摘要
针对基于高斯滤波的重要性采样方法运算量的明显增加主要集中在使用高斯滤波生成更好的重要性密度函数的问题,提出了一种新的高斯衍生粒子滤波算法(GDPF).该算法将一种类似光子衍射的粒子衍生重要性采样方法与现有的高斯辅助粒子滤波算法(GAPF)相结合,通过粒子的扩张与收缩,在保证不减少参与状态估计的粒子数的条件下减少更新粒子数,根据粒子权值大小自适应地调整衍生粒子数,能很好地缓解精度与运算量之间的矛盾,抑制粒子退化等问题.对衍生粒子进行理论分析,证明了其与高斯采样粒子的等效性.仿真结果表明,当选取了相同的参与状态估计的粒子数时,所提算法保持了与原算法相当的估计精度,同时运算量大大降低.
The particle filter combined with Gaussian filter can restrict particle degeneracy with a secondary result that the new particle filter has a high calculation cost.In order to reduce the expensive calculation cost in the Gaussian aided particle filter（GAPF）,a Gaussian diffracted particle filter（GDPF） is proposed by introducing a light-diffracting-like particle diffracting sampling method into the current GAPF.The proposed method predicts fewer particles,and keeps more particles to be re-sampled from each Gaussian importance density function,so that the overall particles in the estimation of the system state are maintained the same by extending and contracting of particles.The number of particles is also adjusted according to particle weights.Therefore the calculation cost in GDPF is significantly reduced when the accuracy is required the same as the GAPF method,and the sample degeneracy problem is successfully improved.Theoretical analysis indicates that the efficiencies of both GDPF and GAPF are the same.The results of Monte Carlo simulations with the same number of particles in state estimation show that the improved particle filter can preserve the same accuracy of estimation while computation burden is greatly reduced.
出处
《西安交通大学学报》
EI
CAS
CSCD
北大核心
2010年第6期72-77,共6页
Journal of Xi'an Jiaotong University
基金
国防预研项目(51309060302)
关键词
粒子滤波
高斯滤波
粒子退化
粒子衍生
particle filter
Gaussian filter
particle degeneracy
particle diffracting
