Font Size: a A A

Multi-domain Distributed Network Alarm Fuzzy Association Rules Mining

Posted on:2016-08-23Degree:MasterType:Thesis
Country:ChinaCandidate:Z LiuFull Text:PDF
GTID:2308330473955093Subject:Communication and Information System
Abstract/Summary:PDF Full Text Request
With the information industry soaringly developing in the modern society, people have been much more dependent on communication networks than ever, both in daily life and at work. Undoubtedly, it is important to guarantee the quality of the network running because the temporarily out of normal functioning of any communication network will bring inconvenience sometimes even economic losses. Therefore, it is necessary to find the root faults and solve the problems in time when the network is broken.When a communication network fails to work, the network nodes will produce a large number of alarms. Identifying the correlation between the alarms is helpful to locate the root fault. How to find the correlation among these massive alarm data is a hot spot. One method is data mining which is able to deal with huge amounts of data. Data mining technology gives most works to computer to deal with so that it can not only improve the efficiency but also save lots of manual labor. In recent years, many researchers have used association rules mining which is one of data mining technology methods in network fault diagnosis. However, the corresponding relationship between alarms and root faults is not one-to-one, but has strong fuzziness. The traditional association rules mining does not take the fuzziness into account so that the accuracy of fault diagnosis is not very high. Additionally, large communication network is usually divided into different domains, namely it has distributed structure. It is not enough to just mine alarm association rules in each subdomain. It is necessary to mine the alarm association rules among domains.According to the fuzziness between alarms and root faults, and under the background of multi-domain distributed network, this paper aims at mining fuzzy association rules of communication network by combining the fuzzy theory with mining association rules in data mining. Specific research points and innovations are summarized as follows:1. Because the characteristics of the original network alarms do not suit to association rules mining, information field extraction method was chosen for unifying information model of alarm in pretreatment stage. Then the paper set up alarm transaction database by using time-window and sliding-step mechanism and established fuzzy alarm transaction database after quantifying each property and fuzzing the alarms by FCM. In the paper, initialization of FCM algorithm based on the point density has been improved. There are slightly differences between synchronization scheme of establishment of transaction database and alarm fuzzing process of global-local distributed system model and that of local-local distributed system model.2. For two classic fuzzy association rules mining algorithms, Apriori algorithm and Fuzzy FP-tree algorithm, have shortages, which is that Apriori algorithm has low time efficiency and Fuzzy FP-tree algorithm is insufficient in aspect of processing the Fuzzy attributes, the paper put forward a fuzzy association rules mining algorithm named LLB-FARM based on linear list which has better time efficiency than Apriori algorithm and better ability of processing alarm fuzzy attributes than Fuzzy FP-tree algorithm.3. For two different kinds of distributed system models, the paper proposed two different distributed mining algorithms, global-local algorithm and local-local algorithm. And two schemes in local-local algorithm, data transmission efficiency priority scheme DTEPS and station average mining time efficiency priority scheme SAMTEPS were proposed to suit different scenes. Finally, the paper has verified that the proposed algorithms do have good scalability of time and space and can mine fuzzy association rules of multi-domain distributed network alarms. And those rules provide the alarm correlation data for fuzzy reasoning module of the network fault diagnosis system.
Keywords/Search Tags:network fault diagnosis, network alarms, data mining, multi-domain distributed network, fuzzy association rules
PDF Full Text Request
Related items