principle of each part of the SINS/GPS integrated navigation system was studied and the corresponding error modeling was carried out.As for Bayesian filtering method,the state equation and observation equation of SINS/GPS integrated navigation system was constructed,and the simulation of the full set of SINS/GPS integrated navigation system platform was completed in the MATLAB.The simulation trajectory based on the basic mode of motion,and test simulation data was collected for filtering.And based on this simulation platform,the corresponding algorithms were simulated.