Research On The Improvement Of Robust And Adaptive Filter For GNSS/INS Integrated Navigation

The two sub-navigation systems of GNSS/INS integrated navigation system exhibit complementary advantages and disadvantages.In outdoor scenarios,the navigation system can provide the position,velocity,and attitude information of the carrier.However,with the update of sensor technology and the changes in the positioning application market,the performance of GNSS/INS integrated navigation in civilian field is limited by hardware cost and cannot meet the user’s requirements for availability,reliability,and accuracy.Therefore,improving the navigation performance by algorithm has important research value.This thesis focuses on the robust and adaptive filter of GNSS/INS integrated navigation system.The main research contents are as follows:(1)Base on GNSS positioning principle and INS state update and error model,the lever arm and dynamic initial alignment in GNSS/INS integrated navigation system are expounded,and the system model and measurement model of GNSS/INS loosely and tightly integrated navigation system are summarized.The measured data are used to compare and analyze the PPP and RTK positioning modes,the accuracy of dynamic initial alignment used different speed measurement models,and two different integrated navigation methods.(2)The excellent performance of GNSS/INS integrated navigation filter is based on accurate function model and random model.But GNSS is vulnerable to environmental,low-cost inertial sensors are less stable and reliable than higher-level sensors.Considering the cost limitation in practical application,the analysis and processing of the sensors output data is a more economical and feasible scheme.Therefore,the main error sources of GNSS are analyzed.Aiming at the problem of GNSS observation cycle slip caused by hardware equipment or environmental impact,the MW-GF cycle slip detection method is utilized.Aiming at the problem of random noise and systematic error of INS sensor,the random noise can be determined by Allan variance analysis method,and the systematic error of IMU can be corrected by the rotation calibration of turntable.Multiple experimental data are used to verify the effectiveness of the calibration scheme.(3)Aiming at the problem of time-varying error parameters caused by the instability of sensors in GNSS/INS integrated navigation,an variational Bayesian adaptive Kalman filter is studied and simplified according to the characteristic of closed-loop feedback.The variational Bayesian adaptive Kalman filter can improve the accuracy of covariance matrix by posterior estimation.And the difference and mechanism between robust and adaptive filter and the filter are analyzed.Through simulation experiments,the accurate and inaccurate error model scenarios are simulated to verify that the variational Bayesian adaptive Kalman filter can adjust the covariance matrix in the integrated navigation system.And the field experiments under different observation environments are carried out to analyze the performance of these filters.The experimental results show that compared with the robust and adaptive filter,the position,velocity and attitude accuracy of the variational Bayesian adaptive Kalman filter in the open observation environment are improved by more than 12.4%,15.2% and 3.4%,respectively.While in the urban environment,the accuracy is slightly improved except for the velocity of up direction and pitch,and the improvement range is 1.1% to 18.1% compared with Kalman filter.(4)The traditional robust and adaptive filter is aimed at the influence of GNSS gross error and system model deviation,and its sensitivity of error detection depends on the accurate covariance matrix.It does not take into account the error caused by low-cost IMU instability and observation environment,which may cause the filter estimation results to diverge.Aiming at the phenomenon that low-cost MEMS-IMU is more susceptible to the influence of observation environment(temperature change,vibration,electromagnetic,etc.)due to the limitation of cost and technical process,this thesis proposes a motion model-assisted robust method.Different from the traditional robust methods based on innovation inconsistency,this method introduces the vehicle motion model.In the process of system update,INS mechanization and motion model are updated in two states.The state vector of motion model can be used to test and correct INS information or to fuse with GNSS.The effectiveness of the method is verified in navigation systems with different precision inertial sensors and integrated methods.The experimental results show that the horizontal accuracy of position and velocity is improved by more than 68.5% in the consumer MEMS-IMU loosely and tightly integrated navigation system,and the horizontal accuracy of velocity and attitude is improved by more than 20.3% in the smartphone MEMS-IMU loosely and tightly integrated navigation system.