Research On Data Preprocessing And Fault Diagnosis Methods For Out-of-band Monitoring System

With the development of information technology,human lives and behavior increasingly relies on the network and communication,and the core device to provide service is the server,once the server fails,it will have a huge impact on the service involved,even cause unimaginable loss and disaster.So,in order to ensure the safety and stability of the server,it is very significant to monitor and manage the server comprehensively and deeply.The current server monitoring technology research pays too much attention on the monitoring of process and management,and most of the monitoring technology is built on the operating system.However,the research on the quality of the original monitoring data and the cause of the fault is relatively few.Therefore,data preprocessing and fault diagnosis methods for out-of-band monitoring system are proposed in this thesis to improve the quality of out-of-band monitoring data,and to diagnosis the current server hardware status rapidly and accurately.The thesis firstly introduces the existing type in-band and out-of-band monitoring system,and the advantages and disadvantages of these monitoring systems are analyzed in detail.Then,the framework of out-of-band monitoring system with fault diagnosis function is proposed,and the function modules of the system are introduced in detail.The data preprocessing and hardware fault diagnosis in the framework are mainly studied in the rest of the thesis.In term of the data preprocessing,from the actual monitoring process it can be found that the monitoring data appears data anomalies,data loss and other issues.Also the out-of-band monitoring data has the characteristics of a wide range of monitoring items.To address the problems above in the monitoring data,a variety of preprocessing methods are used to deal with them in the thesis;For the data preprocessing,according to the properties and characteristics of the monitoring items,the attribute set is used to classify it to facilitate the subsequent management and storage;For monitoring data anomalies,the preprocessing method of exponential smoothing and PauTa rule criterion are combined to eliminate abnormal data.Aiming at the deficiency of the original loss and the pretreatment of the monitoring data,Multiple Interpolation method is used to fill the data to ensure data accuracy and completeness.Finally,the data is normalized processing,and provide data support for the establishment of fault diagnosis model.In view of hardware fault diagnosis,firstly hardware failures are divided into three categories.According to the characteristics of multi-dynamic nonlinear data processing and function fitting based on Neural Network,the improved BP neural network is used as a model for fault diagnosis in the thesis,and the monitoring data which has been processed is used for training sample set in the fault diagnosis model.Finally,through the experiment it can be proved that the proposed preprocessing method in the thesis can effectively solve the quality problem of the monitoring data;Confirme the feasibility that the neural network is introduced into the feasibility of monitoring system for hardware fault diagnosis;Through the failure data of a diagnostic test,it is verified that the diagnosis model has a high diagnostic accuracy.