Research Of Alarm Correlation Analysis In Communication Network Based On Distributed Mining

Communication network is a large loosely coupled system,which consists of various transmission equipments,switching equipments and terminal equipments.It's effective operation depends on the network management system."NE-level network management system" manages the network devices directly,while "Network-level network management system" manages the entire network.The Network-level NMS(Network Management System)collects data from the underlying networks and analysis it.With the development of NMS,NE-level NMS may also support functions such as topology management Currently,as the Network-level NMS does.Then,a hierarchical,distributed NMS architecture appears.Research on the distributed network management strategies can help us save a lot of communication costs and reduce the waste of computing resources.Network alarm data are messages which are issued by the managed objects when the communication network's operation status changes,or a fault occurs,Thus,the fault management module is one of the most important components of NMS.Using scientific methods to analysis these alarm data can help administrators diagnose and troubleshoot problems quickly,by finding the main cause.Research of the correlation analysis on distributed data mining is the main work of this article :First,by describing the network management and alarm data basics,The paper identifies the possibility of distributed alarm correlation analysis.And then the paper chooses a appropriate solution and system architecture to realize it by comparison,which is: using the distributed association rule mining,rather than the clustering method for distributed alarm correlation analysis.Secondly,the paper analyzes current algorithms for the extraction of alarm's transaction data,and proposes a improved algorithm.This Mean-shift based algorithm is well adapted to the uneven character of the alarm data,solving the redundancy problem of the transaction data.Then,by studying the existing frequent pattern mining algorithms,a new improved algorithm is proposed,which is well suited for the distributed systems.Compared with the previous approaches,it reduces more communication overhead.Lastly,due to the large computing capacity and low efficiency of the existing rule generation algorithm,the study of pruning strategy in this paper greatly lowered the number of candidate rule antecedent,and improves the efficiency of the algorithm greatly.The credibility of the rules of measurement are also researched.