Research On The Algorithm Of Internet Traffic Identification Based On Deep Learning

This thesis is supported by the state grid science and technology project called Research and Application of Key Technologies of Power Information and Communication Network Flow Prediction and Pipeline Intelligence,whose research direction is the service-oriented traffic identification and sensing.Main research is the study of network traffic classification technology based on deep learning.The classical model of deep learning is the convolutional neural networks,which is applied to the network traffic identification to improve its performance.A feature selection algorithm based on similarity and uniqueness is proposed for the problem of network traffic characteristic attribute.And its performace is compared with gain ratio,Pearson correlation coefficient,symmetrical uncertainty attribute and correlation-based feature selection algorithm.The method of improved convolutional neural networks recognition based on hybrid cross training strategy is proposed for the problem of the non-uniform characteristics of network traffic and verified on the Moore data set.The full text is divided into five chapters,the main contents are:The first chapter introduces the background and research significance of the subject,expounds the research status of the network traffic identification technology,and gives the chapter arrangement of the paper.The second chapter summarizes the basic theory of deep learning,and analyzes the basic principle,parameter initialization,training method and evaluation criterion of the convolutional neural networks of the classical deep learning model.In the third chapter,a feature selection algorithm based on similarity and uniqueness is proposed for the problem of network traffic characteristic attribute and compared with many existing feature selection algorithms.In the fourth chapter,the method of improved convolutional neural networks recognition based on hybrid cross training strategy is proposed for the problem of the non-uniform characteristics of network traffic and verified on the Moore data set.The fifth chapter summarizes the research work of this thesis,and points out the further research direction.