Network Alarm Association Rule Mining System Research And Design

Fault diagnosis and localization is the vital core of the network management. When the faults take place in the networks, it is necessary to find the locations and the causations of the faults in time in order to get rid of the faults and recover the networks'function rapidly. The alarm correlation analysis, an important approach of fault diagnosis, plays a crucial role in network fault management. Data mining provides a new approach of the knowledge updating during the alarm correlation analyzing. In modern networks, some of the alarm's attributes has different levels, and services with different QoS requests need different treatment of alarms. With the background of the project, the alarm correlation in communication networks based on data mining, that is supported by National Natural Science Foundation of China, this thesis focused on mining weighted alarm association rules in networks, including the alarm's pretreatment, mining weighted association rules, the rules'post-treatment and the simulation and validation of the system of mining weighted alarm association rules in networks.The alarm's pretreatment was carried out by expert system. The problem of alarm's synchronization was settled by setting time window and slip length. The alarms in the same time windows were regarded as an alarm transaction. The attributes of the alarm that reflect the faults were picked out to form an item of an alarm transaction. The redundant alarms were got rid of by alarm compressing. The analytic hierarchy process was applied to deciding the weight of the alarm. The simulation result indicates that the weight decided by this method can reflect the users'concentration on different alarms and the dynamic change of the networks.Based on the existent algorithms of mining weighted association rules, aimed at the networks'characters that are having amount of alarms that can burst out randomly, a pattern tree based algorithm-WFPTA of mining weighted alarm association rules was constructed. The big advantage of the WFPTA algorithm is it is unnecessary to traverse the database repeatedly and to construct the conditional pattern tree recursively. The performance test of the algorithm indicates that compared with MINWAL (O), this algorithm is improved in both the executive time and the used memory. Based on the concept of redundant rules and structural rule cover, a quantificational algorithm of rule post-treatment was constructed, which can delete the whole redundant and repeated rule effectively by the minimum confidence increment threshold provided by the user.The simulation and validation of the system of mining weighted alarm association rules in networks indicates that it can find out effective alarm rules which reflect the faults in networks rapidly. It is also validated that the system which provides the user the compact alarm rules with complete information is valuable and useful to the alarm correlation analysis and fault diagnosis and localization in networks.