Modified Elman neural network and its application to network traffic prediction

Abstract

The predictability of network traffic is of significant interest in many domains, including adaptive applications, congestion control, admission control, wireless and network management. An accurate traffic prediction model should have the ability to capture the prominent traffic characteristics, e.g. short and long dependence, self similarity in large-time scale and multifractal in small-time scale. For these reasons time series models are introduced in network traffic simulation and prediction. Accurate traffic prediction may be used to optimally smooth delay sensitive traffic or dynamically allocate bandwidth to traffic streams. A modified Elman neural network model is proposed for the network system in this paper. Compared to the traditional Elman neural network model, the proposed model is nonlinear, multivariable and time-varying and has higher accuracy and better adaptability. By the model, a abnormal behavior of network traffic can be found on time through the test of adaptive boundary value. The experimental results show the model is effective and feasible for Network traffic prediction.