基于角度的多数据流异常检测方法  

Multiple Data Streams Outlier Detection Algorithm Based on Angle

作　　者：黄东东 徐建 张宏 HUANG Dongdong;XU Jian;ZHANG Hong(School of Computer Science and Engineering,Nanjing University of Science and Technology,Nanjing 210049)

机构地区：[1]南京理工大学计算机科学与工程学院,南京210049

出　　处：《计算机与数字工程》2020年第4期851-856,共6页Computer & Digital Engineering

基　　金：国家自然科学基金项目(编号:61300053)资助

摘　　要：同构计算环境下,计算系统中异常节点检测问题越来越受到关注。而同构环境下多个计算节点产生的信息可以看作多条高维数据流,则对于多条高维数据流中的异常数据流检测则显得极为重要。然而现有的异常检测研究主要集中在静态数据或是多条单维数据流方面,不能很好地应用在多条高维数据流的异常检测中。论文提出了一种新的多数据流异常检测方法,其基本思想是先在当前时刻利用角度作为数据流间的相似性度量指标,充分利用角度在高维数据中表现稳定的特性,对数据流进行当前相似性度量,然后结合数据流历史信息,对数据流间的整体相似性进行度量,最后以此来找出异常数据流。通过与传统的方法相比,验证了这种方法在高维多数据流异常检测时能获得更好的检测效果,并通过实验数据和分析得到了有益的结论。In homogeneous computing systems,the outlier detection problem has attracted more and more attentions.The data that the multiple computing nodes generated can been seen as a number of high-dimensional data streams.So it is extremely important to detect abnormal data streams in multiple high-dimensional data streams.However,at present researches focused on static data or multiple one-dimensional data streams.And these methods cannot be well used in multiple high-dimensional data streams.In the present paper,a new multiple data streams outlier detection algorithm is proposed.This algorithm uses angle to calculate the current similarity between multiple data streams,and consider the historical information to calculate the overall similarity between multiple data streams.Experimental results show that our algorithm can achieve better effect.Some useful conclusions are obtained through the analysis and explanation of the experimental data,which lay a solid foundation for further research.

关 键 词：高维 多数据流 异常检测 角度 

分 类 号：TP311.13[自动化与计算机技术—计算机软件与理论]