摘要
滤波的目的是从序贯量测中在线、实时地估计和预测出动态系统的状态和误差的统计量。从递归贝叶斯估计的框架出发,对非线性滤波算法作了统一描述,并根据对后验概率密度的近似方法的不同,把非线性滤波划归为3类:基于函数近似的滤波方法、基于确定性采样的滤波方法和基于随机采样的滤波方法。对这些非线性滤波的原理、方法及特点做了分析和评述,最后介绍了非线性滤波研究的新动态,并对其发展作了展望。
The main goal of filtering is to obtain, reeursively in time, optimal estimation and prediction of the dynamical systems and error statistics from the sequential observations. Various nonlinear filtering algorithms are reviewed and interpreted in a unified way using the recursive Bayesian estimation. According to different approximation methods, these approximate nonlinear filters can be categorized into three types:analytical approximations, deterministic - sampling based approaches, and stochastic - sampling based filters. Then the principles, methods and characteristics of above nonlinear filters are analyzed and reviewed in detail. Finally, some representative new developments of the nonlinear filtering are described,and further research prospects are introduced.
出处
《电光与控制》
北大核心
2008年第8期64-71,共8页
Electronics Optics & Control
关键词
非线性滤波
递归贝叶斯估计
差分滤波
无味卡尔曼滤波
粒子滤波
nonlinear filtering
recursive Bayesian estimation
differential filter
unscented Kalman filter
particle filter