On-Line VBR Video Traffic Prediction Using Neural Network
SU Xiao-xing · Dianzi xuebao · 2005
An adaptive neural network model for VBR video traffic prediction is proposed in this paper.Firstly,adaptive training and pruning algorithm based on Extended Kalman Filtering(EKF) approach is used to train the Time Delay Neural Network(TDNN).By pruning the unimportant hidden weights,the corresponding redundant hidden neurons can be deleted,as a result a compact TDNN architecture can be obtained.The pruning process results in better generalization ability and lower computational complexity for the online stage.During on-line training stage,the TDNN's weights will be updated using Recursive Least Square(RLS) algorithm according to current prediction error.Since EKF and RLS are second order algorithms,they can estimate the learning step automatically,faster convergence speed and more precise prediction can be obtained.By simulation and comparison,the adaptive neural network model proposed in this paper is shown to be promising and practically feasible in obtaining the best adaptive prediction of real-time VBR video traffic.