Research On Network Traffic Forecasting Based On Echo State Network

Network traffic is defined as the quantity of data transmits through a network. Because network traffic contains the key information when the network operates, and it is the important fundamental data when supervising and optimizing network, the analyzing and forecasting of network traffic have become vital. With the incessant abundance of business, network traffic has exhibited increasing complex characteristics. The using of the forecasting model of network traffic can assist people to prophetically perceive the future tendency of network, which profoundly influence several key technologies such as network programming, resource disposition and network security etc. How to establish an efficient and accurate model for network traffic forecast problem has become a challenging research focus.As a novel recurrent artificial neural network, echo state network(ESN) has attracted wide attentions from various fields due to its high-efficiency computing capacity and easy-to-use advantage. Furthermore, ESN adopts reservoir framework to compute, leading it has the ability of dynamic memory and powerful for temporal processing, which has been successfully applied at different engineering practices of time series. History network traffic record is viewed as time series data, and the network traffic forecasting is also viewed as a standard forecasting problem of time series, thus it is possible to utilize ESN to operate it. However, due to the complex multi-scale features of modern network traffic, such as self-similarity, long-range dependent and multi-fractal, the standard ESN has a loss on forecast accuracy.This paper develops a forecasting model of network traffic based on ESN, improved symbiotic organisms search(SOS) and ensemble empirical mode decomposition(EEMD). EEMD method is employed to decompose the original network traffic data into different scales. The high-frequency signal and the reconstructed data then are put into the input terminal of ESN. Meanwhile, the presented chaotic local SOS(C-SOS) is used to optimize the input weights of ESN. Two real network traffic dataset are adopted to verify the accuracy of presented model. The experiment result demonstrates that the proposed model has more high forecasting accuracy compared to other models and is an efficient network traffic forecasting model.