The Characteristics Analysis And Prediction Model Research Of Network Traffic

As the rapid development and application of Internet, the scale of Intemet is becoming largerand larger. Since increasing network traffic is composed of datas from various applications,e.g.video, audio, image and so on，the characteristics of network traffic are becoming more and morecompliated, and correspondingly the requirements on the service quality and network securityare more and more stringent. All of these require a deeper understanding of network traffic andtraffic conditions. Network traffic prediction can provide strong supports for the networkmanagement, planning and maintenance.In this thesis, the time series theory and Hurst parameter estimation methods are introduced.Then, the performances of different estimation method for Hurst parameter are compared.Furthermore, a novel estimation methods for Hurst parameter based on MF-DFA algorithm isproposed. Simulation results are carried out to compare with the most commonly used R/Smethod validing the effective of our proposed method.Concerning the prediction model for network traffic, this thesis analyzes the predictability ofnetwork traffic. Then various predication models are introduced, including some commonshort-dependence models, long-dependence models, etc., with their performance comparison areprovided. Based on these models, AIC criteria is used for model order determination, and GeversWouters-LS algorithm is used to obtain the coefficients of the ARMA forecasting model.Specifically this model is composed of an AR model and an MA model. On the base ofStationarity and reversibility assumptions of ARMA model, the ARMA model is equivalent to aMA model with infinite and calculated by GeversWouters algorithm, and then the LS method isused to estimate the parameters of AR model. Our simulation results show that the proposedmodel can predict the network traffic more accurately.At the end of this thesis, some characteristics and shortcomings of the network traffic modelare summarized, and then prospect the future development direction of the network trafficprediction.