复杂网络社团探测方法及在轮机故障诊断中应用的研究
发布时间：2018-11-11 20:17
【摘要】：复杂网络作为一门结合了数学、物理学、计算机图形学和社会学等多种知识的新兴技术,是21世纪各领域研究人员关注的一个重点。复杂网络由大量的节点和边组成,绝大部分真实网络都由一些内部连接稠密而彼此之间连接稀疏的节点群组成,即具有社团结构。社团探测是复杂网络理论的一个重要研究方向,帮助人们从中观角度了解复杂系统及其代表的各种现象。Newman快速算法与标签传播算法是两种经典的社团探测方法,由于探测速度快且不需要预先指定社团数目,得到了普遍的关注。社团探测方法的应用多在于真实网络聚类,对于聚类问题的另一大分支——数据聚类则研究甚少,而数据聚类一直是解决船舶柴油机故障诊断问题的一个重要手段。船舶柴油机是船舶的心脏,利用Newman快速算法和标签传播算法的优势解决船舶柴油机故障诊断问题对维护航行安全有着重要意义。本论文从实际应用的需求出发,研究了标签传播算法的推广与改进策略和基于社团探测理论的船舶柴油机故障诊断方法,主要研究工作包含以下几方面内容。1.利用Newman快速算法在聚类问题中自行确定类数的特点,提出基于Newman快速算法的船舶柴油机故障诊断方法。以样本为节点、样本间相似度为边权,构建有权无向的复杂网络,并以Newman快速算法中的准则函数作为自底向上的层次聚类的准则函数,建立聚类方法模型,对船舶柴油机故障样本进行数据聚类,并使用聚类结果对待识别样本进行故障类型识别。诊断实例和影响因素试验结果表明,该方法对类数等初始条件要求低、运行时间短、准确率高且具有一定的稳定性,能够识别出历史数据中不存在的故障类型。2.为提高标签传播算法的实用性,推广了标签传播算法,使其适用于有权网络,从而能够用于船舶柴油机故障数据聚类。通过分析得知标签传播算法的三个关键因素为标签初始分配、标签传播规则和传播终止条件,根据多重边的原则计算两相邻节点同社团的概率,加权了标签传播规则和标签传播的终止条件,从而将标签传播算法推广到有权情况。网络社团探测试验结果表明,推广后的标签传播算法适用于有权网络社团探测；同时经典测试数据集和柴油机供油系统故障数据集的聚类试验结果表明,推广后的标签传播算法适用于数据聚类。3.针对标签传播过程中容易出现平凡解的问题,提出了基于逾渗转变预测过程的标签传播算法。原标签传播算法的随机性导致了平凡解的出现,影响了算法的速度和准确性。通过转化标签传播过程为网络构建过程,将随机网络生成过程中的逾渗转变现象与平凡解的出现联系起来,从而通过在标签传播过程中添加逾渗转变的预测过程来减少平凡解的出现。推广邻居纯度的概念到有权网络,并给出考虑被更新标签的节点度的不完全更新条件来节省计算时间。网络试验结果表明改进后的标签传播算法对小社团的敏感度与解的稳定性,不完全更新条件使算法更加省时；船舶柴油机故障数据集上的聚类试验结果表明,改进后的算法不容易遗漏规模较小的类,对故障诊断中样本不均的情况同样适用。4.针对故障诊断过程中单次聚类方法容易引起信息损失,多重聚类方法需要调节预设参数或方法的问题,利用标签传播算法可能获得多种解的特点,提出了基于多次标签传播的船舶柴油机故障诊断方法。使用改进后的标签传播算法对船舶柴油机的故障数据多次聚类,整合得到的多个结果或确认得到的唯一结果作为最终聚类结果,利用得到的聚类中心判断待识别样本类型。诊断实例和影响因素试验结果表明,该方法无需修改预设参数,信息损失较单次聚类少,运行时间短且具有较高的准确率和较好的稳定性,能够识别出历史数据中不存在的故障类型。
[Abstract]:As a new technology that combines many kinds of knowledge, such as mathematics, physics, computer graphics and sociology, the complex network is an important focus of researchers in all fields of the 21st century. The complex network consists of a large number of nodes and edges, and most of the real networks are composed of a plurality of nodes which are densely connected and sparse with each other, that is, the complex network has a community structure. The community detection is an important research direction of the complex network theory, which helps people to understand the complex system and its representative from the point of view. The Newman fast algorithm and the tag propagation algorithm are two classical community detection methods, which are of general interest due to the rapid detection speed and no need to pre-specify the number of associations. The application of community detection method is one of the important means to solve the problem of the fault diagnosis of marine diesel engine. The marine diesel engine is the heart of the ship, and the advantage of the Newman fast algorithm and the tag propagation algorithm is of great significance to the maintenance and navigation safety of the marine diesel engine. Based on the demand of practical application, this paper studies the extension and improvement strategy of label propagation algorithm and the method of fault diagnosis of marine diesel engine based on community detection theory. The main research work includes the following aspects. This paper presents a new method for fault diagnosis of marine diesel engine based on the Newman fast algorithm. using the sample as the node, the similarity between the samples is the edge weight, the complex network with the right is constructed, and the standard function in the Newman fast algorithm is used as a criterion function of the hierarchical clustering of the bottom up, a clustering method model is established, and the data aggregation of the fault sample of the marine diesel engine is carried out, and the classification result is used to treat the identification sample to identify the fault type. The results of the test show that the method has the advantages of low requirement for initial conditions such as class number, short operation time, high accuracy and certain stability, and can identify the type of fault that does not exist in the historical data. In order to improve the practicability of the label propagation algorithm, the label propagation algorithm is extended so that it is applicable to the network, so that it can be used in the fault data aggregation of the marine diesel engine. By analyzing the three key factors of the label propagation algorithm, the label initial assignment, the label propagation rule and the propagation termination condition are analyzed, the probability of the two adjacent nodes and the community is calculated according to the principle of the multiple edges, the label propagation rule and the termination condition of the label propagation are weighted, thereby promoting the tag propagation algorithm to the right. The results of the network community survey show that the extended label propagation algorithm is applicable to the network community detection; meanwhile, the test results of the classical test data set and the data set of the system fault data of the diesel engine show that the generalized label propagation algorithm is suitable for the data aggregation. In order to solve the problem of trivial solution in the process of label propagation, a label propagation algorithm based on the percolation transition prediction process is proposed. The randomness of the original label propagation algorithm leads to the occurrence of the trivial solution, which affects the speed and the accuracy of the algorithm. The transformation label propagation process is a network construction process, and the percolation transition phenomenon in the random network generation process is linked with the occurrence of the trivial solution, so that the occurrence of the trivial solution is reduced by adding the percolation transition prediction process in the label propagation process. the concept of promoting the purity of the neighbor is to the right network and the calculation time is saved taking into account the incomplete update condition of the node degree of the updated tag. The results of the network test show that the improved algorithm for the sensitivity and solution stability of the small community and the incomplete updating condition make the algorithm more efficient. The results of the poly-type test on the fault data set of the marine diesel engine show that the improved algorithm is not easy to miss the smaller class, The same applies to the case of non-uniform samples in the fault diagnosis. Aiming at the problem that the single-time poly-class method in the fault diagnosis process can easily cause the loss of information, the multi-aggregation method needs to adjust the problem of the preset parameters or the method, and the characteristic of various solutions can be obtained by using the label propagation algorithm, and the fault diagnosis method of the marine diesel engine based on the multi-label propagation is proposed. and using the improved label propagation algorithm to make the fault data of the marine diesel engine multiple times, and the obtained result or the confirmed unique result is used as the final polytype result, and the obtained polytype center judges the type of the sample to be identified. The experimental results of the diagnosis and the influencing factors show that the method does not need to modify the preset parameters, the information loss is less than that of the single sub-group, the running time is short and the operation time is short and has high accuracy and good stability, and the fault type that does not exist in the historical data can be identified.
【学位授予单位】：大连海事大学
【学位级别】：博士
【学位授予年份】：2015
【分类号】：U672
[Abstract]:As a new technology that combines many kinds of knowledge, such as mathematics, physics, computer graphics and sociology, the complex network is an important focus of researchers in all fields of the 21st century. The complex network consists of a large number of nodes and edges, and most of the real networks are composed of a plurality of nodes which are densely connected and sparse with each other, that is, the complex network has a community structure. The community detection is an important research direction of the complex network theory, which helps people to understand the complex system and its representative from the point of view. The Newman fast algorithm and the tag propagation algorithm are two classical community detection methods, which are of general interest due to the rapid detection speed and no need to pre-specify the number of associations. The application of community detection method is one of the important means to solve the problem of the fault diagnosis of marine diesel engine. The marine diesel engine is the heart of the ship, and the advantage of the Newman fast algorithm and the tag propagation algorithm is of great significance to the maintenance and navigation safety of the marine diesel engine. Based on the demand of practical application, this paper studies the extension and improvement strategy of label propagation algorithm and the method of fault diagnosis of marine diesel engine based on community detection theory. The main research work includes the following aspects. This paper presents a new method for fault diagnosis of marine diesel engine based on the Newman fast algorithm. using the sample as the node, the similarity between the samples is the edge weight, the complex network with the right is constructed, and the standard function in the Newman fast algorithm is used as a criterion function of the hierarchical clustering of the bottom up, a clustering method model is established, and the data aggregation of the fault sample of the marine diesel engine is carried out, and the classification result is used to treat the identification sample to identify the fault type. The results of the test show that the method has the advantages of low requirement for initial conditions such as class number, short operation time, high accuracy and certain stability, and can identify the type of fault that does not exist in the historical data. In order to improve the practicability of the label propagation algorithm, the label propagation algorithm is extended so that it is applicable to the network, so that it can be used in the fault data aggregation of the marine diesel engine. By analyzing the three key factors of the label propagation algorithm, the label initial assignment, the label propagation rule and the propagation termination condition are analyzed, the probability of the two adjacent nodes and the community is calculated according to the principle of the multiple edges, the label propagation rule and the termination condition of the label propagation are weighted, thereby promoting the tag propagation algorithm to the right. The results of the network community survey show that the extended label propagation algorithm is applicable to the network community detection; meanwhile, the test results of the classical test data set and the data set of the system fault data of the diesel engine show that the generalized label propagation algorithm is suitable for the data aggregation. In order to solve the problem of trivial solution in the process of label propagation, a label propagation algorithm based on the percolation transition prediction process is proposed. The randomness of the original label propagation algorithm leads to the occurrence of the trivial solution, which affects the speed and the accuracy of the algorithm. The transformation label propagation process is a network construction process, and the percolation transition phenomenon in the random network generation process is linked with the occurrence of the trivial solution, so that the occurrence of the trivial solution is reduced by adding the percolation transition prediction process in the label propagation process. the concept of promoting the purity of the neighbor is to the right network and the calculation time is saved taking into account the incomplete update condition of the node degree of the updated tag. The results of the network test show that the improved algorithm for the sensitivity and solution stability of the small community and the incomplete updating condition make the algorithm more efficient. The results of the poly-type test on the fault data set of the marine diesel engine show that the improved algorithm is not easy to miss the smaller class, The same applies to the case of non-uniform samples in the fault diagnosis. Aiming at the problem that the single-time poly-class method in the fault diagnosis process can easily cause the loss of information, the multi-aggregation method needs to adjust the problem of the preset parameters or the method, and the characteristic of various solutions can be obtained by using the label propagation algorithm, and the fault diagnosis method of the marine diesel engine based on the multi-label propagation is proposed. and using the improved label propagation algorithm to make the fault data of the marine diesel engine multiple times, and the obtained result or the confirmed unique result is used as the final polytype result, and the obtained polytype center judges the type of the sample to be identified. The experimental results of the diagnosis and the influencing factors show that the method does not need to modify the preset parameters, the information loss is less than that of the single sub-group, the running time is short and the operation time is short and has high accuracy and good stability, and the fault type that does not exist in the historical data can be identified.
【学位授予单位】：大连海事大学
【学位级别】：博士
【学位授予年份】：2015
【分类号】：U672
【参考文献】
相关期刊论文 前10条
1 邓国红,曹龙汉,刘进,李建勇,蒲红梅;一种柴油机燃油系统故障诊断的新方法[J];重庆大学学报(自然科学版);2004年10期
2 郭文勇,朴甲哲,张永祥;基于多信息的柴油机缸套磨损故障诊断研究[J];海军工程大学学报;2005年01期
3 冯二浩;陆辉山;潘宏侠;;柴油机故障诊断研究综述[J];机械管理开发;2010年02期
4 王观玉;;基于聚类的复杂网络社团发现算法[J];计算机工程;2011年10期
5 何岳华;王海波;曾利平;;基于聚类的模糊柴油机故障诊断[J];计算机仿真;2011年10期
6 吴欠欠;王直;董贺;;故障树分析法在船舶柴油机故障诊断中的应用研究[J];机械设计与制造;2009年01期
7 杜海峰;王娜;张进华;邵颉;王孙安;;基于复杂网络的故障诊断策略[J];机械工程学报;2010年03期
8 何东晓;周栩;王佐;周春光;王U，
本文编号：2326026
本文链接：https://www.wllwen.com/kejilunwen/chuanbolw/2326026.html
