Adaptive unscented Kalman particle filtering method

Abstract

The invention discloses an adaptive unscented Kalman particle filtering method. The method utilizes the theory of the Sage filtering windowing method, also combines the idea of fading, estimates a true covariance matrix of the observed quantity by collecting an epoch innovation vector, and compares the true covariance matrix with the covariance matrix of a filtering recursive model, when a deviation exists between the two covariances, the observed covariance matrix of the system is adaptively adjusted according to the difference. Based on the process, an adaptive fading factor is designed, theobservation noise is further modified, the modified observation noise participates in the solution of a gain matrix, and thus the state estimation can be adaptively adjusted. The scheme of the invention can effectively perform filtering correction on a strongly nonlinear satellite / inertial integrated navigation system, especially when the external noise is abnormal, the filtering gain can be effectively and adaptively adjusted to improve the robustness and positioning accuracy of the system.

Description

technical field [0001] The invention belongs to the field of nonlinear combined navigation, in particular to an adaptive unscented Kalman particle filter method. Background technique [0002] Conventional Kalman filtering is a model-based linear minimum variance estimation, but in actual engineering practice, the system is often nonlinear. Although the Extended Kalman Filter (EKF, Extended Kalman Filter) can solve the problem of nonlinear system state estimation, it performs Taylor expansion on the system equation, ignores the influence of high-order terms on the system, and has a certain truncation error. The Unscented Kalman Filter UKF (Unscented Kalman Filter) determines the optimal gain matrix based on the covariance of the estimated quantity and quantity measurement, and the covariance matrix is ​​calculated based on the recurring double σ sample point, and the sample point is calculated according to the system The nonlinear model calculation. So it can be used for bo...

AI Technical Summary

Problems solved by technology

[0005] Most filtering methods are poor for state estimation of strongly nonlinear systems, and when the external noise changes, the system often cannot perceive it, resulting in poor filtering results, and even leading to filtering divergence in severe cases.

Images