Study On The Network Traffic Prediction Model Based On The Wavelet Neural Network

Along with the development of IT and technology are developing rapidly,people on the network requirements are increasingly high, how to realize network QoS control, better for network management and maintenance, is a problem to be solved. And network traffic prediction for the QoS control and network intrusion detection plays a very important role. The network traffic prediction research by more and more realistic significance to people's attention.If want to accurately predict the network traffic, must establish a suitablemodel of network flow characteristics, and the detailed understanding is to establish a proper model foundation. But through a lot of analysis and experimental research shows that the reality of network traffic is very complicated. A model if not accurately capture actual network traffic statistical characteristic, will lead to network performance is bad, will appear on the performance of network too high or low estimate.Since the study of network flow characteristics has been using Erlang model, namely Poisson process, Package is no memory to arrive, packet arrival interval obeys index distribution. But in 1994 as W.L eland 10m Ethernet LAN by analyzing the flow data, off-line measurement revealed after the scale of the network traffic network traffic study, characteristics of enter a new historical stage. Network flow has various characteristics:self-similar, long related, heavy-tailed, chaos characteristics, fractal and scale features including time and space). In this case a single model has not suitable for analysis and prediction of the complex flow, wavelet analysis with its multi-resolution sex and to deal with emergency advantages, is widely used. Another neural network with its excellent nonlinear fitting characteristics also more and more attention. Both the combination model combined with the above advantages have broad application space.Based on wavelet neural network and the characteristics, puts forward a based on wavelet and neural network model. This experiment USES data for Changchun some operators switch port flow were collected 100 days,24 hours a day on the hour, before the port flow moments as 90 days neural network training data,10 days after the prediction by wavelet transform, the original flow data sequences decomposed into the details and approximate part, then to approximate part and each detail part single reconstruction respectively; After the reconstruction of the sequence, which were obtained using BP neural network and RBF neural network was forecast; Finally combining these prediction of original flow data that get predict. Forecast with real flow and forecast after compared to flow. And get the variance comparison error. We use the method of this paper forecast experiments, the result is satisfactory, it is proved that this model can accurately forecast in a certain extent network traffic.