| In modern information age, the rapid development of electronic informationindustry promotes the growing of the communication network. The strong desire ofpeople for communication anytime, anywhere is increasing. Therefore, the dependenceon communication network covers the whole life in respective corner accordingly.When network fails, it will generate varities of loss ranging from affecting normalbusiness needs to economic losses. Fault Management as the first function formanagement of communication network, its importance is increasingly prominent.It’s urgent for fault loacation and recovery when it occurs. The management sitecarries out fault location basing on a kind of report event collected in real time, which iscalled alarm. However, node failure will also affect those which directly adjacent to it orhaving a communication demand relationship with it. So that, these devices generatealarms accompanied. Only by the way of alarm correlation analysis will we get the rootone in order to compress the redundant alarms.Fault management for multi-domain distributed network is my article’s researchcenter. According to this theme, the fuzzy theory with association rules data miningtechniques are combined together and applied into real time fault diagnosis. Specificresearch points and innovations are summarized as follows:1. For the characteristics of the original alarms do not suit to association rulesmining, this artical studied information field extraction method used for unifinginformation model of alarm. One the one hand, we discussed the method forestablishing alarm transaction database basing on original alarm sequence. On the otherhand, depending on the characteristics of distributed multi-domain networkmanagement, we proposed a strategy for establishing alarm transaction database amongglobal management site and the local ones synchronously.2. Based on the degree of approximate between one alarm and the root one, fuzzyclustering method had been applied for the alarm fuzzy process. After fuzzy process foralarms, it makes network administrator easier to understand. Alarm correlation analysis,fault fuzzy inference, and real-time network fault diagnosis become more scientific and reasonable based on it.3. Usually in a large-scale communication system, the network is divided intodifferent management domains, each of the management domain can contain severalsubnets. Characteristic of large communication network is multi-domain management,so that network management model must be distributed, i.e., the global management siteand local management ones satisfy a distributed architecture. In this artical, a newalgorithm used in multi-domain distributed network for alarm fuzzy association ruleparallel mining was created, which was called PFAARM (parallel fuzzy alarmassociation rules mining algorithm). Fuzzy association rules can be achieved withininner-and inter-domain alarms. On the basis of alarm correlation analysis in singlemanagement domain, the introduction of inter-domain alarm fuzzy association rules,which are based on inter-domain communication relationship, gives another essentialclue for fault location. Meanwhile, it’s of great significance for quick and efficient faultlocation.4. Real-time fault diagnosis has been a problem in the field of fault diagnosis. Thereason is that all the mining algorithms execute alarm association analysis depending onhistorical alarm database. But, the establishment of transanction database relies on theaccumulation of alarm in time series, which inevitably delays the time of the faultdiagnosis. If the network topology has not large-scale frequent changes, the type andfeature of network fault have a large number of identical characteristics. Inspired by thissituation, we made full advantage of the historical association rule mining experienceand conclusions in the process, the fuzzy clustering and fuzzy matching method hadbeen used in the real-time network fault diagnosis. It provided a new idea for rapiddiagnosis of the fault. |
PDF Full Text Request