The Research And Application Of Computer Network Fault Diagnosis Based On Machine Learning

The computer networks,as one of the most indispensable infrastructure in the world of today,play an extremely significant role both in industry and in our daily lives.Meanwhile,the scale of the computer networks is becoming larger and larger,and its complexity is also getting higher and higher,as a result,the ISPs(Internet Service Providers)are facing severe challenges to the effective network management.Fault diagnosis is one of the essential and difficult tasks in network management,If the faults existing in the network cannot be diagnosed and repaired quickly,it will not only increases the operating costs of ISPs,but also reduces the quality of service.Thus,this thesis researches on adopting machine learning methods to network intelligent fault diagnosis and apply it to IPTV fault diagnosis system.The main work and contributions are as follows:(1)Studied several methods of network fault diagnosis.In the face of today's complex network environment,the existing network fault diagnosis methods have a disadvantage that the diagnostic knowledge is difficult to obtain or update on different levels,but the network fault diagnosis technique based on machine learning can automatically learn the knowledge of fault diagnosis from the complex network environment,and uses the learned diagnostic knowledge to quickly detect and locate faults.(2)Studied the current related work of network diagnosis technique based on machine learning,and introduced the theory and mathematical derivation of four machine learning algorithms including the Centroid Based Classifier(CBC),Logistic Regression,Support Vector Machine(SVM)and Artificial Neural Network(ANN).(3)Proposed a new centroid based model named Gravitational Centroid Model(GCM)to solve the CBC's model misfit problem on highly skewed data.The experimental results show that the GCM's performance measure of micro-averaging 1 and macro-averaging 1 consistently outperform the CBC model.(4)Analyzed and compared the performance of CBC,GCM,Logistic Regression,SVM and ANN on IPTV fault detection dataset and fault location dataset.The experimental results show that the GCM is more suitable for IPTV fault detection scene,while the SVM and ANN is appropriate for IPTV fault location scene.(5)Designed and developed a prototype software named the IPTV intelligent fault diagnosis system.The system adopts Gravitational Centroid Model and Support Vector Machine by default for fault detection and fault location respectively,and can visually provide the fault diagnosis results to the relevant personnel.