Research On Intelligent Analysis Technology Of Optical Network Alarm Based On Machine Learning

With the expansion of the scale of optical transmission network(OTN),the number of optical links and OTN devices is increasing.Once the network fails,it will often cause a series of devices to produce many alarms.It is difficult to find useful information from a large number of alarms in a short time by traditional artificial methods,to correctly determine the location of the fault and to repair it in time.Machine learning is an emerging intelligent technology in recent years.It plays an increasingly important role in optical communication.It has important practical value to analyze and process optical network alarms.In this paper,some problems existing in the original alarm of optical networks(i.e.,information redundant,time asynchronous,ambiguous importance of alarm attributes,etc)are studied in depth.An OTN alarm pre-processing method based on time series and machine learning is proposed.The association rules between alarms are analyzed,the existing association rules mining methods are improved,and the quantization based on machine learning is proposed and demonstrated.A weighted association rule mining method for scoring.This method can automatically compress a large number of alarms,and then analyze the compressed alarms.It can help network managers to find out the source of the fault in time,eliminate the fault and ensure the normal operation of the network.The main innovations of this paper are as follows:First,in the process of alarm pre-processing,this paper uses time series segmentation and sliding time window combination method to extract alarm transactions.In this method,the original alarm is divided into several alarm sequences,so as to realize the initial extraction of alarm transactions.At the same time,through sliding time window,alarm synchronization and redundancy elimination are carried out in each time period of the alarm sequence.Second,aiming at the problem that there are many alarm attributes but their importance is not clear,this paper quantitatively evaluates the importance of alarm attributes by machine learning algorithm,and obtains the importance weight of each alarm attribute.We call the aforementioned alarm transaction extraction method and the alarm importance assessment method together as the alarm pre-processing method,which can deal with the original alarm more effectively and objectively,and is conducive to the subsequent further alarm compression and analysis.Third,in the process of alarm correlation analysis,we improve the classical rule mining algorithm Apriori algorithm to weighted Apriori algorithm,analyze the association between alarm transaction sets,find out the association rules between alarms,realize the further compression of alarms,and help to locate faults quickly.