Research On Deep Learning Based Network Traffic Classification Technology And Its Applications

Network traffic classification has been widely accepted as an important means to realize effective network management and ensure network security.With the identification and classification of network traffic,the traffic of different applications can be subdivided,so that network service providers can accurately understand network traffic and provide personalized services for various cunstomers,by which the quality of service can be improved,respectively.The performance of traditional traffic classification methods has been unable to meet the current users' requirements for the quality and efficiency of network services.Therefore,how to identify various network traffic and distinguish different services through effective technical means has become one of the challenges in the field of network operation and maintenance management.This thesis presents a comprehensive research on the network traffic classification technology and its application based on deep learning technology using theoretical analysis and numerical simulation.Firstly,this thesis briefly introduces the principles and major technologies of network traffic classification.Aiming at the problem that traditional network traffic classification methods are difficult to apply to complex network environments,this thesis discusses the machine learning and deep learning methods,and compares the pros and cons between there methods.The advantages and disadvantages of the application of the classic convolutional neural network model in deep learning are used to classify the KDD data set that simulates the real network condition,which verifies the feasibility of deep learning in the problem of traffic classification.Aiming at the degradation problems such as the disappearance of gradients that may occur when the designing model is too deep in the convolutional network,this thesis proposes an improved residual neural network traffic classification algorithm.The improved algorithm can avoid such degradation problems by adding residual modules.Theoretical analysis and simulation experiments show that the improved algorithm does not have degradation problems when classifying the KDD data set and introduces better accuracy at the same time.Aiming at the application problem of traffic classification,this thesis proposes an in-depth research on data center network,analyzes the traffic characteristics and topology in the data center network comprehensively.It finds that by classifying the traffic in the data center network according to the amount of data carried,the routing scheme can be effectively improved and the transmission performance of the data center can be improved simultaneously.On this basis,this thesis proposes an improved algorithm for multi-class random routing optimization.The improved algorithm divides the traffic in the data center network into three or more categories through the principle of multi-class expansion,and then performs random routing according to the size of the traffic load,i.e.the probability of traffic routing optimization is directly linked to the size of the traffic load.Theoretical analysis and simulation experiments show that the improved algorithm can achieve better optimization effects than the two-category routing optimization algorithm in the Fat-Tree topology,and can further reduce the time to complete the stream transmission,respectively.