摘要
数据融合作为多源协同导航方案中的一环对状态估计质量影响重大。粒子滤波由于其在非线性非高斯系统中具备的独特理论优势已逐渐成为众多融合方法焦点。但粒子退化及其导致的样本枯竭却制约粒子滤波在复杂工程中的应用。提出采用鲁棒容积裂变粒子滤波解决上述问题。首先在容积法则框架下利用Huber函数将L2范数与L1范数结合来改进重要性密度函数,抑制观测噪声并通过高斯分布融合拉普拉斯分布进一步优化建议分布,以此缓解粒子退化;在重采样前对粒子群进行裂变衍生,通过对高权值粒子进行裂变并覆盖低权值粒子重构粒子权值实现对样本枯竭的抑制。多源协同导航车载实验表明,在相同条件下,相对于扩展卡尔曼滤波、容积粒子滤波、强跟踪粒子滤波,提出的算法在精确度上分别提高了23.04%、42.62%、37.74%,为缓解粒子退化和多源协同定位提供了新的思路。
As a part of the multi-source cooperative navigation scheme,data fusion has significant impact on the quality of state estimation.Because of its unique theoretical advantages in the nonlinear non-Gaussian system,the particle filter has gradually become the focus of many fusion methods.However,particle degradation and sample depletion restrict the application of particle filter in complex engineering.In this article,a robust cubature fission particle filter is proposed to solve the above problems.Firstly,in the framework of cubature rule,the Huber function is used to combine L2 norm with L1 norm to improve the importance density function,suppress the observation noise,and further optimize the proposed distribution by integrating Gaussian distribution with Laplace distribution.In this way,the particle degradation is alleviated.The particle swarm is fission derived before resampling,and the sample depletion is suppressed by fission of high weight particles and covering low weight particles to reconstruct particle weights.The vehicle experiment of multi-source cooperative navigation shows that under the same conditions,compared with extended Kalman filter,cubature particle filter and strong tracking particle filter,the root mean squared of the proposed algorithm is improved by 23.04%,42.62%and 37.74%,respectively.It provides a new idea for alleviating particle degradation and multi-source cooperative localization.
作者
孙伟
刘经洲
Sun Wei;Liu Jingzhou(School of Geomatics,Liaoning Technical University,Fuxin 123000,China)
出处
《仪器仪表学报》
EI
CAS
CSCD
北大核心
2022年第2期166-175,共10页
Chinese Journal of Scientific Instrument
基金
2019辽宁省“兴辽英才计划”青年拔尖人才(XLYC1907064)
2018年度辽宁省“百千万人才工程”人选科技活动资助项目(辽百千万立项【2019】45号)
辽宁工程技术大学学科创新团队(LNTU20TD-06)项目资助
关键词
协同导航
粒子滤波
粒子退化
鲁棒性
容积
cooperative navigation
particle filter
particle degradation
robustness
cubature
