Research of Prediction Technology Based on Network Traffic Data Character Analysis
|School||National University of Defense Science and Technology|
|Course||Computer Science and Technology|
|Keywords||Network traffic Adaptive Forecast|
Expanding the scale of the Internet, online applications increasing network traffic continues to increase, and the Internet, there have been more and more problems. How effective management of the Internet has become the new challenges facing the people. For this reason, more and more researchers began to study the network traffic and hope to make some new ways to deal with these challenges in the study of the network traffic. Network traffic has self-similar characteristics, can be part of network traffic analysis to grasp the nature of the overall network traffic. Self-similarity of the network traffic, the burst of network traffic and drift. And for the nature of this paper, an adaptive traffic prediction method of genetic algorithm optimization neural network model. Compared to traditional traffic prediction algorithm, the prediction algorithm fully consider the burst of network traffic and drift on the accuracy of the forecasts greater improvement. The content of this article is divided into the following aspects: 1. Use of classic R / S rescaled range method to analyze network traffic self-similarity, and further study of this nature network traffic. The experiment shows network traffic with the characteristics of the power-law, and the superposition of the nature of the power-law is an important cause of sudden cause network traffic. 2 according to the the burst network traffic and drift, an adaptive genetic algorithm. The algorithm can be adjusted according to changes in flow adaptive convergence speed, the experiment shows that the algorithm in greater improvement compared to the traditional genetic algorithm performance. Using improved genetic algorithm to optimize BP neural network model and RBF neural network model, and compared with the use of traditional methods to optimize both neural network model. The prediction experiments through the use of real network traffic data to compare the improved genetic algorithm to optimize the model than the traditional method for greatly improved performance. 4 implements a network traffic prediction system. The system can predict network traffic on multiple time granularity, and can be selected according to the actual needs of the corresponding model. Network managers can make reasonable decisions based on the results of this system and timely response to emergencies in the network.