基于小波聚类的QAR数据异常检测算法研究

发布时间：2018-01-23 05:47

本文关键词： QAR数据小波聚类算法异常点检测相似性匹配故障分析　出处：《中国民航大学》2017年硕士论文　论文类型：学位论文

【摘要】：随着社会的发展进步,选择民航客机出行的人数越来越多,飞机的飞行安全得到越来越多的重视,直接或间接影响了民航的发展,如何有效的发现飞机飞行过程中潜在的安全隐患变得尤为重要。目前QAR被应用于存储飞机上机载设备中获取的参数,因此分析QAR数据中潜在的故障信息为飞机的后续故障定位、解决故障提供保障有很重要的实用意义。本文为了发现QAR数据中的异常点信息,研究了小波聚类算法在QAR数据异常点检测上的应用,并通过将获取的异常点与专家经验下的标准故障样本进行相似性匹配来发现发生故障的原因,从而有效的帮助飞机维修人员定位故障原因并解决故障。为此完成了以下所述的研究:1、完成了QAR数据的异常点检测算法与故障分析。该方法首先获取飞机飞行过程中巡航阶段的数据,通过使用小波聚类算法对其做聚类分析来发现QAR数据中隐藏的异常信息。然后根据构建的标准故障模型,使用得到的异常点数据和标准故障模型做相似性匹配来定位故障类型,并检验该方法查找的故障是否正确以检测该方法的效率。2、对小波聚类算法进行深入细致的分析,发现了小波聚类算法标记满足密度阈值的连通单元为同一个簇,而该连通单元中的网格有可能是几个簇的边界,也可能不属于任何一个簇,导致不同的簇最后被标记为同一个簇的MUAP问题;另外不满足密度阈值的网格也可能存在属于簇的数据对象和簇的边界区分不准的问题,从而提出改进的小波聚类算法划分非均匀网格,并对边界网格进一步细化及对不满足密度值的网格进一步处理,最后形成聚类并验证此方法的可行性。即改进的小波聚类算法较传统的小波聚类算法在不提高其时间复杂度数量级的基础上提高了其效率。3、研究相似性度量并细致地分析了方法的原理,结合对QAR数据特点的分析发现QAR数据属性间的相关性并选择基于马氏距离对异常点数据和标准故障模型做相似性匹配来发现故障类型。
[Abstract]:With the development and progress of the society, more and more people choose civil aviation passenger aircraft to travel, and the flight safety of aircraft is paid more and more attention, which directly or indirectly affects the development of civil aviation. It is very important to find the potential safety hazards in the flight process. At present, QAR is used to store the parameters obtained from the airborne equipment on the aircraft. Therefore, it is of great practical significance to analyze the potential fault information in QAR data for the subsequent fault location of the aircraft and to provide protection for the failure. In order to find the abnormal point information in the QAR data, this paper aims at finding the outliers in the QAR data. The application of wavelet clustering algorithm in outlier detection of QAR data is studied, and the reason of fault is found by matching the obtained outliers with standard fault samples under expert experience. Thus effectively helps the aircraft maintenance personnel locate the malfunction reason and solve the malfunction. For this reason, completed the following research: 1. The outlier detection algorithm and fault analysis of QAR data are completed. Firstly, the data of cruising phase in the flight process of aircraft are obtained. Wavelet clustering algorithm is used to cluster analysis to find the hidden abnormal information in QAR data. Then according to the constructed standard fault model. The outlier data and the standard fault model are used for similarity matching to locate the fault type and to check whether the fault found by this method is correct to detect the efficiency of the method. 2. The wavelet clustering algorithm is analyzed in detail, and it is found that the connected cells that meet the density threshold are the same cluster, and the grid in the connected cell may be the boundary of several clusters. It may not belong to any cluster, which leads to the MUAP problem that different clusters are finally labeled as the same cluster. In addition, the mesh which does not satisfy the density threshold may have the problem of inaccuracy between the data objects belonging to the cluster and the boundary of the cluster, so an improved wavelet clustering algorithm is proposed to divide the non-uniform mesh. The boundary mesh is further refined and the mesh that does not satisfy the density value is further processed. Finally, clustering is formed and the feasibility of this method is verified. That is, the improved wavelet clustering algorithm improves its efficiency on the basis of not increasing the order of magnitude of its time complexity compared with the traditional wavelet clustering algorithm. The similarity measurement is studied and the principle of the method is analyzed in detail. Based on the analysis of the characteristics of QAR data, the correlation between the attributes of QAR data is found and the fault type is found by matching the outlier data with the standard fault model based on Markov distance.
【学位授予单位】：中国民航大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：V241;TP311.13

【参考文献】