【作者】 徐前方；

【导师】 郭军；

【作者基本信息】 北京邮电大学 ， 信号与信息处理， 2007， 博士

【摘要】 现代电信网络管理的主要任务是对网络进行实时监控，确保电信网络高效、可靠、经济和安全的运行。随着现代电信网络的规模越来越大，结构日益复杂，对电信网络的告警数据进行相关性分析尤其重要，因为从告警数据分析出的相关性知识，可以帮助网络管理人员及时定位故障，保证电信网络的正常运行。传统的相关性分析方法由于过多地依赖专家知识而难以适应网络复杂、多变的情况，采用数据挖掘的方法则可以弥补这方面的不足。随着网络规模的增大，告警数量的增加，如何从海量告警数据中发现电信网络中的告警相关性知识，帮助网络管理人员处理网络故障，是当前网络故障管理所面临的主要问题。本文将数据挖掘技术应用到网络故障告警相关性分析中，研究了频繁告警序列和非频繁告警序列关联规则的挖掘方法，取得了一定的成果。本文的具体工作与创新包括以下几个方面：1、告警序列模式(频繁告警序列)的挖掘研究了电信网络中告警序列模式的挖掘问题。序列模式挖掘是在关联规则挖掘的基础上发展起来的。目前序列模式挖掘方法大多基于WINEPI的算法框架，由于该算法需要多次遍历数据库，执行效率较低，因此本文基于FP-growth算法框架，提出一种基于FP-树的序列模式挖掘FSPM-FP算法。并且分别对其重要参数(最小支持度)和数据库发生变化的情况，提出了相应的增量式挖掘算法——SFSPM-FP和DFSPM-FP，并通过实验证明了算法的有效性。2、非频繁告警关联规则挖掘针对目前告警序列模式挖掘算法受到最小支持度的限制，仅能够得到高支持度、高置信度条件下频繁发生的告警关联规则的问题。本文结合实际电信网络告警的特征，提出了一种以高相关度、高置信度为条件，基于相关度统计的告警关联规则挖掘算法AARSC；同时为了适应告警数据动态增加的特点，提出了其改进算法——增量式挖掘算法UAARSC。实验表明AARSC和UAARSC算法可以同时发现频繁和非频繁发生告警序列间的关联规则，从而提高了告警关联规则的完整性和准确性。3、告警模式的可视化由于电信运营商经常会根据业务需求，对网络进行优化，为了有利于网管人员对设备进行维护，可以将当前网络中存在的告警以可视化的方式呈现给网管人员，他们根据可视化的结果，有效地发现故障的告警模式，进而预测告警，定位故障。本文提出一种基于谱图理论的ACASG算法。该算法基于谱图理论发现高维数据空间中潜在的低维映射结构；通过分析低维空间中点结构之间的相似性，实现告警模式挖掘的目的。实验结果表明，该算法不仅可以发现告警间的相关性，而且还可以通过分析谱图的变化，预测、定位网络故障。更多还原

【Abstract】 The task of telecommunication networks management focuses on monitoring the status of network, which ensures the network to run realiablely and efficiently. As the modern telecom network becomes large scale and its construction goes complex, it is much more important to analysis the alram correlation. Because the result can help the network administrators locate the fault to ensure the network running smoothly. However, traditional alarm correlation analysis methods can hardly work well when networks are complex and changeful due to its relying on expert knowledge, the data mining method can overcome the shortage of traditional methods. With the development of telecommunication networks, it is key issue for network management to extract the alarm correlation rules from massive alarm data to help network administrator to handle the fault.In this thesis, we apply data mining technology to network fault alarm correlation analysis and study mining methods in frequently and non-frequently alarm sequence. The research and innovations are described in details as follows:1. Alarm sequence pattern mining (high-frequency alarm sequence) We research alarm sequence pattern mining in telecom network. Sequence pattern mining is developed based on association rules mining. Nowadays, most sequence pattern mining methods are based on WINEPI algorithm frame. But WINEPI algorithm needs to traversal the database for many times so that it is not very efficiently. In order to overcome this shortage, we proposed a sequence pattern mining FSPM-FP algorithm, which is based on FP-Tree, and then we proposed two increments mining algorithm which are SFSPM-FP algorithm based on the minimal support changed and DFSPM-FP algorithm based on the database changed. Experimental results demonstrate The validity of the algorithm.2. low-frequence alarm association rules miningCurrently those algorithms to mine the alarm association rules are limited to the minimal support, so that they can only obtain the association rules among the frequently occurring alarm events based on high support and high confidence. In the thesis the alarm association rules mining algorithm AARSC is presented, which is used high correlativity and the high confidence. At the same time, we proposed its improved algorithm UAARSC to adapt the database increasing. Experimental result shows AARSC algorithm and UAARSC algorithm can discovery both high-frequency and low-frequency alarm association ruls so that it can make the rules are much more completly and correctly.3、Alarm pattern visualizationTelecom network often changes as the service changes. Alarm visualization can help administrators maintain network devices, so that they can discovery the alarm pattern efficiently and locate and predict the alarm and fault. A new ACASG algorithm is introduced, which is based on spectral graph theory. The algorithms discover the underlying mapping structure lying on a low-dimensional structure based on spectral graph theory and mine alarm pattern by analyzing point construction similarity. Experimental results demonstrate that the algorithm not only discovered correlation among alarms but also acquire the fault in the telecommunications network based on the spectral graph transformation.更多还原