IP骨干网络流量矩阵估计算法研究  

Algorithm of Traffic Matrix Estimation in IP Backbone Networks

作　　者：蒋定德[1,2] 胡光岷[2] 倪海转[1] 

机构地区：[1]东北大学信息科学与工程学院,沈阳110004 [2]电子科技大学宽带光纤传输与通信网技术重点实验室,成都610054

出　　处：《电子科技大学学报》2010年第3期420-424,共5页Journal of University of Electronic Science and Technology of China

基　　金：中央高校基本科研业务费专项资金(N090404014)

摘　　要：流量矩阵估计的高度病态特性,使得要精确地估计流量矩阵变得非常困难,如何克服这一问题的病态特性是当前面临的主要挑战。该文研究大尺度IP骨干网络流量矩阵估计问题,并利用BP神经网络的强大建模功能来捕捉流量矩阵的特征,通过将流量矩阵估计描述成约束条件下的最优化过程,能成功地克服这一问题的病态特性。仿真结果表明基于BP神经网络的估计算法具有明显的性能改善。Traffic matrix estimation is significantly difficult because it is a highly ill-posed problem.How to overcome the ill-posed nature of this problem is the main challenge faced at present.This paper studies the large-scale IP（Internet Protocol） traffic matrix estimation and uses backpropagation neural network to capture the characteristics of traffic matrix.By describing traffic matrix estimation into an optimal process under the constraints,ill-posed nature of this problem can successfully be avoided.Simulation results show that the estimation algorithm based on backpropagation neural network brings the evident performance improvement.

关 键 词：病态问题 网络流量 神经网络 流量矩阵估计 

分 类 号：TP393[自动化与计算机技术—计算机应用技术]