Predicting Bursty Network Traffic with Self-similarity Characteristic over Echo State Covariation Orthogonality Network
Xiaochuan Sun, Yingqi Li, Minghui Zhang · International Journal of Multimedia and Ubiquitous Engineering · 2017
Network traffic depicts the network characteristics and users behaviors.Accurate network traffic prediction is essential for dynamic network management.In this paper, the echo state covariation orthogonality network (ESCON) is proposed in a linear unbiased estimation framework based on echo state mechanisms for network traffic prediction.The ESCON inherits the basic idea of ESN learning in an unbiased estimation framework, but replaces the commonly used least square method with a covariation orthogonality one, which can reflect the tendency of network traffic more accurately, to solve the optimal output weights.We perform a comprehensive performance evaluation, considering publicly available nonstationary H.264 video traces.In all traces, we show that the ESCON can more effectively capture the characteristics of self-similarity and bursty, and yield superior prediction accuracy than the considered prediction schemes.