Research On Technologies Of LAN Network Fault Diagnosis Based On Convolutional Neural Network

With the development of Ethernet,the complexity of Local Area Network（LAN）is also increasing with the number of users and network terminals.In order to ensure the health and stability of LAN,network administrators need to control the whole network running state,analyze the causes of network faults and make empirical diagnosis.However,LAN often carries special services,traditional methods are very time-consuming and labor-intensive,so at this stage of the fault diagnosis response and the degree of intelligence put forward new requirements.Therefore,from the perspective of deep learning,this paper studies the network fault diagnosis method based on convolution neural network model.The main work of this thesis is as follows:First of all,the paper introduces the research status of network fault diagnosis,and the two core problems in the process of fault diagnosis:network fault information collection and analysis,network fault problem discovery and detection.Finally,the main contents of this paper are given.In the second part of the paper,the fault cause theory of LAN fault is analyzed,and the fault diagnosis techniques of expert system,graph search model,support vector machine and software defined network（SDN）are explained in detail.In the third part,based on the KDD99^[48]dataset,the convolution neural network is studied,and the data conversion operation of gray matrix is proposed.The structure of convolution neural network is designed according to the scale of data feature,and a series of optimization studies including dropout learning,gradient optimization algorithm and data enhancement are carried out.In the fourth part of this paper,the advantages and disadvantages of the existing data acquisition methods are analyzed,and this thesis propose a new passive distributed data capture method without new traffic and network structure.At the same time,a dataset construction method based on feature engineering is proposed,and a complete data dictionary with hierarchical structure is put forward.Finally,this thesis designs and simulates the real LAN fault scene,where this thesis completes data capture,data set construction.And this thesis model verification on KDD99^[48]data set and self-built data set.By testing different activation functions,learning rates and data enhancement of the model,the generalization ability of the model is improved.The final results show that the model is trained on KDD99^[48]dataset with an accuracy rate of 96.8%and that on self-built dataset with an accuracy rate of 88.3%.Comprehensive experiments show that the fault diagnosis based on convolution neural network has a good diagnosis efficiency and has the potential to land completely.