Self-similar network traffic prediction based on burst decomposition

Traffic prediction plays an important role for congestion control and resource allocation of network management. For self-similar network traffic, in virtue of aggregated burstiness on all time scales, the prediction accuracy of traditional traffic prediction methods is generally low. In order to deal with the burstiness in traffic data, a burst decomposition based traffic prediction method is proposed in this paper. More specifically, before performing traffic prediction, the burst will be separated from the traffic by control chart, which leads to the decomposition of original traffic into the burst part and non-burst part. Then Artificial Neural Network(ANN) based method and adaptive template matching based method are used for the prediction of non-burst part and burst part respectively. Finally, the prediction results of original traffic are obtained through combining the predictions of the two parts. The real traffic based experimental results show that the prediction accuracy of the proposed method is higher than that of traditional methods.