On Modeling, Forecast And Simulation For Network Traffic Based On Wavelet Analysis

In recent years, the Internet has developed so quickly that, on one hand, the network traffic carried by the Internet has increased significantly and the network traffic traits also becomes more and more complicated; on the other hand, the network traffic information is having more extensive and further impact on people’s daily life. Thus, only by effective means can one perform the statistics and analysis on all the business traffic on the Internet, and then further boost the network operations such as network management, network maintenance and network planning and so on. With the rapid development of Internet, it is becoming quite difficult to obtain the Traffic Matrix(TM), which is impossible to get only via direct measurement, so it has become a useful method to acquire the TM by building a prediction model which can describe the network traffic traits precisely and effectively and can predict the change trend of the future network traffic.Concerning the highly ill-posed nature of network taffic when predicting along with the self-similarity and multifractal chatacter of the network taffic, we propose the multifractal wavelet model-based network traffic prediction model. The multifractal wavelet model is a multiplicity model of fractality, of which the edge is asymptotic logarithmic normal distribution. It has relatively strong global analysis ability and has opend up a new way to analysize traffic. The multifractal wavelet model has the characters of wavelet multi-scale as well as of the multifractal analysis, so it can deal with the nonlinear part of the network traffic effectively. The model can better depict the traffic traits than the traditional network prediction models, thus improving the prediction accuracy and predicting the actual network traffic precisely.Concerning the bad performance of depicting local traffic of multifractal wavelet model and those traffic traits of long range dependence, short range dependence, time-varying non-stationary and time-space Correlation itself, we propose the network traffic prediction model of combining Autoregressive (AR) model and Back Propagation Neural Network (BPNN) model. The model firstly decompose the traffic into the high-frequency part and low-frequency part through wavelet decomposition, and then explore the BPNN model to predict the high frequency part which has the characters of non-stationary and short range dependence, the AR model to predict the low-frequecy part which has long range dependence likewise. At last, we combine the separate two results into the final prediction results to finish the job of precise prediction.Concerning the bad performance of depicting local traffic of multifractal wavelet model, the insufficiency of AR model when seizing global traffic and the slow-learning-speed of BPNN model as well as the problem of too easy to reach those points which are not local optimum, we put forward the network traffic prediction model of combining the Grey Model (GM) and White Gaussian Noise (WGN). The model firstly decompose the traffic into the approximate part and detail part through wavelet decomposition, and then explore the GM to predict the approximate part, the WGN to predict the detail part likewise. At last, we combine the separate two results into the final prediction results to finish the job of precise prediction.