基于非负矩阵分解的网络距离预测  

Network Distance Prediction Based on Non-negative Matrix Factorization

作　　者：陈善雄[1] 廖剑伟[1] 张林[1] 彭茂玲[2] 

机构地区：[1]西南大学计算机与信息科学学院软件学院,重庆400716 [2]重庆城市管理职业学院,重庆401131

出　　处：《计算机工程》2011年第5期106-108,共3页Computer Engineering

基　　金：重庆市自然科学基金资助项目(CSTC;2010BB2006);中央高校基本科研业务费基金资助项目(100030-2120131007)

摘　　要：大规模网络中的节点存在高度动态性、不可达性等问题,使得直接测量节点之间时延有时根本无法进行。为此,提出一种非负矩阵坐标分解预测方法,通过获取普通节点与基准节点的RTT值构建一个特征矩阵(非负矩阵),并在欧式空间坐标计算的基础上,根据距离矩阵的非负性,确保其无损分解成基矢量和权矢量乘积,降低距离计算的维度,加快预测的收敛,其预测精确性符合预测误差趋势分布模型。As the existence of nodes＇ highly dynamic character and nodes＇ non-reachability in large-scale network,direct measurement of the node-delay is sometimes simply impossible.This paper proposes a method of Non-negative Matrix Factorization Prediction（NMFP）,constructs a characteristic matrix for the RTT between common node and benchmark node.Utilizing method of coordinate calculation in European space,according to the non-negative characteristic of distance matrix,it ensures the matrix non-destructively broken down into the product of the base vector and weight vector.This approach reduces the dimension of distance calculation,speeds up the convergence of prediction,the accuracy of the prediction in line with the trend forecast prediction error distribution model.

关 键 词：距离预测 非负矩阵 预测误差 

分 类 号：N945[自然科学总论—系统科学]