【作者】 周斌；

【导师】 刘大有；

【作者基本信息】 吉林大学 ， 计算机应用技术， 2014， 硕士

【摘要】 随着多传感器技术的快速发展，涌现出大量各种应用背景下的多传感器系统，信息融合技术在许多领域得到应用。在这些传感器系统中，信息往往呈现出复杂性，这种复杂性主要体现在数量上的巨大性和表现形式上的多样性。但是传统的信息融合大都针对简单的、少量的数据进行应用研究。随着海量数据时代的到来，传统的信息融合方法存在着局限性。本文以环境舒适度评价为研究背景，通过对基于证据理论的海量信息融合的研究，围绕多源海量信息进行了实验探索，为高效、合理地利用海量数据提供理论和技术基础。本文提出一种结合聚类和凸函数证据理论的海量信息融合方法。该方法的主要思想是，首先基于BIRCH算法对采集的海量信息进行聚类预处理，形成多个簇。然后，计算出每个簇的质心，将质心作为各个簇的代表信息，再基于广义三角模糊隶属函数对每个质心信息进行基本概率指派。最后，基于凸函数证据理论完成对各证据的组合，从而完成对海量信息的融合。本文通过在一个室内环境舒适度判定系统中验证了方法的有效性，该方法也可以很方便地推广到其它状态评价系统中。本文以室内环境舒适度为应用背景，选取温度、湿度和PM2.5这三个重要的影响因素，根据连续多天不同时段收集的大量真实的温湿度传感器数据和PM2.5数据，对算法进行实验验证。由于温湿度传感器数据收集的数据量比较大，我们先对温湿度数据进行聚类融合，再将结果与PM2.5数据进行融合。本文在使用BIRCH对温湿度数据聚类时，根据数据的特点和实验确定了节点阈值和分支因子，用聚类后的多个簇来代替原数据，并用簇的质心代表各个簇。本文在采用三角模糊隶属函数的方法构造基本概率函数时，由于本应用中存在着一种舒适度对应多个温度或湿度区间的情况，我们在基于模糊隶属函数方法的基础上做了一些改变，针对每一个区间都建立一个隶属度函数，最后选择似然函数大的作为该温湿度对应的舒适度的似然值，再处理生成证据的mass值。在凸函数证据理论的组合公式中，一般不同证据的影响因子是不一样的，为此，我们根据数据的特点，设计以每个簇中数据的条数在整个数据中的比例来作为影响因子，这样能够比较合理地指派权重。我们先使用凸函数证据理论组合公式求得温度和湿度对各舒适度的支持程度，再将温湿度的融合结果与PM2.5数据组合，得出最终的实验结果。通过与直接融合算法实验数据对比可知，本文提出的算法能在保证融合结果正确的前提下，大幅度减少海量信息的融合时间，是一种合理高效的海量信息融合新方法，为海量信息融合技术的发展提供了一条新的探索途径。最后本文给出了基于hadoop平台下的本文算法实现。更多还原

【Abstract】 With the rapid development of multi-sensor technology, a variety of complexapplications-oriented background sensor system are emerged, information fusion technologyhas been applied in many fields. In these sensor systems, the information is often presentedcomplex, this complexity is mainly reflected in the huge number and diversity forms ofexpression. However, most of the traditional information fusion research is used for simple andsmall amounts of data. With the advent of the massive data era, there are limitations oftraditional information fusion. In this paper, evaluation of environmental comfort as theresearch background, through the mass of information fusion based on evidence theory, andmassive multi-source information conducted experiments to explore, for the efficient andrational use of massive amounts of data to provide theoretical and technical basis. This paperproposed a massive information fusion method based on evidence theory, it can effectivelysolve the information is incomplete, uncertain, large volumes of data and other difficulties.This paper presents a combined mass of information clustering and convex function ofevidence theory fusion method. The main idea of this method is to first collected pretreatmentmass of information by BIRCH clustering algorithm, forming a plurality of clusters. Secondly,the centroid of each cluster is calculated to be the representation of every cluster. Then, to formthe evidence provided by the information in each cluster, the centroid information is given theBasic Probability Assignment based on the generalized triangular fuzzy membership function.Finally, evidences are combined according to the combination rule of the Convex EvidenceTheory. So far, the massive information fusion is completed. In this paper, an indoorenvironmental comfort judging system to verify the effectiveness of the method, the methodcan also be easily extended to other state assessment system.In this paper, indoor environmental comfort as application background, select thetemperature, humidity and PM2.5three important factors, according to a large number of realtemperature and humidity sensor data and number of continuous days PM2.5data collected atdifferent times, the algorithm experimental verification. Since the data of temperature andhumidity sensor data collection more, before the data integration, we need to use BIRCHclustering the data, this paper identifies the node threshold and branching factor according to the characteristics and experimental data, using multiple cluster after cluster instead of theoriginal data, and with representatives of each cluster centroid clusters. Information obtainedafter each cluster, we need to generate the basic probability function, we use the method oftriangular fuzzy membership functions. Since the application of the existence of acorresponding plurality of comfort temperature or humidity range, for which we have the basisof fuzzy membership function based on the method made some changes have been establishedfor each interval of a membership function, and finally chose a large likelihood functioncorresponding to the temperature and humidity, as the likelihood of comfort, and thenprocessed to generate mass value of the evidence. In combination formula convex function ofevidence theory in general, evidence of the impact of different factors are not the same, and wehave the characteristics of the data, the number of design data to the proportion of each clusterin the whole data as a factor this can be quite reasonable to assign weights. Finally, the use ofconvex combination formula fusion. By compare with the direct data fusion algorithm,experiment shows that the proposed algorithm can result in ensuring the correct premiseintegration, and significantly reduce the time of massive information fusion, it is a reasonableand efficient mass of information fusion method. It is provides a new way to explore themassive information fusion technology. Finally, we give the realization of the algorithm basedon hadoop platform.更多还原