Research Of Esn-Based Network Traffic Prediction Algorithm

With the continuous expanding of network business scope and sharp increasing of network users, the network pressure increases day by day. The future network with good network awareness and intelligent decision-making ability has been the focus of network research. The network traffic awareness can largely improve the quality of services and performance of network as a key aspect of network awareness ability. The short-term and long-term forecast of network traffic is extremely helpful for network control and network resources adjustment. With the analysis and prediction of history network traffic data, it can effectively improve the performance of network to adjust the network resources accordingly.Firstly, the background of this paper is presented as three parts. In the first part, we introduce the new network model under the future network architecture. The second part, we talk about the definition of prediction of network traffic and its current state of research. In the third part, we introduce the background of neural network and its application in network traffic forecast.Secondly, this paper focuses on ESN (Echo State Network). MCESN (Minimum Complexity Echo State Network) is introduced in this paper. We propose two new MCESN prediction algorithms with wavelet injection in reservoir of ESN:minimum complexity ESN with wavelet injection and wavelet minimum complexity ESN.Thirdly, this paper will use real datasets just like network traffic dataset to test our new MCESN models on the aspects of prediction accuracy, the range of effective spectral radius, short-term memory capacity, the robustness and so on in detail. Experimental results show that our new algorithms have excellent prediction accuracy, good range of spectral radius, stable prediction ability. This paper also gives the advices of the choice of input weights and reservoir weights, ratio of wavelet injection, symbol distribution of input weights and so on. Wavelet minimum complexity ESN algorithm gets the perfect prediction ability with few time cost and it has a very good application value.Finally, we discuss how to predict the network traffic under a future network architecture, which is designed hierarchically based on virtualization. And we also propose a prediction scheme for future network traffic based on Wavelet minimum complexity ESN model.