Research On Key Technologies Of Network Traffic Identification And Prediction System

With the continuous development of the Internet,the Internet has developed into a massive nonlinear system,and the network is increasingly diversified.Therefore,people put forward new requirements for the operation and maintenance of the network,which are mainly reflected in network traffic identification and prediction,abnormal traffic monitoring and control.Traffic identification algorithm and network traffic model play an important role in network design,quality of service,network management and monitoring.At present,operators have to face a large number of "resource occupancy,low value profit",such as the P2 P downloading traffic is constantly squeezing up the network bandwidth,and the quality of service and network security is difficult to solve.The network traffic model reflects the real network traffic by establishing a mathematical model.The traditional Poisson model has failed in the modern communication network.Since Leland and others first explicitly proposed the self similar phenomenon in the network traffic in the early 90 s,the self similar model of network traffic is constantly emerging.At present,there are mainly port based flow recognition methods,methods for accurate identification of variable port services,identification methods based on traffic statistics characteristics and cross layer services.Because more and more multimedia applications are introducing secure encryption technology and P2 P computing technology,and multimedia applications are constantly changing versions,so the original traffic identification method is no longer applicable,and more and more difficult to accurately and efficiently identify the multimedia information flow.For this reason,in the face of known,unknown,encrypted and unencrypted multimedia information flows,recognition technology is becoming more and more integrated,intelligent,automated,efficient,fast and accurate.This paper first introduces the network business recognition model based on multi recognition engine,the DPI business analysis system based on machine learning and the business recognition method based on the decision tree model.Then,starting from the self similarity of the network,this paper introduces the 1R/S statistics,periodic graph method and wavelet analysis in self similarity,and studies and validates the network flow.The volume time series model focuses on the fitting and prediction of network traffic by using the FARIMA model which can reflect both the long-term correlation and the short correlation.Wavelet transform is a branch of applied mathematics developed in the late 1980 s.It can decompose the complex network traffic into approximately unrelated time series and smooth the time series.It has multi-scale features and provides a method to observe signals from coarse and thin.The wavelet model has a unique advantage in describing network traffic,that is,the sequence after wavelet transform has better stability.Both the independent wavelet model and the multifractal wavelet model can generate self similar traffic.The latter is more suitable for describing the actual traffic because of the non negativity of the signal.The core of wavelet transform is multi-scale analysis,which studies information at different levels.The independent wavelet model can generate self similar network traffic,but because it generates a large number of negative signals,which is inconsistent with the actual situation.One of the important functions of network traffic modeling is to predict traffic.Network traffic has high predictability.This paper introduces the flow prediction problem into the wavelet domain,combines the wavelet analysis method and the time series model to predict the flow rate,and improves the method.First,the approximate part and the detail part of the wavelet decomposition are reconstructed,and then the FARIMA model is used to predict the error.