Research On Radar Target Tracking Method Under Model Uncertainty

The purpose of radar target tracking is to track space targets through radar,and then realize accurate estimation of the target state,which is the premise of the modern military tracking system for data fusion,command decision-making,target recognition and other tasks.However,in the battlefield environment,due to the influence of measurement noise uncertainty caused by electromagnetic interference and thermal noise interference and target model parameter uncertainty caused by target maneuvering,which makes it difficult for the existing adaptive Kalman filter algorithm to solve the problem of maneuvering target tracking under model uncertainty.Therefore,to carry out research on the problem of radar target tracking under the condition of model uncertainty,the main work is as follows:(1)Aiming at the target tracking problem under the condition of heavy tail skew noise caused by electromagnetic interference,equipment failure,etc.,a variational Bayes filtering algorithm based on Skew-T is proposed.The algorithm uses the Skew-T distribution to model the heavy-tail skewed noise,establishes the joint posterior probability density function of Gaussian distribution,Half normal distribution and Gamma distribution,and uses variational inference to estimate the system state and noise parameters,and the simulation results show that the proposed algorithm has good filtering performance.For the target tracking problem under nonlinear system,a robust volumetric Kalman filter algorithm based on variational inference is proposed,which replaces the probability density function of the Skew-T distribution with an approximate form,uses the spherical radial rule to solve the function numerically integrally,and then approximates the system state and covariance by the volume point obtained by sampling,and combines variational inference learning to estimate the approximate posterior distribution,and updates the system state to obtain an accurate target state.(2)Aiming at the problem of estimating the target state under the condition of uncertain target model parameters and noise,a general interactive multiple model adaptive filter algorithm is proposed.Combined with the Skew-T distribution measurement noise statistical model,the algorithm uses the inverse Wishart distribution to describe the covariance of the system prediction error,constructs the system joint probability density function under each model.Besides,the algorithm uses variational inference to calculate the approximate posterior distribution of the system state and noise parameters of each model,then the fusion of multiple model states by weights,so as to achieve the accurate estimation of the system state in the case of uncertainty of noise information and target model parameters.Then the simulation results show that the proposed algorithm has well robustness and high estimation accuracy.(3)Aiming at the problem that the multi-model adaptive filtering algorithm does not consider the influence of constraints in practical applications,which leads to the degradation of system tracking performance,an interactive multiple model adaptive filter algorithm based on probability density truncation is proposed.Firstly,the joint posterior probability density function of the system state,prediction error covariance and noise parameters is established,and then variational inference is used to solve the problems of time-varying noise and mutual coupling between noise information and state.After obtaining the updated filtering results,construct the linear inequality constraint equation,and truncate according to the constraint boundary,and calculate the mean value and the variance of the truncated probability density function.Finally,by updating the weights,the system state and covariance after each model constraint are fused to complete the estimation of the target state.The simulation results show that the proposed algorithm improves the state estimation accuracy of the system.