Research Of Prediction Technology Based On Network Traffic Data Character Analysis

With the size of Internet expanding fast, the online application growing and network traffic increasing, there are more and more problems in Internet. How to manage Internet effectively is a new challenge that people face. Because of this, more and more researchers begin to study the network traffic, they hope to develop some new methods to deal with the challenge.As the self-similar character of network traffic, it is possible to analyze part of the network traffic for grasping the nature of the overall network traffic. By making use of its self-similar nature, this paper analyze the burst and drifting character of network traffic. And to deal with the character, this paper propose an adaptive genetic algorithm which is used to optimize neural network model on traffic prediction. Comparing it with the traditional traffic prediction algorithm, because of this method takes burst and drifting character into account, so it improves much on accuracy of prediction.The main contents of this paper are as follows:1. By using the classical R/S method, the self-similar character of network traffic was analyzed, and some further research based on this character was carried out. The power-law character of network traffic was illustrated through the experiment, and it was testified that the superstition of power-law is the main reason of network burst.2. According to the network traffic burst and drifting character, an adaptive genetic algorithm was proposed. This algorithm could adjust its convergence rate according to the traffic change. The experiment shows that the performance has a larger improvement when compare it to the traditional genetic algorithm.3. The BP neural network model and RBF neural network model were optimized by the improved genetic algorithm, and they were compared to the traditional algorithm improved model. Through the contrast of real network traffic prediction experiment, it was illustrated that the performance of improved genetic algorithm optimizing model was much better than traditional method.4. A network traffic forecasting system was implemented. This system could be used for network traffic prediction in multiple time granularity, and an appropriate model could be selected according to actual needs. Network manager could make decisions by the prediction results to deal with all kinds of emergencies.