基于局部变分贝叶斯推断的分布式交互式多模型估计  

Distributed interactive multi-model estimatation based on partial variational Bayesian inference

作　　者：胡振涛[1] 杨诗博 侯巍 HU Zhen-tao;YANG Shi-bo;HOU Wei(School of Artificial Intelligence,Henan University,Zhengzhou Henan 450000,China;School of Artificial Intelligence,Nankai University,Tianjin 300381,China)

机构地区：[1]河南大学人工智能学院,河南郑州450000 [2]南开大学人工智能学院,天津300381

出　　处：《控制理论与应用》2024年第4期681-690,共10页Control Theory & Applications

基　　金：国家自然科学基金项目(61976080);河南省科学技术厅科学技术重点项目(212102310298);河南大学研究生教育综合改革试点建设项目(SYL20010101);河南省学位与研究生教改项目(2021SJGLX195Y)资助.

摘　　要：针对目前部分多模型算法预先设定运动模型转移概率矩阵对状态估计精度的不利影响,本文提出了一种基于局部变分贝叶斯推断的分布式交互式多模型估计算法.不同于传统交互式多模型估计中运动模型转移概率矩阵为先验已知的假设条件,在分布融合估计框架下,首先基于最小化Kullback-Leibler散度准则的递归优化策略实现对运动模型转移概率矩阵的预测与更新;在此基础上,结合变分贝叶斯推断实现对当前时刻目标状态与模型概率的联合估计;最后依据协方差交叉融合策略完成对局部状态估计融合.仿真结果表明:新算法通过对运动模型转移概率矩阵以及模型概率自适应在线估计,有效提升了机动目标的状态估计精度.In view of the adverse effect of presetting transition probability matrix of motion model on state estimation accuracy in some multi-model algorithms,a new distributed interactive multiple model estimation algorithm based on parital variational Bayesian inference is proposed in this paper.Different from the assumption that the motion model transfer probability matrix is a priori known in the traditional interactive multiple model estimation,in the framework of distributed fusion estimation,the recursive optimization strategy based on minimizing Kullback-Leibler divergence criterion is used to predict and update the motion model transfer probability matrix.On this basis,the joint estimation of target state and model probability at current time is realized by variational Bayesian inference.Finally,the local state estimates’fusion is completed based on the covariance intersection fusion strategy.The simulation results show that the new algorithm effectively improves the state estimation accuracy of the maneuvering target by adaptively estimating the motion model transition probability matrix and the model probability online.

关 键 词：机动目标跟踪 变分贝叶斯推断 模型转移概率矩阵 分布式融合 协方差交叉融合 

分 类 号：O212.8[理学—概率论与数理统计] TP212[理学—数学]