| The scale of communication network is becoming more and more complex,and the network equipment is increasing,which brings huge burden to the network.When the burden of devices is overloaded,faults occur and a large number of alarms are received during network control.There is a lot of information about the root cause of the fault and the source of the fault in the alarm data.Further analysis of the alarm data can assist the network fault maintenance work.When a fault occurs,how to analyze the alarm data,generate the root cause of the fault,and locate the fault location is a huge challenge for network operation and maintenance personnel.Through data analysis and correlation research on alarm data,potential information,such as frequent alarms and association rules,can be obtained.These data can help maintenance personnel find the root cause of the fault and locate the fault location.Therefore,alarm data analysis is very important in network management part.By analyzing and researching alarm data,this paper proposes an improved weighting method for alarm correlation analysis and a fault tracing algorithm based on data mining and alarm correlation.The results mainly include the following aspects.(1)By reading a large number of literatures,the research status of data processing,alarm analysis and fault location is summarized and organized.The process of the classical association analysis algorithm is introduced through data examples,and the formula derivation,algorithm improvement and algorithm advantages and disadvantages of the ensemble learning algorithm are introduced respectively,which lays a theoretical foundation for the follow-up research.Analyze the alarm data set to understand the attributes and data characteristics contained in the alarm data,and find that there are dirty,incomplete,and redundant data in the data.To facilitate subsequent data analysis and correlation analysis,the data set is cleaned.(2)Aiming at the problem that the difference of alarm importance is not considered in the previous alarm correlation analysis algorithms,an alarm correlation analysis algorithm with improved weighting method is proposed.An alarm data correlation analysis algorithm with improved weighting method is proposed.First of all,considering that the importance of each alarm data is different,the information contained in the alarm is also different.In order to distinguish the difference,the XGBoost algorithm model is used to train the alarm data,and the weight of the alarm attribute is obtained and the importance weight value of each alarm is obtained.Aiming at the problems existing in the traditional sliding window division transaction,the method of dividing the alarm transaction set by the sliding window is improved,and the alarm data is divided into transactions with reference to the network topology data,and the divided alarm transactions are more authentic and reliable.Then use the weighted FP algorithm to perform association analysis and association rule mining on the transaction data containing the alarm importance weight value to improve the mining efficiency.Finally,the experimental results verify that the alarm correlation analysis algorithm of the improved weighting method has good performance in mining frequent alarms,important association rules and running time.(3)First of all,considering that the importance of each alarm data is different,the information contained in the alarm is also different.In order to distinguish the difference,the XGBoost algorithm model is used to train the alarm data,and the weight of the alarm attribute is obtained and the importance weight value of each alarm is obtained.Firstly,the data mining method is used to construct and filter the features of the alarm data,and the ensemble learning model is used to classify the faults.Then,based on the obtained fault category,combined with the alarm correlation data and the alarm cause table,analyze the root cause of the fault,and locate the fault location.Finally,through comparative experiments,it is verified that the fault source tracing algorithm based on data mining and alarm correlation has good effects in fault classification,root alarm location,root cause analysis of alarms,fault source location and running time. |