| With the rapid development of computer network, the network isbecoming more colossal and complicated. Therefore more malfunctionedtroubles might happen. The quality of network is in turn influenced by thesetroubles .In the mean time, the difficulty of managing the network increases.So the maintenance and management of network should be given greatemphasis in order to serve the customers well. In traditional networkmanagement, the potential problems will be resolved after receiving the alerts,and it is very possible for the network to have been affected before beingsettled. This method is called as "response mode". Such as traffic monitoring,the network often are affected badly when alerts are sent out because ofexcess of traffic threshold, and it is usually no time to avoid it. Therefore, ifthe malfunctioned troubles can be predicted in advance by certain networkmanagement system, it will do a lot to help the network managers to takeactions in time, control the occurrence of the trouble and keep the networksmooth.In this paper, a method of network traffic prediction based on wavelettransform and autoregressive model is proposed. Antonini9-7 biorthogonalwavelet is used in Wavelet Transform.The relevant research backgrounds of network traffic prediction andinnovation of this paper are introduced in the preface.Wavelet transform, an important tool used in this paper is introduced inChapter Two. Firstly the definition of wavelet and several examples ofapplying wavelet are given. Then the definitions of successive wavelettransform, discrete wavelet transform are introduced as well asmulti-resolution and the famous Mallat algorithm. Finally single branchreconstruction algorithm is presented. Instead of reconstructing theapproximate part and detailed part in the same time, single branchreconstruction algorithm reconstructs them separately. That is when a certainpart is reconstructed, it is set as zero. The illustrative pictures are as follows:originalA1D1A2 D2……Aj Dj(A0)Fig.1 Decomposition process with Mallat algorithmAj 0Aj-1 0A1 0……ResultFig.2 Single branch reconstruction process of series AjSeveral basic definitions are presented in Chapter Three firstly. Thesedefinitions include traffic stability,long relativity and short relativity oftraffic model. Then Poisson Model and some Auto-regression traffic modelsare introduced. The models mainly include AR, ARMA, ARIMA, F-ARIAand GARMA.Chapter Four and Five are the cores of this paper. In Chapter Four, theoriginal discrete series consisting of network traffic data is decomposed intoapproximate series and several detail series. The prediction of the originalseries can be obtained by the synthesis of each reconstructed series'prediction result.Chapter Five is about experiment and error analysis. The collectionmethod of traffic data is introduced in the beginning of experiment part.Secondly the pictures of wavelet decomposition and single branchreconstruction are given. These pictures are based on the traffic data collectedbefore. The model's number of parameter,parameter estimation, trafficprediction of the sequence after reconstructed are handled in the end. Somedefinitions including absolute error,relative error,average absolute error,average relative error are introduced in error analysis part. Thereafter, theerror reference forms of different wavelet transform decomposition layers aregiven while applying the methods proposed in this paper. Finally acomparison is made among the application of this method and the other twomethods (one is the traffic prediction directly based on AR model. The otheris based on AR model after the proposition of traffic stable operation. Furtherpredication is presented as well as the error reference forms of five stepsprediction and ten steps prediction.As shown in our experiment, the novel method is of higher accuracy in comparisonwith the traditional ones. The outcome and method of our research and systemimplementation are helpful to the similar research. |
PDF Full Text Request