The Design And Implementation Of Intelligent Alarm Analysis Aystem Based On AIOps

Nowadays,with the rapid development of the Internet,modern information networked interactive business systems have replaced traditional paper,books,tapes and CD-ROMs that rely on physical media for business communication.The business efficiency of various industries has greatly increased by information networking.With the establishment of related business systems in various industries,and the application and development of a large number of related computer network hardware equipment,it has brought about exponentially increasing operation and maintenance problems and the demand for IT operation and maintenance work in various industries.In recent years,with the development of artificial intelligence,various enterprises have gradually integrated artificial intelligence algorithms,bringing solutions to industry problems for many enterprises.When IT operation and maintenance and artificial intelligence are combined with each other,intelligent operation and maintenance has emerged.(Artificial Intelligence for IT Operations),AIOps.This thesis mainly studies the alarm root cause analysis and KPI anomaly detection under the AIOps framework,and on this basis,designs and implements an intelligent alarm analysis system based on AIOps.The specific work content of this thesis is as follows:1.For the research of KPI(Key Performance Indicators)anomaly detection technology,a trend model and a seasonal model based on the Prophet model are proposed.Perform trend factor decomposition and seasonal factor decomposition on the KPI indicator time series,automatically identify the trend and periodicity,and then estimate the corresponding parameters to obtain the corresponding model.The obtained model is used to predict the real-time index of a time interval,and the corresponding confidence interval is obtained.Then enter the result into the program.The more the actual value exceeds the confidence interval,the lower the score.The health score will reflect the comprehensive situation of each indicator of a device unit in a period of time.For KPI anomaly detection,the trained model is called to predict,and the predicted value and confidence interval are obtained,and compared with the actual value.If the confidence interval is exceeded,it is judged as abnormal.2.Aiming at the research of alarm root cause analysis technology,we proposed a model algorithm based on supervised learning.When the business/physical logic in the network is relatively dispersed,the sub-network division can avoid the interference caused by accidental information between the sub-networks,and can more accurately analyze the alarm propagation law within the sub-network.The basic idea of the algorithm is to convert the alarm root cause location problem into a multi-classification problem,use the root cause information marked in the alarm information that occurred in a continuous period of time to classify the alarm transaction,train the multi-classification model through the continuously accumulated annotation data,and finally analyze The root cause of the output alarm.3.This thesis designs and implements an intelligent alarm analysis system based on AIOps.By adopting a browser/server architecture,and choosing the python open source framework Django combined with front-end design technologies HTML,CSS and JS for platform development.This platform includes data source management,alarm root cause analysis,topology model management,and KPI anomaly detection functions.It provides auxiliary decision-making for the operation and maintenance process from three aspects:data preprocessing,root cause location and anomaly detection.