Research On AIOps Technologies Based On Machine Learning

With the rapid development of global information technology and the increasing scale of the network,the safe,efficient and high-quality operation and maintenance of IT systems has become a hot spot of concern and research in the industry.In recent years,with the development of machine learning technology,artificial intelligence and IT operation and maintenance are combined,and AIOps(Artificial Intelligence for IT Operations)technology is proposed.This thesis mainly researches two major problems under AIOps framework,namely KPI(Key Performance Indicators)anomaly detection and IT operation and maintenance fault analysis.Based on this,AIOps intelligent operation and maintenance system is designed and implemented in this thesis.The specific work of this thesis is as follows:(1)For the research of KPI anomaly detection technologies,this thesis analyzes the characteristics of KPI indicators in AIOps and designs the overall framework of KPI intelligent anomaly detection module.This thesis collects KPI indicators and extracts different features from multiple levels,then adopts SMOTE oversampling method and random undersampling method to balance the data set and uses Z-Score method to normalize the extracted features.This thesis selects BP neural network as KPI anomaly detection training model,and tries different network structures to select the optimal one.To improve the performance of the model,the redundant features are reduced through feature screening experiments.Then the thesis analyzes the defects of BP neural network and improves it from two aspects:optimizing weight initialization and optimizing gradient descent,then analyzes it through experiments.Finally,this thesis proposes the ensembled KPI anomaly detection model,and applies the improved BP neural network to it,further improving the model performance.(2)For the research of IT operation and maintenance fault analysis technologies,this thesis analyzes the characteristics of operation and maintenance faults,and designs the overall framework of fault analysis module,then selects Support Vector Machine to construct the fault analysis model,and conducts experiments to selects the appropriate kernel function type of Support Vector Machine.In this thesis,the grid search method is used to basically determine the parameters of Support Vector Machine,and the heat map is drawn.Based on this,an improved simulated annealing algorithm is proposed,and experiments are carried out to improve the performance of the fault analysis model.(3)This thesis designs and implements AIOps intelligent operation and maintenance system.Firstly,it analyzes the requirements of AIOps intelligent operation and maintenance system.Secondly,the overall architecture,the system function framework and the system database are designed according to the system requirements.Finally,the functional modules of the system are implemented in detail and the implementation results are shown.