Research And Application Of Alarm Correlation Algorithm Based On Data Mining

With the rapid development of telecommunication networks and the expansion of network scale,people have put forward very high requirements for the stable operation of telecommunication networks.Faced with this situation,telecom network operation and maintenance personnel must accurately find the root cause of the failure and quickly resolve the failure.Although the faults in the network are presented in the form of alarms,due to the correlation between devices and components in the network,a device failure will cause the associated component or device to also fail,which will trigger an alarm storm and increase the difficulty of finding the root cause.In the face of massive alarm data,accurately analyzing and locating the root cause of the alarm is the key to troubleshooting,which is of great significance for improving the stability of network operations.This dissertation will use data mining methods to mine the correlation between alarms to help network operation and maintenance personnel accurately locate the root cause of the alarm and solve the fault.The main research work of this dissertation is as follows:1.If the alarm data generated by the telecommunications network system directly mines the relationship of the alarms without processing,it will consume a lot of time and resources to deal with massive repetitive and unimportant alarms,which will affect the quality of mining.Therefore,the alarm data needs to be pre-processed to filter out the high frequency and short duration flutter alarms in the network,reduce the impact on the alarm correlation analysis,and improve the accuracy of locating the root cause of the fault.2.In a telecommunications network,network equipment or components are related to each other,and when certain components in the network fail,they will propagate along the call chain.Establishing an alarm network topology diagram from the perspective of data analysis to analyze the law of alarm propagation is an important method for alarm correlation analysis and root cause analysis.Therefore,this dissertation will learn the alarm Bayesian network from the alarm data set,and combine it with the Spark computing engine to improve the learning speed,and combine the minimum spanning tree algorithm to extract the backbone path of the network topology to optimize the network topology.3.Sequence pattern mining can find potential relationships between data,and its research significance is huge,and this method is widely used in the field of alarm correlation analysis.Therefore,this dissertation will mine the correlation between alarms based on time series patterns.In view of the fact that the GSP algorithm needs to traverse the sequence set generous times and a great quantity of candidate sequence patterns will be builded,which consumes much time in algorithm,this dissertation makes corresponding optimizations to the GSP algorithm.Based on the dictionary search tree,the improved algorithm can effectively decrease running time,but consumes more memory space.In view of the defect that the Prefix Span algorithm consumes huge memory and may cause memory overflow,this dissertation makes corresponding optimizations to the Prefix Span algorithm.The improved algorithm is based on the data structure of the two-tuple linked list index tree,which can effectively reduce the use of memory space,and the running time is similar to the original Prefix Span algorithm.4.Based on the above theory and experimental results,the alarm traceability AIOps system is designed,and will introduce module design and implement at length.