小波分析与神经网络的网络流量预测模型

摘要

网络流量具有长相关、非平稳性与多时间尺度特性。提出了一种基于小波分析与AR（p）人工神经网络相结合的网络流量预测模型，即WPBP算法。该算法采用小波分析得到网络流量在不同尺度下的近似信号和细节信号，并运用AR（p）的相关性理论确定近似信号序列和细节信号序列的相关程度（p值），与神经网络进行耦合，以P＋1划分数据，前P项作为输入，后一项作为输出对网络进行训练，从而使得神经网络的输入与输出的选择更加合理，预测的结果也更加准确。用小波重构得到最终的流量预测值，用实际网络流量对该模型进行验证。仿真结果表明，该模型的预测效果较好。%The network traffic has such natures of long-range dependence,non-stationarity and multi-time. In this paper,a new network traffic prediction model which eombines the wavelet transform and neural network is presented,that is WPBP algorithm. Using wavelet analysis we ean get approximation signal and detail signal at different scales. Approximation signal has not only maintained relevance of network traffic but a very good data smoothing, using the theoretieal of relevance of AR（p） to determine the value of p and then coupled with the neural network. And the data item is divided by P ＋ 1 as a pre-p input,the latter as the output of the network training. Thus the input-output selection of neural network is more rational and the prediction results are more accurate. Finally, wavelet reconstruction is used to get the final traffic predicted value. The simulation results on real network traffie show that WPBP model is more successful than traditional methods in network traffie prediction.