Research On Network Traffic Classification Based On Deep Learning

In recent years,the rapid growth of the Internet has accelerated the generation of network traffic.According to the Cisco Visual Web Index,2016 to 2021,the global IP traffic is expected to grow three times,average value will be increased from 1.2ZB to 3.3ZB,which will occupy a huge network bandwidth,and will make the problem of network congestion more serious.How to analyze and use this huge network traffic effectively,will directly affect the control of network security and network resource management.Network traffic classification technology is an important technology for analyzing and utilizing network traffic.It is of great significance to the analysis of network traffic and the management of network systems.This thesis focuses on the classification of network traffic,mainly about network application classification and network user behavior analysis.By identification and analysis of network applications,we will understand the development needs of network applications for hardware devices,understand the potential relationships between network applications and users,and improve the service quality of network applications.According to the analysis of network user behavior,we will understand personal characteristics or group characteristics of network users such as Internet habits and preferences,which provide the basis for the network security policy updates.Aiming at the deficiency of the existing network traffic classification technology,this thesis proposes to apply Deep Learning to the classification of network applications,so as to improve the classification accuracy of network applications.Deep Belief Network(DBN)is used as a deep learning research model for network application classification.Using UDP public network application data as the research object.The design and implementation of constructing DBN model are described in detail.Compared with the traditional BP neural network model,this DBN model can effectively improve the classification accuracy of network applications.This thesis also proposed a DBN-based network user identification model.The data set used in this model is the real traffic data of network users in a college.The entire research process includes the model construction,initialization,training optimization and performance testing.The final result shows that the model achieves high accuracy of network user identification.