【作者】 李艳；

【导师】 肖德宝；

【作者基本信息】 华中师范大学 ， 计算机应用技术， 2003， 硕士

【摘要】 故障管理作为网络管理的五大功能之一，负责对网络故障进行检测、诊断和恢复，而故障诊断又是其中的难点与重点，其有效与否和功能强弱直接关系到被管网络的可用性和可靠性。早期的故障管理是通过网管人员分析告警信息以人工方式实现，该方式代价高，效率低，无法进行实时、有效的故障诊断。引入人工智能技术的目的就是希望通过尽可能少的人工干预，使故障诊断能够独立的驱动智能诊断过程，实现其目标的自动控制。 但目前的智能故障诊断技术，多采用某种单一的人工智能技术运用于整个故障诊断过程，难以满足对复杂网络故障进行准确、有效的定位和诊断的全部要求。本文针对以上问题，将故障诊断分为故障定位和故障原因诊断两个阶段，针对各阶段任务的不同，采用不同的智能化技术，并提出了一种基于事件关联和数据挖掘的故障诊断技术，以实现各技术之间的优势互补： 采用基于CBR／MBR的分布式事件关联进行故障定位，对大量的告警事件进行时间、空间上的关联，以准确、快速地分离出故障源。 对事件关联过程确定的故障源，采用基于CBR的诊断方法进行的故障原因诊断。该方法将决策树法和近邻检索法等数据挖掘技术相结合，对事例检索过程中必需的特征项相似度进行挖掘。在一定程度上解决了故障诊断中知识获取这一瓶颈问题。 为了深入研究，客观评价新算法，作者对本文提出的技术方案在真实网络环境中进行测试，结果显示该方案使故障诊断具有较高的准确性和实时性，性能表现良好。更多还原

【Abstract】 As one of the five greatest Network Management Function Areas, Fault Management takes charge of the detection, diagnosis and restoration of network fault, the effectiveness and power of it correlate with the availability and reliability of managed network. Early in Fault Management, by assisting operators in analyzing the alarm information and pinpointing the nature and location of faults, network downtime, which can be very costly, can be significantly reduced. Considering it, AI techniques will be imported to realize the fault management automatization.At present, it is general to apply a single Al intelligent technology to fault diagnosis, the actual state cannot meet the whole demands of the exact fault identification and diagnosis. With regard to it, fault diagnosis is divided into two parts in this thesis: fault localization and fault cause diagnosis, different Al techniques can be applied to these two parts. Moreover, a fault diagnosis technology based on event correlation and data mining has been worked out by the writer.A distributed event correlation model based on CBR/MBR is presented in this thesis to implement fault localization. It can reduce the amount of information presented to network operator by filtering out unnecessary or irrelevant events. Simultaneously, the semantic content of the information presented can be increased by the correlation process, hence helping to establish the underlying problem or condition which produced the events.For the second part, a fault diagnosis technology based on CBR has been brought forth. This technology which combines Decision Trees and K Nearest Neighbor, can mine the relative similarity of latent-item in the course of case search. This algorithm will settle the bottleneck to a certain extent.For evaluating the new technology in study, the writer tests algorithms in real network environment. They show good performance of fault diagnosis.更多还原