Research And Application Of Correlation Analysis In Telecommunication Network Alarms Management

Communication network produces daily a lot of alarms. Large numbers of unrelated alarms flood the key alarms that reflect the root network failure, bringing about a great challenge to alarm management and fault-locating. Network alarms databases stored a lot of history alarms, which hides useful rules on which the network operates. Data mining is a kind of knowledge discovery technology that can be used in large scale of data to identify potential and valuable knowledge. Association rules mining methods that based on data mining technology can be used to find association rules between the alarms, which can effectively be used in alarm filtering and compression, highlighting the key alarms, thus contributing to alarm management and precise fault-locating.The paper will be presented as follow:Firstly, the communication network faults and the up to date alarm management research status were introduced, as well as the differences and connections between fault and alarm were clarified. The complete alarms life cycle, ranging from the production, acquisition, to normalization, filtering, and finally displaying and storage, was totally introduced. After that, the characteristics of alarms were detailedly analyzed and summarized. The complete introduction to the knowledge of alarms makes a good preparation for the following alarms association rules mining.Then, the paper focused on research on the alarm association rules mining algorithms. The time and space performances of the traditional association rule mining algorithms need to be further improved when dealing with massive data mining. The disadvantages of the traditional mining algorithms were reasonably improved to fit in the massive alarms data mining. As to the problem of low traversal efficiency when searching the FP-tree for conditional pattern bases, a new No-Header-Table FP-Growth (NHTFPG) algorithm was proposed, avoiding traversing the same FP-tree path multiple times. Theoretical analysis and actual mining results showed that NHTFPG effectively improve the mining speed and can meet the rules mining needs.When mining frequent patterns, a simplified representation form of frequent patterns was used, which reduced the storage overhead, meanwhile, avoided the recursive mining waste of time. Alarms were sequential data with strict temporal order, so WINEPI algorithm was used for mining sequential patterns rules. During the association rules generation, the concept of cross-support patterns was employed, and the rules were represented in a simplified form that effectively reduced the number of association rules.The mining data used in this paper was from real alarms database, which guarantees the integrity and validity of the mining results. This study directly intends to application, the mining methods used and results generated can be directly, or make minor modifications, applied to the actual alarm management system, having some theoretical and practical value for reference.