Research On Data Mining And Traceability Algorithm Of Communication Network Faults

With the development of network technology,the network scale is expanding and becoming more and more complex.The in-depth analysis and mining of network alarm is the key task at present.The communication network management technology based on big data drive is the main development trend of network alarm in the future.The main purpose of future communication network development is to ensure the efficient operation of communication services and the high quality of service operation and maintenance of communication network.According to the current research status of network management at home and abroad,the research steps of network management mainly include four steps,namely alarm data collection,alarm data preprocessing,alarm correlation analysis and fault management.For large-scale network alarm data,the research on traceability algorithm in fault management is the main problem.The research contents are as follows:(1)This paper first gives a general description of the steps of network management,then analyzes and compares the relevant research methods and solutions proposed by the majority of researchers at each stage according to the actual network data and problems,and focuses on the key technology of data mining applied to network alarm analysis.Next,it briefly introduces the architecture of the communication network,the collection process of alarm data,then analyzes the data characteristics of the collected original alarm data in many aspects,and carries out one-to-one data preprocessing steps according to the characteristics of the alarm data.Finally,the data preprocessing process and experimental results are given.(2)Aiming at the problem of locating the source of network anomaly in communication network,a method to determine the root alarm location of communication network based on fault tree is designed.Firstly,the clustering algorithm is used to adaptively generate the alarm transaction set,which avoids the randomness caused by artificially setting the step size and window when using the sliding window;Aiming at the problem that the amount of computation of DBSCAN clustering algorithm increases with the increase of data set,KD＿Tree was established to reduce the running time.Then sequence pattern mining algorithm is used to find the strong association rules,and set the threshold for the frequency of the rules to obtain the high frequency sequence patterns.Finally,a fault tree algorithm is designed to determine the root alarm by calculating the influence and priority of each type of alarm.Through the experiments on the alarm data,it can be seen that this method saves time in the transaction set generation stage and improves the correlation between alarm events.In the association rule algorithm stage,the amount of alarms is reduced,and the establishment of fault tree effectively excavates high influence alarms.(3)Aiming at the problem of analyzing the causes of network anomalies in communication network,a method of analyzing the characteristics of alarm data based on decision tree algorithm is proposed.Since communication network equipment generates thousands of alarm data every day and most of them are unlabeled data,adding a label to it is the first step.The algorithm firstly analyzes the alarm data based on regional and time characteristics to obtain root cause information,and then requires multi-dimensional alarm data characteristics.Next,the decision tree method is used to analyze the importance of various features and their impact on the accuracy.Finally,the multi-dimensional analysis results of the data characteristics of real-world network alarm data show that the research method has great advantages in accuracy and execution efficiency,and to a certain extent helps network managers to identify the main alarm data characteristics,which can be based on the data feature first performs network recovery.