基于广义回归神经网络的流量矩阵估计  

Generalized regression neural network-based traffic matrix estimation

作　　者：蒋定德[1,2] 胡光岷[2] 

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：《计算机应用研究》2009年第7期2676-2679,共4页Application Research of Computers

基　　金：国家自然科学基金资助项目(60572092;60872033)

摘　　要：研究大尺度IP骨干网络流量矩阵估计,通过使用广义回归神经网络来捕捉流量矩阵特征,将流量矩阵估计描述成马氏距离下的最优化过程,能成功克服流量矩阵估计的病态特性,获得精确的估计值。仿真结果表明,该估计算法具有更高的估计精度和显著的性能改善。This paper studied traffic matrix estimation in the large-scale IP networks. By using generalized regression neural network to capture the characteristics of traffic matrix and describing the traffic matrix estimation into an optimal process under the Mahalanobis distance, the ill-posed nature of this problem could successfully be overcome and the accurate estimation could be obtained. Simulation results show that the estimation algorithm proposed holds the higher estimation accuracy and evident performance improvement.

关 键 词：流量矩阵估计 网络流量 广义回归神经网络 马氏距离 

分 类 号：TP393[自动化与计算机技术—计算机应用技术]