一种基于SVM重采样的似然粒子滤波算法  

Likelihood particle filter based on support vector machines resampling

作　　者：蒋蔚[1] 伊国兴[1] 曾庆双[1] 

机构地区：[1]哈尔滨工业大学航天学院,哈尔滨150001

出　　处：《控制与决策》2011年第2期243-247,252,共6页Control and Decision

基　　金："十一五"国防预研项目(51309030102;51309030203)

摘　　要：针对弱观测噪声条件下非线性、非高斯动态系统的滤波问题,提出一种基于支持向量机的似然粒子滤波算法.首先,采用似然函数作为提议分布,融入最新的观测信息,比采用先验转移密度的一般粒子滤波算法更接近状态的真实后验密度;然后,利用当前粒子及其权值,使用支持向量机估计出状态的后验概率密度模型;最后,根据此模型重采样更新粒子集,有效地克服粒子退化现象并提高状态估计精度.仿真结果表明了所提出算法的可行性和有效性.To cope with state estimation problems of nonlinear/non-Gaussian dynamic systems with weak measurement noise,an improved likelihood particle filter（LPF） algorithm is proposed based on support vector machines（SVM） resampling.Firstly,the algorithm employs the likelihood as proposal distribution and takes account of the most recent observation,so it is comparably closer to the posterior than the transition prior used as proposal.Then,the posterior probability density model of the states is estimated by SVM with current particles and their importance weights during iteration.Finally,after resampling the new particles from the given density model,degeneration problem is solved effectively by these diversiform particles.The simulation results show the feasibility and effectiveness of the algorithm.

关 键 词：粒子滤波 似然函数 支持向量机 重采样 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]