Particle filter algorithm based on geometric center and likelihood estimation

作     者：Fang, Xing Luo, Yin Wang, Lei Jiang, ShuiBin Xu, Nan Zeng, Daniel Dajun 

作者机构：Tianjin Univ Sch Management & Econ Beijing Peoples R China Wenge Technol Co Ltd Beijing Peoples R China Chinese Acad Sci Inst Automat State Key Lab Management & Control Complex Syst Beijing Peoples R China Univ Chinese Acad Sci Beijing Peoples R China Univ Arizona Dept Management Informat Syst Tucson AZ 85001 USA 

出 版 物：《AIP ADVANCES》 (AIP进展)

年 卷 期：2021年第11卷第8期

页      面：1-8页

核心收录：

学科分类：07[理学] 0805[工学-材料科学与工程（可授工学、理学学位）] 0702[理学-物理学] 

基　　金：National Key R&D Program of China [2016QY02D0305] Key Research Program of the Chinese Academy of Sciences [ZDRW-XH-2017-3] National Natural Science Foundation of China [71621002, 61671450] National Key Research and Development Program of China [2020AAA0103405] Strategic Priority Research Program of Chinese Academy of Sciences [XDA27030100] 

主　　题：Monte Carlo methods 

摘      要：The improved particle filter (PF) based on the geometric center and likelihood estimation is proposed in order to solve the problem of particle dilution and degradation. In the resampling stage, the geometric center is used to resample the particles. The particles are filtered according to the distance between the particles and the geometric center, and then the particles are resampled. The resampled particles are composed of newborn particles and non-resampled particles. The former can help alleviate the degradation problem, while the latter can keep the diversity of the particle set. In order to ensure effectiveness of the PF, the positioning error threshold of the particle filter is introduced. In the phase of particle weighting calculation, in view of the problem of low accuracy and divergence of PF state estimation caused by non-stationary and non-Gaussian noise, it adopts non-Gaussian noise parameter estimation based on likelihood to approximately estimate the measurement noise instead of the Gaussian density function. The proposed model is applied to particle weight calculation to avoid particle degradation caused by Gaussian density function approximation. The simulation results show that, after the improved algorithm, the root-mean-square error is reduced to 0.085, the variance is reduced to 0.014, and the running time is shortened by 14.8% compared with the polynomial resampling algorithm, which can effectively alleviate particle degradation and dilution in the traditional PF algorithm, and the positioning accuracy is also improved. (C) 2021 Author(s).