电信科学 ›› 2020, Vol. 36 ›› Issue (9): 51-58.doi: 10.11959/j.issn.1000-0801.2020270

• 研究与开发 • 上一篇    下一篇

基于密度聚类的网络性能故障大数据分析方法

李想,李原(),张子飞,杨哲   

  1. 中国信息通信研究院,北京 100191
  • 修回日期:2020-06-01 出版日期:2020-09-20 发布日期:2020-09-27
  • 作者简介:李想(1986- ),男,中国信息通信研究院高级工程师,主要研究方向为互联网监测分析、域名系统等|李原(1976- ),男,博士,中国信息通信研究院高级工程师,主要研究方向为互联网网络架构、互联网测量分析、下一代互联网、国际通信等|张子飞(1987- ),男,中国信息通信研究院工程师,主要研究方向为互联网域名系统、网络性能与业务体验分析等|杨哲(1990- ),男,中国信息通信研究院工程师,主要研究方向为宽带网络分析、互联网网络等

A density clustering-based network performance failure big data analysis algorithm

Xiang LI,Yuan LI(),Zifei ZHANG,Zhe YANG   

  1. China Academy of Information and Communications Technology,Beijing 100191,China
  • Revised:2020-06-01 Online:2020-09-20 Published:2020-09-27

摘要:

针对层出不穷的网络安全事件,如何快速在海量监测数据中发现异常数据,并开展网络故障分析成为研究难点。针对该问题,提出一种基于密度聚类的网络性能故障大数据分析方法,通过熵权分析、数据清洗与标准化处理实现关键性能特征提取与数据整形,基于参数调优的DBSCAN聚类算法提取性能故障异常数据。基于实时采集的全国多家运营商海量骨干网链路性能数据验证该算法,结果表明,与人工标注网络性能异常数据相比,其识别的准确性超过90%,可满足开展全国网络运行故障分析的需求。

关键词: 网络性能, 机器学习, 密度聚类, 测量分析

Abstract:

Facing frequent network security incidents,how to quickly find abnormal data in massive monitoring database and carry out network failure analysis becomes a research difficulty.A density-based network performance failure big data analysis algorithm was proposed,which extracted key performance characteristic indicators through entropy weight analysis,implemented data shaping through data cleaning and standardization,and extracted abnormal performance data on the basis of DBSCAN clustering algorithm.Relying on the real-time massive backbone network link performance data of multiple domestic operators to validated this algorithm,the results shows that compared with the manually manner,the recognition accuracy of the algorithm proposed to the network performance abnormal data is more than 90%,which can well fit for the analysis of real-time Internet network operation failure.

Key words: network performance, machine learning, density clustering, measurement analysis

中图分类号: 

No Suggested Reading articles found!