Research On Traffic Prediction Technology Based On Mixed Network Model

Network traffic prediction is the prediction model, based on the collected data of network traffic changes in a future time network traffic prediction, network administrators to master the running status of the network to provide some guidance.The traditional network traffic prediction method, linear regression model, Poisson model, Markov model and time series prediction model, the network traffic data is a time series in essence, so time series model is the most commonly used traditional model.In recent years, nonlinear prediction theory has been developed, such as neural networks, support vector machines and other machine learning methods, which are used in network control management.Firstly, the paper studies and analyzes the wavelet transform technology and several commonly used network prediction model. Through the research, we find that the time series analysis is the base of the traditional network traffic model. With this feature, the traditional traffic model has a good predictive ability in the smooth sequence. But in today’s increasingly complex network, network traffic began to appear unstable characteristics, so that the traditional network model in predicting ability is powerless. Through the research on the wavelet technology discussion, we found that the wavelet technology can be related to the processing of traffic data with long-range dependence, can be a bad deal in the time domain into the frequency domain to change. The characteristics of the wavelet multiresolution technique can efficiently deal with the sudden, network traffic self similarity, related characteristics and so on in the complex situation caused by the tangled together. Although the wavelet technology layer deconstruction to the signals of different frequency domain becomes a single, but more smooth.SO, the paper proposes the introduction of wavelet technology, and combines it with the traditional traffic model, so that the advantages of the traditional traffic model can be fully developed in the field of smooth sequence prediction. First using wavelet decomposition technique will be non stationary time series is decomposed into the appropriate different bands, multiple stationary time series, followed by the use of traditional network traffic model on the stationary time sequence modeling, the sequence in the original scale recovery and get the prediction results.In the introduction of wavelet decomposition technology to improve the prediction accuracy, we realized that if the decomposition of the number of layers were modeled, such a method will seriously affect the complexity of the time. Therefore, this paper considers the use of appropriate traffic prediction model, as much as possible to reduce the number of modeling. The premise is similar sequence characteristics of stable, stationary sequence of algebraic operations is still stable, and can still get reasonable final results. The experimental results show that, prediction accuracy was not affected. Experiments show that this improvement time complexity measures did not affect the prediction accuracy and prediction results are compared in wavelet decomposition technique is introduced to simple all subsequences that are fully modeled or even improved. Therefore, it is feasible to use wavelet and multi network model to predict the network traffic.