Local estimation adaptive two-stage square root cubature Kalman filter method

Abstract

The invention discloses a local estimation adaptive two-stage square root cubature Kalman filter method. The method comprises the steps of firstly deriving a square root two-stage cubature Kalman filter algorithm for adopting a square root of an error covariance matrix to replace a covariance matrix to participate in recursive operation; secondly proposing an adaptive two-stage square root cubature Kalman filter algorithm; and thirdly estimating noises with unknown statistical characteristics in stages in filters in the two stages, improving estimation effects of the respective filters, and correcting noise statistical characteristics of the respective filters again by using obtained new system states and error covariances to form respective recursive loops of the two filters. In the localestimation adaptive two-stage square root cubature Kalman filter algorithm (ATSCKF-G), the estimation of measurement noise statistical characteristics is directly performed by adopting the Sage-Husafilters, and then estimated statistical characteristic values serve as known conditions.

Description

technical field [0001] The invention relates to the field of filter estimation, in particular to a local estimation adaptive two-stage square root volume filter method. Background technique [0002] The nonlinear filtering algorithm is a process of using discrete sensor observations to estimate the continuous state of the target and filtering random noise under the nonlinear system model. At present, several common nonlinear Kalman filters have their own advantages and disadvantages. Extended Kalman filtering (EKF) performs Taylor expansion on nonlinear functions and ignores high-dimensional items for linearization. However, this method is only suitable for weak nonlinear functions whose system models are smooth enough. If the system is a strongly nonlinear system, it will be affected by filtering The error is large and the effectiveness is lost. At the same time, the Jacobian matrix needs to be calculated during the calculation, and the calculation amount is large. The un...

AI Technical Summary

Problems solved by technology

In the filtering iteration process, the existence of some factors will cause the error covariance matrix to be negatively definite, such as rounding errors in numerical calculations, large initial value errors, and large observation noises, etc., which will cause the filter to be unstable or even unable to work.

Images