基于马尔可夫模型的端到端流量预测方法研究.docVIP

  基于马尔可夫模型的端到端流量预测方法 研究# 蒋定德，聂来森，秦文达* 5 10 （东北大学信息科学与工程学院, 沈阳 110819） 摘要：针对当前通信网络中端端流量预测问题，研究端到端流量的准确预测方法，通过将端 到端流量建模为马尔可夫模型来描述其时间相关性和动态变化特性，并利用测量样本建立马 尔科夫模型的转移概率矩阵，考虑到端到端流量所满足的约束条件，首先预测获得单条端到 端流量流的预测值，然后获得所有端到端流量流的预测，最后通过约束调整获得最终的预测 结果。仿真结果表明所提出的预测方法能精确预测端到端流量并能准确捕获它的动态变化, 与以前的方法相比,具有更好的预测性能和更低的预测误差，并具有明显的性能改善。 关键词：网络流量；马尔可夫模型；流量预测；加权平均；预测误差 中图分类号：TP393 15 On Prediction Approach of End-to-End Traffic Based on Markov Model Jiang Dingde, Nie Laisen, Qin Wenda (College of Information Science and Engineering, Northeastern Universiy, Shenyang 110819) 20 25 30 35 Abstract: For the prediction problem of end-to-end traffic in current communication networks, this paper studies the prediction method about end-to-end traffic. End-to-end traffic is modeled as the Markov process to describe its temporal correlation and dynamic nature. The measurement samples about end-to-end traffic are used to build the transition probability matrix of the Markov model. After considering the constraints met by end-to-end traffic, end-to-end traffic is each predicted firstly. Then all of end-to-end traffic is attained. They are adjusted according to the constraints to obtain the final prediction. Simulation results show that the proposed prediction approach can accurately predict end-to-end traffic and capture its dynamic change, and in contrast to previous methods, it holds the better prediction performance and lower prediction errors and has the larger performance improvement. Key words: network traffic; Markov model; traffic prediction; weighted average; prediction errors 0 引言 端到端流量反映了通信网络中网络级流量的变化情况，其作为网络管理的重要输入参 数，是网络设计、规划和优化的重要参考依据。但由于端到端网络流量隐藏在链路流量中， 要直接获得非常困难[1-3]。端到端流量预测是获取端到端流量的有效手段，已成为当前的研 究热点[4-6]。然而，由于其动态变化和随机突发性，要进行准确预测是困难的，端到端流量 预测引起了研究人员的广泛关注 [5-9] . [6] [7] [8]  40  流的结构，利用主成分分析方法来建模和分析网络级流量的特征和属性，并提出相应的预测 基金项目：国家自然科学基金项目(No.; 高等学校博士学科点专项科研基金资助课题(No. 20100042120035); 新世纪优秀人才支持计划项目(No. NCET-11-0075); 中央高校基本科研业务专项资金资 助课题