) 上海交通大学 自动化系; 系统控制与信息处理教育部重点实验室;海洋智能装备与系统集成技术教育部实验室, 上海 200240
收稿日期:2020-01-06出版日期:2020-09-28发布日期:2020-10-10通讯作者:蔡云泽E-mail:yzcai@sjtu.edu.cn作者简介:董祥祥(1995-),女,山西省临汾市人,博士生,主要研究方向为状态估计与滤波基金资助:国家自然科学基金国家重大科研仪器研制项目(61627810);国家科技重大专项(2018YFB1305003)A Variational Bayes-Based Filter with Uncertain Heavy-Tailed Noise
DONG Xiangxiang, LÜ Runyan, CAI Yunze*() Department of Automation; Key Laboratory of System Control and Information Processing of the Ministry of Education; c. Key Laboratory of Marine Intelligent Equipment and System of the Ministry of Education, Shanghai Jiao Tong University, Shanghai 200240, China
Received:2020-01-06Online:2020-09-28Published:2020-10-10Contact:CAI Yunze* E-mail:yzcai@sjtu.edu.cn摘要/Abstract
摘要： 为了解决非线性滤波中量测噪声呈厚尾分布且统计特性不确定的问题,提出一种基于Pearson Type VII 分布的自适应滤波算法.针对传统鲁棒卡尔曼滤波器因尺度矩阵和自由度参数固定不变而无法自适应调整的问题,以容积卡尔曼滤波器为基础,选择Pearson Type VII 分布对厚尾噪声进行建模,将传统鲁棒滤波固定自由度参数的估计转化为Pearson Type VII 分布中可自适应调整的双自由度参数的估计,并通过 inverse Wishart和Gamma分布描述尺度矩阵、双自由度参数和辅助参数的先验分布,利用遗忘因子对各参数进行时间更新;基于变分贝叶斯理论,对系统状态、尺度矩阵、双自由度参数和辅助参数形成的联合后验概率密度函数进行变分迭代,实现对系统状态和未知厚尾噪声的联合估计.仿真结果表明,在不确定厚尾噪声条件下,本文算法的滤波精度高于传统鲁棒容积卡尔曼滤波.
关键词: 厚尾噪声, Pearson Type VII 分布, 变分贝叶斯, 联合估计, 容积卡尔曼滤波
Abstract: An adaptive Pearson Type VII based filter was proposed to deal with measurement noise with heavy tailed distribution and uncertain statistical characteristics in nonlinear filtering. In view of the fact that the scale matrix and degree-of-freedom (DOF) parameters of traditional robust Kalman filter are fixed, which cannot adjust adaptively. The Pearson Type VII distribution was selected to model the heavy tailed noise to realize the adaptive estimation of 2-DOF parameters based on Cubature Kalman filter (CKF). Inverse Wishart and Gamma distributions were used to describe prior distributions of the scale matrix, 2-DOF parameters, and auxiliary parameters. Besides, time updating of these parameters was performed through a forgetting factor. Based on the variational Bayesian theory, the variational iteration of joint posterior probability density function formed by states together with the 2-DOF parameters and auxiliary parameters was performed, and the states and noise parameters were estimated simultaneously. The simulation results show that this algorithm has a higher filtering accuracy than the traditional robust CKF under the condition of uncertain heavy tailed noise.
Key words: heavy tailed noise, Pearson Type VII distribution, variational Bayesian (VB), joint estimation, Cubature Kalman filter (CKF)
PDF全文下载地址:
点我下载PDF
