中文摘要：

文章通过多层采样方式，将样本空间划分为多个部分，集中采样点到使概率密度函数值大的地方，大大减小了采样误差；在重采样阶段嵌入KHM聚类算法，通过将空间特征与权重分布近似的粒子进行聚类，降低总的样本数，提高了计算效率。样本经聚类处理后，在保持粒子状态后验分布的几何特征的同时，状态空间中的粒子数明显降低，计算效率显著提高。

英文摘要：

In this paper, by using the stratified sampling method, the sample space is divided into mul- tiple parts, grouping sampling points to the part of high probability density function value, so the sampling error is greatly reduced. In the resampling phase, the KHM clustering algorithm is embed- ded, the particles with approximate spatial characteristics and weight distribution are clustered, thus reducing the total number of samples and improving the computing efficiency. After the clustering, the geometrical characteristics of particles state posterior distribution is maintained, while the number of particles in the state space is significantly reduced and the computing efficiency is improved.