Research On Traffic Classification Technology In Intelligent Network Management

Traffic classification is the first step of network Quality of Service(Qo S)control mechanism and traffic anomaly detection,and it is also an important research branch in congestion control and network security.In complex network analysis,the traffic classification scheme based on machine learning often needs to obtain a large number of correctly labeled samples,which requires a large number of manual work.Therefore,traffic classification with higher accuracy at low identification rate is still a thorny issue.Secondly,the automatic driving scene needs millisecond real-time response time,which leads to the research of high-speed and low-delay transmission network.The second research content of this paper involves a transmission network of software-defined grid topology,whose functions include centralized control,software-based traffic analysis and dynamic update and forwarding rules.We specifically studied intelligent routing and related traffic classification methods.The main work of this paper is as follows:(1)The realization of intelligent network traffic classification is studied.In this paper,two network representation methods based on graph convolution network(GCN)are tried to avoid the problem of a large number of labeled samples.The research method is convenient to combine representation learning based on graph structure network with network traffic problem,and obtain certain classification accuracy of network traffic with relatively few labeled samples.Preliminary experimental results show that the proposed method achieves good performance in small network scale,and the classification accuracy reaches 97.35%.(2)The routing arrangement,traffic distribution and traffic classification in grid topology and transmission network are studied.The transmission network design is an example of joint routing prediction and forwarding,in which the input data in the training phase comes from a routing scheme of the shortest path,and we use offline training to obtain relevant models.Secondly,using the current network state data(such as service request value and queue length)and the trained model,the predictor will output the vector of routing sequence.The main challenge here is to train a suitable model from the limited training sample data set,which will dynamically update forwarding rules according to current and historical network data.By introducing different traffic classification models,such as logistic regression and parallel neural network,this example realizes an intelligent route arrangement.Experimental results under different traffic modes verify the feasibility of the design.