SVD-CKF Filtering Algorithm And Its GNSS/INS Navigation Application Research

Inertial Navigation System(INS)and Global Navigation Satellite Systems(GNSS)are two types of navigation systems with complementary advantages.Therefore,the GNSS / INS integrated navigation system has become a research hotspot and has been widely used in military and people's daily life.With the growing demand for position,there are higher requirements for the accuracy and reliability of integrated navigation systems.Generally,the accuracy and reliability of integrated navigation systems are improved from two aspects of modeling and parameter estimation.This paper studies the filtering algorithms in the data processing of GNSS / INS integrated navigation systems,and focuses on the nonlinear errors of integrated navigation systems and the model errors of filtering models to improved the filtering model.The main research contents are as follows:(1)The coordinate systems that are commonly used in GNSS and INS systems and their transformation relations,respective characteristics,and error models are briefly introduced.(2)The common filtering methods in GNSS / INS integrated navigation systems are introduced: linear Kalman filter algorithm and nonlinear cubature kalman filter algorithm,and the latter is improved.Singular value decomposition is used instead of Cholesky decomposition to avoid system state covariance matrix appears non-positive.Subsequently,the two algorithms are compared and analyzed experimentally.The results show that the improved SVD-CKF algorithm can more effectively control non-linear errors and achieve higher filtering accuracy.(3)For the integrated navigation system filter's model error and observed gross error,three robust models are established: Robust cubature kalman filter based on singular value decomposition(R-SVD-CKF),Strong tracking robust cubature kalman filter based on singular value decomposition(STR-SVD-CKF)and Categorical adaptive robust cubature kalman filter based on singular value decomposition(AR-SVD-CKF).After comparative analysis of various experimental schemes,the results show that the R-SVD-CKF algorithm can suppress the effect of the observation gross error,and the STR-SVD-CKF algorithm can control the influence of state estimation error and the observation gross error at the same time,AR-SVD-CKF algorithm can also control the influence of state perturbation error while weakening state estimation error and observation gross error,this algorithm further improve the stability of the integrated navigation system,and can provide some reference for GNSS / INS integrated navigation data processing.(4)Considering the instability of GNSS signals,the Elman neural network is introduced into the integrated navigation data processing work,combined with the AR-SVD-CKF algorithm constructed earlier.The network is trained when the GNSS signal is normal,and after the GNSS is locked,the Elman network is used to predict the output.Experiments show that the algorithm can control the filtering accuracy in a short period of time,and the longest speed prediction time is 200 s and the longest eastward position prediction time is 50 s,the longest prediction time for the northward position is 20 s.