跳变马尔可夫系统的最大混合相关熵状态估计  

Maximum Mixture Correntropy State Estimation for Jump Markov Systems

作　　者：沈忱[1,2] MIHAYLOVA Lyudmila SHEN Chen;MIHAYLOVA Lyudmila(School of Information and Electronic Engineering(Sussex Artificial Intelligence Institute),Zhejiang Gongshang University,Hangzhou,Zhejiang 310018,China;Department of Automatic Control and Systems Engineering,University of Sheffield,Sheffield S102TN,UK)

机构地区：[1]浙江工商大学信息与电子工程学院(萨塞克斯人工智能学院),浙江杭州310018 [2]谢菲尔德大学自动控制与系统工程系,谢菲尔德S102TN

出　　处：《电子学报》2021年第11期2225-2233,共9页Acta Electronica Sinica

基　　金：浙江省自然科学基金(No.LQ18F030003);国家留学基金(No.201908330102)。

摘　　要：依托于多模型框架的跳变马尔可夫系统状态估计的性能通常受限于多模型间的信息融合程度.本文以交互式多模型方法为框架,针对跳变马尔可夫系统提出了一种基于最大混合相关熵的状态估计方法.为了能有效处理模型高阶信息,在混合和融合步骤引入最大混合相关熵测度替代常规的二阶统计矩准则,设计了关于系统状态的代价函数,通过最优化该函数得到状态估计的迭代解.仿真实验详尽展示了所提方法的主要特征,并表明其在高斯和非高斯噪声环境下都具有较好的估计效果.State estimation for jump Markov systems based on multiple models is usually influenced by the quality of model fusion.In this article,we propose a novel state estimation approach for the jump Markov systems based on the maxi⁃mum mixture correntropy criterion(MMCC).The proposed approach is implemented within the framework of the interact⁃ing multiple models.To capture high order information from multiple models,we utilize the MMCC instead of second-or⁃der statistical measures at the mixing and fusion stages,respectively.Two cost functions with respect to the system state at different stages are designed and optimized to yield the resultant iterative solutions.Extensive simulated results present the feature of the proposed MMCC based approach,and prove its efficacy for both Gaussian and non-Gaussian cases.

关 键 词：混合相关熵 状态估计 模型不确定 

分 类 号：TP212.6[自动化与计算机技术—检测技术与自动化装置]