电子学报 ›› 2016, Vol. 44 ›› Issue (4): 795-803.DOI: 10.3969/j.issn.0372-2112.2016.04.007

• 学术论文 • 上一篇    下一篇

高斯粒子流滤波器

张宏欣, 周穗华, 冯士民   

  1. 海军工程大学兵器工程系, 湖北武汉 430033
  • 收稿日期:2014-10-28 修回日期:2015-03-23 出版日期:2016-04-25 发布日期:2016-04-25
  • 作者简介:张宏欣 男,1987年12月出生,陕西汉中人.2010年毕业于西安理工大学,现为海军工程大学博士生,从事统计信号处理及目标跟踪相关研究. E-mail:mylifeforthebattle@hotmail.com;周穗华 男,1962年10月出生,广东五华人,1984年毕业于海军工程学院,1990年在海军工程学院获得博士学位.现为海军工程大学教授,从事军用目标特性信息处理及武器系统总体设计方面研究.

Gaussian Particle Flow Filter

ZHANG Hong-xin, ZHOU Sui-hua, FENG Shi-min   

  1. Department of Weapon Engineering, Naval University of Engineering, Wuhan, Hubei 430033, China
  • Received:2014-10-28 Revised:2015-03-23 Online:2016-04-25 Published:2016-04-25

摘要:

粒子流滤波器以粒子流速度场描述随机样本从先验分布到后验分布的演化,实现对系统状态的贝叶斯估计.针对其一般解计算复杂、难于滤波求解的问题,导出一种高斯假设条件下的粒子流滤波器.在线性高斯条件下推导了速度场的解析解;证明了当演化步长趋近于0时,该解析解与Kalman-Bucy滤波器的解具有一致的形式;基于该解导出了非线性高斯系统速度场的表达式,并进一步利用Unscented变换近似求解.通过若干仿真算例表明,高斯粒子流滤波器放宽了系统噪声为高斯型的限制,其精度优于经典非线性高斯滤波器,计算复杂度低于一般粒子滤波器,且具有良好的稳定性.

关键词: 非线性滤波, 贝叶斯估计, 粒子流滤波器, 速度场, Unscented变换

Abstract:

Particle flow filter formulated the dynamics from prior samples with posterior samples with particle flow velocity field to perform Bayesian estimation of system state.To address difficulties of particle velocity field computation in present particle flow filter,a novel particle flow filter based on Gaussian assumption was proposed.The analytical solution of velocity field under linear Gaussian condition was derived.The consistency of this analytical solution with Kalman-Bucy filter for continuous system,when discrete dynamic step goes to zero,was proved.The solution was finally extended to obtain the nonlinear Gaussian velocity field expression which can be approximated by using unscented transformation.Several simulations revealed the effectiveness over classic nonlinear Gaussian assumption on accuracy and particle filter on efficiency and stability.

Key words: nonlinear filtering, Bayesian estimation, particle flow filter, velocity field, Unscented transformation

中图分类号: