【作者】 李超；

【导师】 张振江；

【作者基本信息】 北京交通大学 ， 通信与信息系统， 2021， 博士

【摘要】 可穿戴网络作为物联网的重要组成部分之一,在军事、医疗、工业等领域均有重要的应用。近些年,随着电子产业以及大数据相关技术的迅猛发展,可穿戴网络与各领域的结合更加紧密与深入,对于可穿戴网络中信息的精度也提出了更高的要求。数据融合技术可以利用原始数据的冗余信息进一步提高数据的精度与可用性;数据估计技术则可以根据原始数据的状态信息进行数据的处理,去除或减少噪声的影响。可穿戴网络数据处理中的高精度数据融合估计技术已经成为可穿戴网络的研究热点之一。可穿戴网络主要是对个人或群体的运动数据和状态数据进行采集和处理。一方面,数据的采集过程中不可避免的混入了噪声,使得数据存在不确定性,如何降低不确定性或充分利用不确定性信息提高数据精度是本领域研究的一个重要课题;另一方面,可穿戴网络采集和处理后的数据,需要进一步的数据分析,以得到最终的、可用的结论,如何将数据的融合估计与后续的数据分析紧密结合,充分利用数据信息,也是本领域研究的一个重要内容。本论文结合可穿戴网络数据处理中存在的实际问题,从上述两个角度分别对高精度融合估计技术进行了深入的研究:（1）针对非平稳过程的出现会导致估计结果存在一定的滞后性和精度出现明显下降的问题,提出了一种基于无损卡尔曼估计的突发事件感知融合估计模型,该模型引入了绝对差阈值用于发现数据流的非平稳过程,并且在非平稳过程出现时主动进行模型参数调整,一定程度上避免了非平稳过程导致的数据精度下降的问题,仿真实验表明,非平稳阶段的出现对所提模型的融合估计精度的影响不大。而且基于本模型的分类算法,其分类精度至少提升了4%。（2）针对采集数据的不确定性差异较大时,基于可能世界聚类算法的精度快速下降甚至失效的问题,提出了一种基于可能世界和K-L散度的融合估计模型。该模型将估计结果看作概率分布进行处理,充分利用了数据的不确定性。模拟数据和真实数据仿真分析表明:模型在采集数据的不确定性差异较大时,提高了聚类精度,避免了聚类算法失效,扩展了聚类算法的应用范围。（3）针对可穿戴网络中已经存在的采集数据具有一定关联性时,算法的复杂度高,估计精度低的问题,提出了一种基于吸引子的二次融合估计模型。该模型利用吸引子思想对未知的状态数据关系进行了近似拟合,得到了状态数据间数值上的关系表达式,并利用该表达式对融合估计结果进行二次处理,进一步提高了数据的融合估计精度。仿真分析表明:模型在一定条件下可以提高数据的精度,并使得同类数据的分布更加集中。（4）针对基于卡尔曼思想的融合估计算法无法处理噪声不服从正态分布下的数据,以及目前基于降维和数据离散化的处理能耗偏大的问题,提出了基于混合H₂/H_∞的轻量级融合估计模型。该模型使用了混合H₂/H_∞估计算法对状态转移方程进行融合估计,弱化了模型对噪声的要求;同时考虑到可穿戴设备中能量有限的问题,使用数据压缩方法降低可穿戴设备的数据通信消耗,达到降低能量消耗,延长使用寿命的目的。仿真分析表明:该模型在提高整体估计精度的同时,模型中各设备的通信量较其他同类模型有所降低,并在噪声剧烈变化时保持了较强的鲁棒性。更多还原

【Abstract】 As one of the important components of Internet of Things（Io T）,wearable body network（WBN）has various of applications in military,medical,industrial,and other fields.In recent years,the rapid development of electronics industry and big data-related technologies has put forward higher requirements on the accuracy of information in WBN.On one hand,data fusion can further improve the accuracy and availability of data by using the redundant information of original data.On the other hand,the data estimation can remove or reduce the influence of noise according to the state information of the original data.Therefore,high precision data fusion estimation technology in WBN data processing is one of the research hotspots.WBN mainly collect and process movement data and status data of individuals or groups.Because of the limitations of the equipment and surrounding environment,noise is mixed into the data,which makes the data uncertain.Therefore,reducing uncertainty or making full use of uncertainty information to improve data accuracy is an important research topic in this field.Besides,after data collection and processing,further data analysis is needed to obtain the final and usable conclusion.Therefore,it is another important research in this field to combine data fusion estimation with subsequent data analysis and make full use of data information.In this paper,combined with the actual problems in WBN data processing,the high-precision fusion estimation technology is studied from the above two perspectives respectively:（1）The emergence of non-stationary processes would lead to a certain delay and a significant decrease in the accuracy when calculating the estimation results.To solve the above problems,an Unscented Kalman Filter-based perceptual fusion estimation model for emergencies is proposed.The model introduces the absolute difference threshold to discover the non-stationary process of data flow,and adjusts the model parameters adaptively when the non-stationary process occurs,which avoids the problem of data accuracy decline caused by the non-stationary process to some extent.The simulations show that the non-stationary phase has little influence on the fusion estimation accuracy of the proposed model.Moreover,the classification accuracy of KNN algorithm based on this model is improved by at least 4%.（2）The clustering accuracy of the possible world-based algorithm declines rapidly or even fails when the uncertainty difference of the collected data is large.To solve this problem,a fusion estimation model based on possible worlds and K-L divergence is proposed.The model treats the estimated results as probability distribution and makes full use of the uncertainty of the data.Simulation analyses of simulated data and real data show that the model improves the clustering accuracy,avoids the failure of clustering algorithm,and extends the application range of clustering algorithm when there is a big difference in the uncertainty of data collected.（3）When there is a certain correlation between the collected data in the wearable network,the related algorithms are of high complexity and low estimation accuracy.To avoid above situation,a attractor based quadratic fusion estimation model is proposed.The model uses cluster attractor to approximate the unknown relationship in state data,obtains the numerical relation expression between the state data.And then uses this expression to carry on the second processing to the fusion estimation result,further enhances the accuracy of data fusion estimation.Simulation analyses show that the model can improve the precision of data under certain conditions and make the distribution of data in the same cluster more centralized.（4）The fusion estimation algorithm based on Kalman’s idea cannot deal with the data under normal distribution without noise,and the current processing energy consumption based on reduction and data discretization is too large.To solve this problem,a mixed H₂/H∞-based lightweight fusion estimation model is proposed.The model uses mixed H₂/H∞estimation algorithm for fusion estimation,which weakens the requirements of the model on noise.At the same time,considering the limited energy in wearable devices,the data compression method is used to reduce the data transmission consumption.The simulation analyses show that the model improves the overall estimation accuracy,the traffic of each device in the model is lower than that of other similar models,and maintains strong robustness when the noise changes dramatically.更多还原