多率量测下随机跳变系统迁移交互多模型估计  

Transfer Interacting Multiple Model State Estimator for Markovian Jump Linear Systems With Multi-rate Measurements

作　　者：高爽 栾小丽 赵顺毅 刘飞 GAO Shuang;LUAN Xiao-Li;ZHAO Shun-Yi;LIU Fei(Key Laboratory for Advanced Process Control of Light Industry of the Ministry of Education,Institute of Automation,Jiangnan University,Wuxi 214122)

机构地区：[1]江南大学自动化研究所轻工过程先进控制教育部重点实验室,无锡214122

出　　处：《自动化学报》2023年第1期210-218,共9页Acta Automatica Sinica

基　　金：国家自然科学基金(61991402,61833007,61991400);江苏省研究生科研与实践创新计划(KYCX21-2007)资助。

摘　　要：实际工业过程中,量测数据除了在线仪表采集的快速率数据,还有离线化验等慢速率辅助量测数据.为了更好地利用离线化验数据,增加在线估计的精度,针对随机跳变系统,引入迁移学习思想,提出迁移交互多模型估计(Transfer interacting multiple model state estimator,IMM-TF)新策略.首先,将离线化验数据的边缘分布作为可以迁移的知识,迁移到贝叶斯后验分布,实现辅助量测数据的充分利用.其次,利用KL(Kullback-Leibler)散度度量知识迁移前后任务间的差异性,求解最优的贝叶斯迁移估计器.同时,结合慢速率量测,利用平滑策略获取待迁移的估计值,解决多率量测下的迁移估计难题.然后,利用影响力函数构建辅助量测数据与估计性能之间的解析关系,从而对迁移效果进行定量评价.最后,通过在目标跟踪实例中的应用,表明所提方法的有效性及优越性.In industrial processes some measurements are sampled frequently while other measurements are available infrequently and often slow rate.To utilize the slow rate measurements better for improving the accuracy of online estimation,this paper proposes a powerful transfer interacting multiple model state estimator(IMM-TF)for Markovian jump linear systems with multi-rate measurements based on the transfer learning strategy.First,the form of knowledge to be transferred to the Bayesian posterior distribution is designated as the observation predictor derived by using the slow rate measurements.We define universal evaluation of relatedness between the distribution transferred knowledge and ideal posterior distribution from the perspective of Kullback-Leibler(KL)divergence to obtain the optimal Bayesian transfer state estimator.Integrated with the slow rate measurements,the smoothing strategy is then proposed to obtain the transferred estimates for solving the difficult problem of transfer state estimator facing multi-rate measurements.Furthermore,the influence function is defined to construct the analytical relationship between the slow rate measurements and the estimation performance,so as to quantitatively evaluate the transfer effect.Finally,the effectiveness and superiority of the proposed method are illustrated by an example of target tracking.

关 键 词：跳变系统 迁移交互多模型估计 多率量测 KL散度 平滑策略 

分 类 号：O211.6[理学—概率论与数理统计] O212.8[理学—数学]