Study Of Campus Network Traffic Analysis And Prediction

With the expands of network size，network administrators need to grasp network’srunning state to take measures accordingly so as to realize network Qos control. Themodel meeting the features of complex statistics of network flow can help the analysisand simulation of network flow, what’s more, the design and control of network. As thebasis of implementation of Qos control, the measure and prediction and network floware being increasingly valued by researchers.The features of network flow include self-similarity, periodicity, chaos and bursty.In order to accurately predict network flow and more effectively manage and regulatethe running of network, BP neural network and wavelet neural network will be appliedin the process of the prediction in this paper. Through the study of forward nets, thestructure and the parameters of wavelet neural network will be determined. Furthermore,through the replacement of forward nets’ hidden layer’s transfer function with waveletfunction, the model of mixed net can be given, namely, the model of wavelet neuralnetwork. In this paper, It is pointed out that better prediction can be achieved by the useof the model of mixed net that can combine advantages of other various models.In this paper, the campus network of Xi’an University of Posts and Telecommunications is takenas a research platform. Through the collection of data of real network flow, it is confirmed that thecampus traffic information does have self-similarity. Besides, the model of BP neural network andthe model of wavelet neural network are established to predict the real collection of network flow.After analyzing the results of experiments, it is demonstrated that both of the models can wellpredict the flow of the campus network in the near future. What’s more, it is proven that the modelof wavelet neural network is more effective than the sole model of BP neural network in predictingnetwork flow and the model of integrated type is a better one than the loose type.