Research On Communication Alarm Correlation Model Based On Hierarchical Attribute Similarity

At present, the network size is increasing and the network structure is becoming more and more complex. In order to ensure the robustness of the network, the integrated management of network management system is more and more important. The network alarm is kind of real-time information reported to the network management system when abnormal behavior occurs in the network element. The alarm expresses the status of the network directly. In general, the abnormal network behaviors refer to the network faults. By analysing the alarms, the administrators can locate the network faults, then take appropriate measures. Alarm correlation is an important technicque to analyse massive alarm data. This paper established a hierarchical attribute similarity clustering model.In general,it is thought that alarms triggered by the same fault are similar, thus the original alarms can be classified into meta alarms. The meta alarm includes more semantic information about the fault than one single alarm. The fault information expressed by meta alarms can help administrators improve analysing efficiency and reduce the workload.This papper analysed the reserch status of alarm correlation, and classified the alarm correlation techniques from different angle of view. The advantages and disadvantages of various alarm correlation techniques were compared. It was found out that, so far there was not an universal alarm correlation method applicable to all alarm correlation cases. The importance of alarm correlation in the network management system was analyzed. The main content of alarm correlation was summarized. Associated with the characteristics of an actual power communication network ASON alarms, a communication alarm correlation model based on hierarchical attribute similarity was established. Compared with the traditional clustering method, the novel method introduced the conception of "super attribute". The novel method defined the hierarchical attribute similarity function, which improved the ability of expression for fault information. History alarm database complexity、clustering expansion rate and meta alarm support count were put forward to evaluate the clustering result without expert knowledge. Finally, the traditional alarm clustering method and the novel method were both applied to the actual power system communication network ASON alarm correlation. The results of the two methods were compared and analyzed. It could be concluded that, by comparing the results, in the premise of satisfying the certain project execution efficiency requirement, the novel method had better clustering result for complex fault conditions.The novel method proposed in this paper was proved to have certain practical value in engineering.