Research On Filtering Algorithm Of Vehicle Integrated Navigation Under Uncertain Noise

Navigation and positioning technology is the core of automatic driving of driverless vehicles.Global Positioning System and Inertial Navigation System integrated navigation,as a common positioning and navigation system,can meet the actual demand of driverless vehicles in different working conditions.However,the positioning accuracy of the GPS/INS integrated navigation system decreases due to the non-negligible error of the external measurement sensor system.In order to improve the estimation accuracy of the navigation system,this paper designs three filtering algorithm for the navigation system state estimation problem with measurement noise uncertainty.The specific research contents are as follows:First,considering the gap of filtering accuracy degradation,caused by the absence of measurement noise variance in GPS/INS navigation system and the inconsistency of the statistical characteristics of the noise prior and real noise,a cubature Kalman filter based on variational Bayesian inference is proposed.The algorithm assumes that the measurement noise obeys Gaussian distribution,which the mean of measurement noise is zero,and the variance is the inverse Gamma distribution with unknown parameters.The algorithm takes the system state and noise parameters together as the variables to be estimated,and the joint posterior distribution of the state and noise variance is obtained by a factorized free form distribution.The estimate of the state and noise parameters by separately utilizing cubature Kalman filter and the Variational Bayesian theory is to obtain the optimal posterior distribution.Second,the position information of the INS is susceptible to interference from unknown external environments during the measurement process,and the measurement noise characteristics are expressed as the bounded uncertainty.Aiming at the filtering problem of the GPS/INS navigation system with bounded noise,this paper adopts the set membership filtering algorithm to estimate the state of the system.The algorithm is based on the outer bounded ellipsoid set containing the real state of the system,and the Minkowski sum is used as the standard,and then the feasible solution of the system state is obtained by combining with the system model.Third,due to the error caused by carrier maneuver and drift of components as well as the influence of uncertain factors,it is difficult to use random or bounded noise to describe in the integrated navigation system.Aiming at the filtering problem under the coexistence of random noise and bounded error in GPS/INS navigation system,this paper proposes a variational Bayesian Expectation-maximization filtering algorithm.The double uncertain noise is modeled by Gaussian mixture model.We construct a joint probability density function that is composed of system states and noise model parameters.The system state is estimated by the variational Bayesian expectation method,and the Gaussian mixture parameters are updated by the variational Bayesian maximization method.In order to verify the effectiveness of the algorithm proposed in this paper,we have done comparative experiments on integrated navigation system.The result shows that the proposed method has an improved estimation performance compared with other conventional approaches.