The Research Of SDN Traffic Classification Based On Deep Learning

With the rapid development of the Internet,the number and scale of network applications have grown exponentially.While bringing convenience to users,they are usually followed by huge traffic.Accurate and effective identification of network applications will contribute to network security control and resource management.As a next-generation network architecture,Software Defined Network(SDN)currently only supports network layer 2/3/4-based policy enhancements,but lacks of application awareness at higher layers of the network.If application identification can be implemented in SDN to obtain application layer information of traffic,network services can be provided more intelligently.In summary,this thesis will study the application method of SDN network traffic application based on deep learning.Firstly,an SDN application recognition framework based on deep learning is proposed,and the operation mechanism of the framework is fully researched and explored to ensure the stable operation of the framework.The function and function of each module involved in the framework are studied.Each module cooperates with each other to complete the application identification function,and the recognition result is used in upper application management or lower layer routing planning.Then,for the application identification module in the framework,a deep learning model is proposed to identify the application of SDN traffic.Based on Theano deep learning framework and Moore dataset with explicit application markup information,a deep learning model of SdA-LSSVM(stack denoising autoencoder-least squares support vector machine)with high application recognition accuracy is constructed.Ten application types such as WWW,MAIL and FTP_DATA are identified,and the recognition accuracy is up to 91.07%.Finally,for the problem that LSSVM is easy to fall into the local optimal solution,we use particle swarm optimization(PSO)to optimize the parameters of the deep learning model.The experimental results show that the accuracy of recognition based on the improved deep learning model based on PSO algorithm is improved.