Multilayer Network Alarm Correlation Analysis Based On Fuzzy Association Rule Mining

With the development of science and technology and the improvement of people’s living standard, the communication network has greatly developed, and, people’s life and work has gradually transformed into a pattern that more and more dependent on the network. So when there is a fault in the network, the impact on people’s life and work will be stronger. It is a great challenge now facing the network workers that ensuring the steady of network and fast accurate recovery of the network fault.A function of network cannot be completed by an isolated network equipment, but need for communication between a series of equipment to complete, this result a network fault propagation phenomena. How to pick out the alarm of the root fault in a large number of network alarms is the key of network fault diagnosis, to locate the root alarm will need to find the correlation between network alarms. Because of the advantages of data mining in the processing of large amounts of data make people turn to the association rules mining in network alarm. And because of the fuzzy relation between the network fault and network alarm, fuzzy theory combined with data mining technology is applied to analysis the correlation between network alarm.The study found that, network alarm has its unique characteristics, only if fuzzy association rules mining in network alarm adapt to the characteristics of network alarm can we effectively mining out the alarm correlation. First of all, the association rule mining is a mining algorithm for the transaction database, to mining association rules in the alarm database, we must transfer the database to a transaction database or we need to improve the algorithm. Secondly, network alarm also has distinctive characteristics in distribution, alarm of different network types and alarm of different layer of multilayer network have different distribution characteristics. Furthermore, the network alarm propagation has one to many or many to one of the characteristics in different network layers, and associated with the usage of business.In view of the above problems, deep research has being done in this paper. This paper has established a network alarm model, proposed the method of the alarm fuzzification, established fuzzy alarm model. This paper presents a mining algorithm based on dynamic time window that can mining fuzzy association rules directly in alarm database, it greatly improved the accuracy of association rule confidence. This paper presents a design method of fuzzy minimum support, the algorithm can well adapt to the distribution of network alarm. This paper puts forward a new multilayer fuzzy association rules mining algorithm that to mine multilayer association rules between layers of multilayer networks alarms. This paper proposed an improvement to the generating method of alarm correlation rules, which can greatly improve the rule generation speed and reduce the redundancy of the rule database.