| MEMS(Micro Electro Mechanical Systems)inertial sensors have the advantages of low cost and high reliability,and their accuracy has been gradually improved in recent years,which has shown great application prospects in civilian car navigation fields such as unmanned vehicles,networked vehicles,and smart high-speed train.However,in complex situations such as highly dynamic maneuvering turns of vehicles,outliers in satellite positioning observations,and rejection of satellite signals,vehicle navigation systems based on low-cost MEMS inertial devices often have large positioning errors,which are difficult to meet the requirements of seamless navigation.Therefore,the paper focuses on several typical problems in the application of high-precision seamless navigation in vehicles.The main work and conclusions are summarized as follows:In view of the low efficiency of calibration and temperature compensation of MEMS Inertial Measurement Unit(MIMU)in traditional calibration methods,and MIMU cannot complete heading alignment autonomously,a heading alignment method using accelerometer data and satellite navigation data and methods based on MIMU simplified error model calibration and temperature compensation are proposed respectively.First,combined with the error model of the inertial device,the linear error model and the second-order nonlinear error model of MIMU are established respectively,and the corresponding simplified calibration scheme is designed to complete the calibration,and the effect of the calibration is verified by static experiments.Then,in order to further improve the accuracy of MIMU,establish the second-order and third-order temperature compensation model for MIMU,design the calibration scheme and complete the experiment in the high-precision temperature control turntable.Finally,the attitude initialization using accelerometer and GNSS navigation information is studied,and the attitude alignment experiment is completed in a vehicle-mounted environment based on the high-precision optical fiber main inertial navigation system,which verifies the feasibility of the initialization method.The designed simplified calibration and temperature compensation method further improves the efficiency while meeting the accuracy;the new heading alignment method not only obtains the initial value of the MIMU heading installation angle on the car body,but also improves the alignment The result is the ability to resist carrier maneuver interference.Aiming at the problem that the low-cost integrated navigation positioning accuracy decreases due to the vehicle’s high dynamic turning maneuvers and the outliers in GNSS observations,a strong tracking maximum correlation entropy filtering algorithm based on the nested sparse grid integration method is proposed.First,the influence of model error and observation error on the estimation result is analyzed,and the error propagation model is deduced.Secondly,based on the traditional extended Kalman filter,an extended strong tracking maximum correlation entropy filter method is designed by forcing the residual sequence to be orthogonal and deriving the filter gain with the maximum cross entropy criterion.In order to further improve the filtering accuracy and reduce the computational complexity,under the assumption of approximate Gaussian distribution,the paper combines the strong tracking maximum correlation entropy filter with the nested sparse grid integration method,and proposes a nested sparse strong tracking maximum correlation entropy filter.Based on actual car navigation experiments,the paper compares EKF,Extended Strong Tracking Filter(ESTF),Maximum Correntropy Kalman Filter(MCKF),and strong tracking maximum correlation based on the nested integration rule(NIR-STMCKF)four algorithms have navigation accuracy in the presence of model deviations and observations with outliers.The results show that the accuracy and reliability of the NIR-STMCKF algorithm in both cases are significantly improved compared to the other three algorithms,which verifies the effectiveness of the method.Aiming at the problem of the large positioning error of the smoothly running train in the tunnel that only relies on the inertial navigation system composed of MEMS,a method using Motion Constraint(MC)and Odometer(ODO)to suppress MIMU error divergence is proposed.This method makes full use of the train motion constraint information and the odometer installed on the wheel axle.In order to better exert the potential of motion constraint,the installation angle of MIMU relative to the car body is considered in the motion constraint model.When both GNSS and motion constraint conditions are met,the installation angle of the system is estimated online.In tunnels,train motion constraints provide lateral and longitudinal speed observations,and odometers provide forward speed observations.Experiments conducted in a real tunnel environment and artificially disconnected GNSS signals show that the accuracy of the MIMU/ODO/MC integration method is significantly higher than that of the pure inertial navigation system(P-INS),and the traditional installation angle is not considered Method of suppressing inertial navigation error(TMC-INS)error and considering the installation angle of MIMU motion constraints to suppress inertial navigation(CIAMC-INS)error method,which can realize seamless navigation inside and outside the tunnel.Aiming at the problem of the inapplicability of motion constraint assumptions during turning and acceleration/deceleration maneuvers during driving,a vehicle motion prediction assisted MIMU dead reckoning method based on CNN(convolutional neural network)is proposed.The paper analyzes the influence of heading turn,pitch turn and acceleration and deceleration maneuvers on motion constraints,and obtains the factors that affect the null hypothesis of lateral velocity and vertical velocity.On this basis,a dynamic adaptive estimator based on CNN is constructed to predict the lateral velocity and vertical velocity of the car body that are difficult to accurately model,and use them as observations to expand the position and velocity in the Kalman filter.Estimation of the deviation from the sensor.This method is trained on the KITTI data set,and the real-time output speed of the trained adaptive estimator is combined with MIMU to obtain the three-dimensional position of the vehicle.The entire algorithm only needs MIMU data and no other sensors,fully tapping the potential of motion constraints,and improving the accuracy and reliability of positioning without adding any additional sensors. |
PDF Full Text Request