Robust autocovariance least-squares noise covariance estimation algorithm
Robust autocovariance least-squares noise covariance estimation algorithm
复制标题
鲁棒自协方差最小二乘噪声协方差估计算法
DOI：
10.1016/j.measurement.2021.110331
复制
发表时间：
2021-10
期刊：
影响因子：
5.6
通讯作者：
Chen Changxin
中科院分区：
文献类型：
--
作者：
Li Wei;Lin Xu;Li Shaoda;Ye Jiang;Yao Chaolong;Chen Changxin
Kalman filter (KF) is the main algorithm in the field of optimal estimation. KF can realize the optimal estimation only under the premise that both the function model and the random model are known. In most practical applications, random models are often unknown, and unknown or biased random models will make filtering suboptimal. Besides, the gross errors will cause the innovation to appear abnormal, making the innovation statistics abnormal. The abnormal innovation statistics will lead to the deviation of the estimated noise covariance information. Therefore, when there are gross errors, how to ensure the random model's correct estimation is a major problem faced by the current random model estimation method. To solve this problem, we propose a robust autocovariance least-squares noise covariance estimation algorithm (RALS). In the new method, we first introduce the robust Kalman filtering method (RKF) based on the chi-square test to resist the influence of abnormal innovation on subsequent epoch state estimation; Then, based on the basic idea of the correlation robustness method, we constructed the correlation robust innovation function sequence model, taking into account the correlation between the innovation sequences, to repair the abnormal innovation statistics, so as to obtain accurate post-test innovation statistics; Subsequently, we use the autocovariance least square method to eliminate the coupling effect between the two types of noise covariance information, to correctly estimate the two types of noise covariance information; Finally, we adopt an iterative strategy to eliminate the influence of a priori random model deviation, to realize the separation of the gross error coupled in the abnormal innovation and the prior random model deviation. Two experimental results show that: compared with the KF, RKF, and ALS methods, the new method has higher noise covariance estimation accuracy and filtering accuracy.
登录
查看更多内容
影响因子：
2.6
作者：
J. Duník;O. Kost;O. Straka;E. Blasch
通讯作者：
J. Duník;O. Kost;O. Straka;E. Blasch
影响因子：
4.4
作者：
Y. Yang;M. Cheng;C. Shum;B. Tapley
通讯作者：
Y. Yang;M. Cheng;C. Shum;B. Tapley
基于马氏距离作为异常值判断标准的鲁棒卡尔曼滤波
影响因子：
4.4
作者：
Chang, Guobin
通讯作者：
Chang, Guobin
影响因子：
4.4
作者：
Yang, Yuanxi;Gao, Weiguang
通讯作者：
Gao, Weiguang
影响因子：
4.4
作者：
Yang, Y
通讯作者：
Yang, Y