基于预测和量测的多建议分布粒子滤波  

Prediction-and-measurement-based Multiple Proposal Distributions

作　　者：龚佑斌 刘金澎 GONG Youbin;LIU Jinpeng(No.29 Institute of CETC,Chengdu Sichuan 610036,China)

机构地区：[1]中国电子科技集团公司第29研究所,四川成都610036

出　　处：《通信技术》2022年第12期1555-1559,共5页Communications Technology

摘　　要：粒子滤波可以表征任意分布,因此具有广泛的适用性,但其会面临粒子退化的问题。重采样步骤可以有效地解决粒子退化,但会引入粒子贫化的新问题。为了有效地解决粒子退化和贫化问题,基于多建议分布的思路,提出了一种高估计精度和计算有效的基于预测和量测的多建议分布粒子滤波。其思路为,首先分别将预测和量测作为基本建议分布,并分别从两个基本建议分布中选取粒子;其次用粒子的似然决定每个基本建议分布的权重,并基于最新的量测值修正每个粒子点的权值;最后通过粒子点加权求和的方式获得对于状态后验估计的一阶矩和二阶矩。仿真结果证明,相比于经典算法,所提算法可以以较低计算代价获得最好的滤波估计结果。Particle filter can characterize arbitrary distributions and therefore has wide applicability, but it faces the problem of particle degradation. The resampling step can effectively solve this problem, but introduces a new problem of particle depletion. In order to effectively solve the particle degradation and depletion problems, based on the idea of multiple proposal distributions, this paper proposes a predictionand-measurement-based multiple proposal distributions particle filter with high estimation accuracy and computational efficiency. The idea is to first use prediction and measurement as the basic proposal distributions, respectively, and to select particles from each of the two basic proposal distributions. Then, the likelihood of particles is used to determine the weight of each basic proposal distribution, and the weight of each particle is modified based on the latest measurements. Finally, the first and second order moments of the posterior estimates are obtained by means of the weighted sum of particles. Simulation results demonstrate that the proposed algorithm can achieve the best filter estimation results at a lower computational cost compared to the classical algorithm.

关 键 词：粒子滤波 建议分布 后验估计 多模型采样 

分 类 号：TN911[电子电信—通信与信息系统]