高维数据聚类算法及应用研究
发布时间：2018-07-17 01:16
【摘要】：随着航运业的发展,船舶的数量和规模不断扩大,作为航行安全的重要保障,船舶故障诊断技术得到越来越多的重视。由于船舶设备种类繁多且参数复杂,导致船舶管理系统中采集的数据量庞大且维数较高,对故障诊断模块的数据处理性能提出了挑战,如何有效地处理海量高维数据成为故障诊断过程中的研究重点。本文以某海事局船舶管理系统为背景,重点研究了高维数据聚类技术,设计并实现了故障诊断模块,主要研究内容如下。本文在深入分析了高维数据聚类算法和传统聚类算法的基础上,设计了一个基于高维数据聚类算法的故障诊断框架,并详细阐述了该框架中各个组成部分的功能。针对故障诊断中出现的高维数据及其噪声信息,本文重点研究了基于正交非负矩阵分解的聚类算法和基于相似矩阵补全的集成聚类算法。为了降低高维数据的维数,提出了一种基于正交非负矩阵分解的K-means聚类算法,该算法对原始数据进行非负矩阵分解,并加入正交约束,保证低维特征的非负性,增加数据原型矩阵的正交性,降低了数据的维数特征,最后进行K-means聚类并验证该算法的有效性。为了解决高维数据中存在大量噪声的问题,提出了一种基于相似矩阵补全的聚类集成改进算法。该算法利用正交非负矩阵算法生成基聚类,在此基础上采用高维数据相似性度量函数Hsim构造每个基聚类的相似性矩阵,然后采用增广拉格朗日乘子法对相似性矩阵中缺失的元素进行补全,最后采用性能优越的谱聚类得到最终的数据划分。本文的研究成果初步应用于某海事局船舶管理系统中的故障诊断模块,以高维数据聚类算法为基础,实现了系统的故障诊断模块,取得了良好的应用结果。
[Abstract]:With the development of shipping industry, the number and scale of ships are expanding. As an important guarantee of navigation safety, ship fault diagnosis technology has been paid more and more attention. Because of the variety of ship equipment and the complexity of parameters, the data collected in the ship management system is huge and the dimension is high, which challenges the data processing performance of the fault diagnosis module. How to deal with massive high-dimensional data effectively becomes the focus of fault diagnosis. In this paper, based on the ship management system of a maritime bureau, the high-dimensional data clustering technology is studied, and the fault diagnosis module is designed and implemented. The main research contents are as follows. Based on the deep analysis of high-dimensional data clustering algorithm and traditional clustering algorithm, a fault diagnosis framework based on high-dimensional data clustering algorithm is designed in this paper, and the functions of each component of the framework are described in detail. Aiming at the high dimensional data and its noise information in fault diagnosis, this paper focuses on the clustering algorithm based on orthogonal nonnegative matrix decomposition and the integrated clustering algorithm based on similarity matrix complement. In order to reduce the dimension of high-dimensional data, a K-means clustering algorithm based on orthogonal non-negative matrix decomposition is proposed. It increases the orthogonality of the data prototype matrix and reduces the dimension feature of the data. Finally, K-means clustering is carried out and the validity of the algorithm is verified. In order to solve the problem of large amount of noise in high dimensional data, an improved clustering algorithm based on complement of similarity matrix is proposed. The algorithm uses orthogonal nonnegative matrix algorithm to generate base clustering. On this basis, the similarity matrix of each base cluster is constructed by using the high-dimensional data similarity measure function Hsim. Then the elements missing in the similarity matrix are complemented by the augmented Lagrangian multiplier method, and the final data partition is obtained by using the superior spectral clustering method. The research results of this paper have been applied to the fault diagnosis module of a ship management system of a maritime bureau. Based on the high-dimensional data clustering algorithm, the fault diagnosis module of the system has been realized, and good application results have been obtained.
【学位授予单位】：南京航空航天大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：U672.74
，
本文编号：2128384
[Abstract]:With the development of shipping industry, the number and scale of ships are expanding. As an important guarantee of navigation safety, ship fault diagnosis technology has been paid more and more attention. Because of the variety of ship equipment and the complexity of parameters, the data collected in the ship management system is huge and the dimension is high, which challenges the data processing performance of the fault diagnosis module. How to deal with massive high-dimensional data effectively becomes the focus of fault diagnosis. In this paper, based on the ship management system of a maritime bureau, the high-dimensional data clustering technology is studied, and the fault diagnosis module is designed and implemented. The main research contents are as follows. Based on the deep analysis of high-dimensional data clustering algorithm and traditional clustering algorithm, a fault diagnosis framework based on high-dimensional data clustering algorithm is designed in this paper, and the functions of each component of the framework are described in detail. Aiming at the high dimensional data and its noise information in fault diagnosis, this paper focuses on the clustering algorithm based on orthogonal nonnegative matrix decomposition and the integrated clustering algorithm based on similarity matrix complement. In order to reduce the dimension of high-dimensional data, a K-means clustering algorithm based on orthogonal non-negative matrix decomposition is proposed. It increases the orthogonality of the data prototype matrix and reduces the dimension feature of the data. Finally, K-means clustering is carried out and the validity of the algorithm is verified. In order to solve the problem of large amount of noise in high dimensional data, an improved clustering algorithm based on complement of similarity matrix is proposed. The algorithm uses orthogonal nonnegative matrix algorithm to generate base clustering. On this basis, the similarity matrix of each base cluster is constructed by using the high-dimensional data similarity measure function Hsim. Then the elements missing in the similarity matrix are complemented by the augmented Lagrangian multiplier method, and the final data partition is obtained by using the superior spectral clustering method. The research results of this paper have been applied to the fault diagnosis module of a ship management system of a maritime bureau. Based on the high-dimensional data clustering algorithm, the fault diagnosis module of the system has been realized, and good application results have been obtained.
【学位授予单位】：南京航空航天大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：U672.74
，
本文编号：2128384
本文链接：https://www.wllwen.com/kejilunwen/chuanbolw/2128384.html
