Communications Network Alarm Correlation Rules Mining Method

The characteristics of communication networks are great dimension and complex structure. It is more and more difficult to manage such communication networks, especially in alarm processing. Technicians are demanded to rapidly and accurately estimate the source reason bringing the alarms, and then quickly take action to do with it. There are two mainly shortages if the alarms are analyzed one by one by hand. On the one hand, technicians should have profound knowledge, long term maintenance experience, and be quite clear on the structure and devices of the communication network. Seldom people are competent for such extensive network. On the other hand, it is difficult to satisfy the demands of high reliability and high efficiency on such communication network by manual processing.In alarm correlation analysis, relative alarms are united and translated into a source alarm. In this way, an alarm can be sent to network monitor terminal instead of lots of relative alarms. Network administrator can quickly locate faults and filter alarms using alarm correlation rules and so technicians can concentrate on alarms filtered and do it rapidly.This project researches following contents. Firstly, Alarms from network monitors are analyzed by means of data mining. Secondly, some shortages for current data mining implements are analyzed, and then a data pre-process method is proposed and a rule compare module is developed. Raw data is pre-processed using time window concept, and the problem of network time losing synchronous is efficiently solved. Fourthly, some aspects of alarm such as alarm time, alarm clear time and so on, are compared, and then useful alarm rules are gotten. Furthermore the veracity and reliability of rules is improved. Lastly, some experiments are done.