加权逼近粒子滤波算法及其应用

针对常规粒子滤波存在的粒子退化、粒子枯竭和运算量大等问题,提出了一种新颖的加权逼近粒子滤波算法(weight approaching particle filter,WAPF).在重采样前按粒子–权值大小将粒子集分组,用两组粒子的加权值覆盖权值较小的粒子,这样便可以使部分粒子从低似然区向高似然域逼近.借用Kullback信息描述加权逼近产生的粒子分布与似然分布的差别,通过迭代发现Kullback信息是递减的,这说明加权逼近粒子滤波算法是合理的.混沌摄动重采样算法,用类似载波的方法将具有全局遍历性的混沌变量引入,更增加了支持粒子集的多样性,且具有较少的运算量.仿真结果显示了所提算法的有效性.

Weight approaching particle filter and its application

To solve the problem of particle degeneracy, sample impoverishment and expensive computing cost in conventional particle filter, we propose the weight approaching particle filter(WAPF) to increase the particle diversity before resampling step. In the resampling step, particles are classified into two groups according to their particle-weights, and then the particles with the smaller weights are replaced by the mean of the two group particles, so that the particles can approach from the low likelihood region to the high likelihood region. The difference between the particle distribution produced by the behavior of mean approaching and the likelihood distribution is described by Kullback information. Kullback information decreases with increasing iteration degree, which proves that the new algorithm is rational. Similar to the carrier wave method, the chaotic perturbation resampling method adopts the chaotic variable with the property of global ergodicity to ameliorate the diversity of samples and reduce the computation load. Simulation results demonstrate the feasibility of the improved particle filter.