【作者】 乔娟；

【导师】 陈仲民；

【作者基本信息】 华中农业大学 ， 资源环境信息工程， 2009， 硕士

【摘要】 随着网络技术的发展,网络资源呈现出异构性和动态性,网络管理的功能日益复杂,传统的网络管理技术已经无法满足对大型复杂网络管理的需要。结构复杂,规模巨大的电信网络每天产生大量告警。告警是一个异常的有害事件,通常是一个自动监测到的故障,提供给网管人员一定的信息。因为某一个故障可能引起一系列告警,所以并不是所有的告警都表明故障原因,故障管理系统想要准确定位网络故障相当困难。传统的网络管理系统和网络管理员只能依靠自身有限的经验和网络管理系统有限的功能进行故障诊断、定位和恢复。但在网络日益扩大且迅速演变的情况下,这些知识已无法满足需求。目前国内外学者对网络告警进行了很多的研究,有多种方法被应用于网络故障告警相关性分析中。其中,数据挖掘的方法被广泛研究和应用。本文首先对数据挖掘的概念、功能、基本流程等进行了概述。其次对电信网络基本结构以及电信网络告警数据特征进行了介绍。最后提出了数据挖掘技术在电信网络告警相关性分析中的应用,详细介绍了关联规则挖掘算法和序列模式挖掘算法。关联规则挖掘算法是一种常用的方法,其中Apriori算法是关联规则挖掘领域中最经典的算法之一,也是一种最具有影响力的挖掘布尔关联规则频繁项集的算法。关联规则算法的核心思想是基于频集理论的递推方法,所采用的是逐层搜索的迭代方法。Apriori算法会产生大量的侯选集,同时多次扫描数据库。针对Apriori算法存在的问题,Jiawei Han等人于2000年提出了另一个经典的算法——FP-growth算法,该算法基于FP-tree(频繁模式树)采取分治策略,从而在挖掘出全部频繁项集时并不产生候选项目集。文章对两种算法进行了比较,并以武汉电信某网管中心的告警数据为实验数据,对实验结果进行分析。序列模式挖掘是关联规则挖掘的延伸,序列是由许多有序事件组成的数据集合,是数据挖掘的一个重要分支,用于提取一维空间上有序集合的频繁子集。如果把网络告警信息库看成按时间排列的有序集合,那么序列模式挖掘就可以用来发现频繁出现的告警序列模式,从而导出告警关联规则。本文采用了基于FP树的告警频繁序列模式挖掘算法FSPMFP(Frequent Sequential Pattern Mining basedon Frequent Tree)的基本思想是:通过对FP树的改进,将告警数据压缩到一棵频繁模式树上,针对频繁模式树自底向上查找频繁告警项集,最后挖掘告警间的时序关系。更多还原

【Abstract】 Along with the development of the network technology.Netwok resource present heterogeneity and Dynamic. The function of Network Management increasinqly complexity. The traditional network management technology was already unable to satisfy to the large-scale complex network management need. Large-scale and complex structure of moden networks have a large number of alarms per day. Alarm is a harmful abnormal events. is usually monitors automatically failure. It offers definite information to network management personnel. As a result of a single fault may lead to a series of alarms, so not all of the alarm indicated that the cause of the malfunction..Fault management system want to accurate positioning netwok fault very difficult. Traditional network management system and network administrators must rely on their own limited experience and limited network management system fault diagnosis functions, location and recovery. However, expanding the network and the rapidly evolving circumstances, such knowledge has been unable to meet demand.At present, Scholars at home and abroad to carry out a lot of alarm network research, there are a variety of methods have been applied to a network failure in the alarm correlation analysis. the data mining methods have been extensively studied and applied.In this paper, first ,summarize the concept of data mining, functional, basic processes and so on. Second, introduce the basic structure of the telecommunication network as well as the characteristics of telecommunications networks of alarm data. Finally, pose the application of the data mining technology in the telecommunication network alarm correlation analysis. Detailed introduction algorithm for mining association rules and sequential pattern mining algorithm.Association rule mining algorithm is a commonly used method, Apriori algorithm is one of the most classic in the field of association rule mining algorithm, is also a most influential mining Boolean association rules algorithm for frequent itemsets. The algorithm core idea of the algorithm is based on the frequency of the recursive method of set theory,used layer by layer search iterative method. Apriori algorithm engender a large number of Candidate Sets, and repeatedly scan the database. Against to the problem of Apriori algorithm, Jiawei Han, who in 2000 proposed the algorithm for another classic - FP-growth algorithm. The algorithm is based on the FP-tree (Frequent Pattern Tree) to take sub-rule strategy. In mining frequent itemsets of all does not have a candidate itemsets, In this paper, two algorithms were compared, and use Wuhan Telecommunication Alarm Data Management Center for the experimental data, the analysis of experimental results.Sequential pattern mining is an extension of association rule mining, in an orderly sequence of events is composed of many data sets.Squence is a data collection by many orderly events,is an important branch of Data Mining. Used to extract one-dimensional space in an orderly collection of frequent subsets. If the network alarm information base is a time-ordered collection, then the sequential pattern mining can be used to find frequent sequential patterns of alarm, in order to export association rules of alarm. In this paper, based on FP-tree warning Frequent Sequential Pattern Mining Algorithm FSPMFP (Frequent Sequential Pattern Mining based on Frequent Tree) The basic idea is: First of all, through the improvement of FP-tree, data compression will be alert to the frequent pattern tree, and then for the frequent pattern Bottom-up tree to find frequent itemsets of alarm, The final excavation of the timing relationship between alarm.更多还原