Design And Implementation Of Alarm Correlation Analysis Module Based On Parameter Self-adaptation

With the increasing number and types of services carried on the network,the devices in the network are also developing towards diversification and location decentralization.This situation not only improves the complexity of the network,but also increases the difficulty of fault management.Therefore,alarm correlation analysis as one of the important means of fault management has been widely concerned.Its main purpose is to find the potential correlation between the alarm data by compressing,filtering and analyzing,and infer the alarm indicating the root cause of the fault from a group of alarm sequences through these correlation information.At present,the research on alarm correlation analysis has made great progress.Common alarm correlation analysis methods include case-based,neural network-based and rule reasoning-based.Among them,the rule reasoning-based method is widely concerned because it can better adapt to the changes of the network,is good at discovering the potential rules between alarm data and has high accuracy.However,the traditional rule reasoning-based alarm correlation analysis methods usually have the following two problems:1)In the process of mining alarm association rules,we need to set the minimum support and other parameters to obtain the qualified association rules.In the existing methods,the minimum support is usually based on expert experience,which is manually specified before the algorithm runs,and it does not change during the execution of the algorithm.However,alarms have the characteristics of sudden,different alarm data items are unevenly distributed and the frequency is different.If we ignore the characteristics of the alarm data itself and always use the same minimum support threshold in the process of rule mining,the quality of mining association rules will be affected.2)In the process of root cause alarm reasoning,a matching network containing rule nodes is usually established.However,because the matching process has the characteristics of combination,a series of intermediate results of partial matching will be cached in each rule reasoning process.For the relatively complex and large amount of data such as alarms,the intermediate results of partial matching will occupy more space and reduce the utilization of cache space.Therefore,this paper proposes an alarm correlation analysis algorithm based on parameter self-adaption.Firstly,aiming at the problem of low efficiency of extracting alarm transaction by using fixed size time window,this paper proposes an adaptive method of extracting alarm transaction based on alann flow rate,so that the window size can be adaptively adjusted according to the alarm data flow rate,so as to realize the dynamic extraction of alarm transaction.Secondly,aiming at the problem of solidifying support parameter in traditional alarm association rule mining algorithm,this paper introduces reinforcement learning method to dynamically adjust the support parameter in association rule mining.It can change support parameter dynamically according to the alarm transaction in the mining process,so as to improve the efficiency of association rule mining.Finally,aiming at the problem that partial matching results of existing alarm rule reasoning methods occupy a large cache,this paper proposes an alarm rule reasoning method based on cache optimization.This method improves the efficiency of alarm rule reasoning by introducing the Heuristic Annotated-Linkage matching algorithm(HAL)to establish a global pseudo binary network,and solves the problem of large cache space in traditional HAL algorithm by recycling partial matching results.Simulation results show that the algorithm proposed in this paper can effectively improve the quality of alarm association rules and reduce the cache pressure in the reasoning process of alarm rules.In order to meet the user's requirements of fault operation and maintenance,such as alarm data query,preprocessing,alarm transaction eatraction,association rule mining and alarm rule reasoning,this paper designs and implements an alarm correlation analysis module based on parameter self-adaptation based on Java language and SpringBoot framework.This paper describes the requirement analysis and the outline design of the module in the development process,and introduces the processing process of the alarm data query,data preprocessing,transaction extracton,alarm association rule mining and rule reasoning in detail.By testing the function and performance of the module designed in this paper,it is verified that the alarm correlation analysis module designed in this paper can effectively meet the needs of users in the process of fault operation and maintenance.The module can effectively support the subsequent fault location and diagnosis tasks,and improve the efficiency of alarm correlation analysis.