Research Of Traffic Prediction Model Based On Wavelet And FIR Neural Networks

With the computer network scale being more and more enormous and complicated, the probability of the error increases, which deteriorates the network performance. The network traffic can reflect the network performance directly. If the traffic received by the network exceeds its real carry abilities, the network performance will go to bad. So the model construction and the prediction of network traffic are significant for the lay-out and the design of the large scale network,the management of the network resources,the adjust of the users'behavior and so on. This paper is based on the project named the Multimedia Network Performance Prediction Model Based on Covariation-orthogonality and United Optimization which is supported by NSFC.Firstly, the characteristics of the network traffic are introduced. After analyzing the wavelet theory and the FIR neural network, a network traffic prediction model based on wavelet transform and FIR neural networks is proposed. The model employs wavelet transform which decomposes the traffic into high frequency coefficients and low frequency coefficients, and then these different frequency coefficients are reconstructed by single branch to the high frequency traffic parts and the low frequency traffic parts which are sent individually into different FIR neural networks for prediction. The synthesizer of the FIR neural networks outputs are the predicted results of the original network traffic.Adopted the authorized network traffic data as the testing object, the simulation experimental are completed as follows: the single-step prediction of the LAN network traffic and the WAN network traffic,the single-step prediction performance and its analysis with different wave function,the single-step prediction performance and its analysis with different structures compared to the wavelet neural network model and the FIR neural network model. The experimental results prove the efficiency and the superiority of the proposed prediction model.Finally, the main contributions in this dissertation are summarized and some suggestions and directions for the future work in this field are given.