The Research Of Network Traffic Prediction Technology

Facing the rapid development of computer science and networks, The scale of the internet becomes bigger and bigger, and its bearing business and the field of application involve wilder. Enhancing the network management is a very important issue facing the more and more complicated internet. For the purpose of improving the network utilization and speed, it is significant to effectively predict the network traffic. We can significantly improve the network service’s quality, effectiveness and safety by predicting the network traffic overload situation to avoid problems. As a very cardinal part of the study of network behaviour, the prediction of network plays an important part in many domains as adaptive applications, congestion control, admission control, and wireless management.Dealing with the network traffic forecasting, scholars domestic and abroad have put forward some relevant technology, such as ARMA linear prediction model, neural networks. In this paper, some related model and technology have been analyzed to compare these technologies’ advantages and weaknesses. This paper analyzes the researching of LSSVM, which is one kind of machine learning methods used in non-linearly environment. AS a statistical learning theory, LSSVM can be used in the situation with small-sized samples to get a better precision. However, with the newly appearing characteristic of chaotic, non-stationary and complexity of the network, the existing methods are unable to predict it precisely.Regarding of the chaotic nature of network traffic, this paper produced a network traffic prediction model based on phase space reconstruction and LSSVM. After confirming the chaotic characteristic of network flow by computing the max lyapunov, we use the particle swarm optimized-LSSVM to train the multi-dimensional sequence to predict the future flow.For the non-linear and complex nature of network traffic, this paper produced a model involving wavelet transform and improved LSSVM. First, regarding of the superiority of multi-scale analysis wavelet transform plays out in nonlinear system, the network traffic is decomposed and recomposed into high-frequency components and low-frequency components.Then use the phase space reconstruction theory to determine the optimal delay time and minimum of the components with chaotic characteristic. And then we train the LS-SVM model to predict the best result of each component both reconstructed and not with chaotic characteristic. Finally, the integration of the prediction of each component is used as the prediction of our model. Simulating results show that the prediction of the model we proposed acquires higher precision than the other traditional models.