Research On Combination Prediction Model Of Network Traffic

As the rapid development and application of Internet, the scale of internet is becoming larger and larger, the application of the internet is becoming more and more complicated. Due to network is a very complicated non-line system, in order to realize reliable data transfer and reasonable internet resource distribution, it is very important to comprehend the control mechanism and complicated behavioral character of network. Excellent analysis and modeling could help to evaluate the network character more effectively. A predictable model which totally accord with the complicated network statistic character, could help to analyze and emulate the network accurately, and conduce to the network design and control.The aim of this article is to explore a new network model in order to describe and predict the network character accurately. In the beginning, the article analyze some main character about network, in the actual network environment, it present quite obvious multi-scale character, such as self-similarity, long-range dependence, fraction and multi-fraction; in following, the article analyze and compare the advantage and disadvantage of some traditional network analytic model, such as semi-Markov model, Poisson model, ARMA model and ON/OFF model.Based on the multi-time scale and the nonlinear character of the network traffic time series, a new network traffic prediction model which combines the wavelet transform and neural network was presented. The suggested model has advantage with its absorbing some merits of wavelet transform and artificial neural network. First, the traffic time series were decomposed to the scaling coefficient series and wavelet coefficient series. Then, RBF neural network and Elman neural network were used respectively to make prediction. Finally, the two predictions were combined into the final result through BP neural network. The simulation results on real network traffic show the new model has better predictive precision. In order to get even better predictable result, the article continue to research, introducing recurrent neural network to replace the normal neural network, the experiment result shows that the new model emulate the actual network traffic more effectively.