Research On Nonlinear State Estimation Algorithm And Its Application In Vehicle Integrated Navigation

As an important branch of national defense modernization,vehicle integrated navigation technology is playing a more and more important role in unmanned vehicles,intelligent robots and other fields.Nonlinear state estimation algorithm is the main factor affecting the performance of integrated navigation,so it is important to design advanced nonlinear state estimation algorithm to improve the performance of vehicle navigation system.As the application environment of vehicle integrated navigation system becomes more and more complicated,the nonlinear state estimation algorithm is affected by non-Gaussian noise,large computation burden,strong nonlinear system model and large state dimension,etc.,and faces challenges in estimation accuracy and real-time performance.Based on the vehicle integrated navigation system and the Cubature Kalman filter framework,this paper conducts a series of studies on the nonlinear state estimation algorithm,mainly including the following four parts:(1)To solve the problem of non-Gaussian noise,a Cubature Kalman filter based on the minimum Cauchy kernel loss function is firstly proposed to improve the filtering accuracy and stability of the algorithm in the case of large outliers in the measured noise.Then,a Cubature Kalman algorithm based on a mixture versiera function is proposed to deal with the complex non-Gaussian noise problem which exists both in the process and in the measurement.By combining the excellent robustness of the mixture versiera function with the flexibility of the hybrid kernel frame,the complex non-Gaussian noise can be suppressed effectively.Finally,numerical simulation and vehicle integrated navigation experiment results show that the two robust algorithms proposed in this paper are superior to existing algorithms in numerical stability and estimation accuracy.(2)To solve the problem of large computation of nonlinear state estimation algorithm,the Cubature Kalman filter based on approximate inverse matrix is proposed.Traditional nonlinear state estimation algorithms are always troubled by a large number of numerical computation,but there are few solutions to this problem in the current field of nonlinear state estimation.To solve the above problems,firstly,the Cubature Kalman filter based on Neumann series expansion method is proposed.By using second-order Neumann series to replace the traditional inverse method in Kalman gain calculation,the estimation accuracy is not affected and the time efficiency is improved.Then,a Cubature Kalman filter based on adaptive Gauss Seidel iterative method is proposed.The Kalman gain is calculated in the form of solving linear equations.A Gauss Seidel method is designed to solve the algorithm according to the accuracy requirements,so as to improve the time efficiency of the algorithm.Finally,numerical simulation and vehicle integrated navigation experiment results show that the proposed two low-complexity algorithms can effectively improve the time efficiency of the algorithm without sacrificing the estimation accuracy.(3)For the problem that a single algorithm cannot deal with the inaccurate noise variances and non-Gaussian noises,an adaptive robust nonlinear state estimation algorithm based on improved interactive multiple models is proposed.First,in view of the nonlinear systems with linear measurement function,based on the fram of time update in Cubature Kalman filter and the measurement update in Kalman filter,the modeling process and measurement noise variance for inverse Wishart distribution,the variable decibels bayesian method and estimate the system state,the process noise and measurement noise variance,realize the system noise covariance and measurement noise covariance of real-time estimation,approximate inverse matrix method are adopted to decrease the matrix inverse operation calculation burden,a low complexity adaptive nonlinear algorithm is proposed.At the same time,based on the robust nonlinear algorithm mentioned above,the same improved idea is used to design the low complexity robust nonlinear algorithm.Then,in the interactive multi-model framework,the above adaptive algorithm and the robust algorithm are used as sub-filters,and the weighted Kullback-Leibler is used to perform weighted fusion of the estimated results.Finally,the vehicle integrated navigation experiment results show that the proposed algorithm is robust to non-Gaussian noise and avoids the decrease of filtering accuracy caused by time-varying noise due to the advantages of adaptive algorithm.(4)A nonlinear state estimation algorithm based on spherical radial cubature criterion and data assimilation principle is proposed to reduce the estimation accuracy of nonlinear state estimation algorithm in high-dimensional nonlinear systems.Firstly,an integrated navigation model with odometer calibration coefficient and installation error as the amplifying state quantity is derived to realize the real-time estimation of odometercalibration coefficient and installation error.The integrated navigation model is treated as a strongly nonlinear high dimensional model from the perspective of state estimation algorithm.To improve the estimation accuracy of nonlinear state estimation algorithm in high dimensional strong nonlinear system,this thesis propose to use the spherical radial cubature criterion in the framework of data assimilation,and propose a new nonliear algorithm based on bayesian estimation theory.Finally,numerical simulation and vehicle integrated navigation experiment verify the effectiveness of the proposed algorithm.