基于维度最大熵数据流聚类的异常检测方法  

Data stream clustering algorithm based on the maximum entropy of data dimension and its applications for anomaly detection

作　　者：耿志强[1] 姬威[1] 韩永明[1] 曹健[1] 

机构地区：[1]北京化工大学信息科学与技术学院,北京100029

出　　处：《控制与决策》2016年第2期343-348,共6页Control and Decision

基　　金：国家自然科学基金项目(61374166);教育部博士点基金项目(20120010110010);中央高校基本科研业务费专项基金项目(YS1404)

摘　　要：针对传统数据流聚类算法聚类信息损失大、不准确的缺点,提出一种基于维度最大熵的数据流聚类算法.采用动态数据直方图将数据维度划分为不同的维度组,计算各维度最大熵划分维度空间簇,将相同维度簇的数据聚集成微簇,通过比较微簇的信息熵大小及其分布特点实现数据流的异常检测.该方法提升了聚类速度,克服了传统数据流聚类算法信息丢失的缺点.实验结果表明,所提出算法能够提高数据流异常检测的准确性和有效性.In view of the traditional data stream clustering algorithm clustering information loss, inaccurate faults, a data stream clustering algorithm based on the dimension maximum entropy is proposed. Dynamic data in the sliding window are divided into different dimensions by using data histogram. The maximum entropy of different dimension is calculated to classify dimension spaces to form a cluster dimensions. Data are gathered into small clusters of the same dimension of cluster. By comparing the size of the cluster of information entropy and its distribution features, outlier detection of data stream is realized. This method improves the clustering speed, and overcomes the traditional shortcomings of the data stream clustering algorithm information loss. Experimental results show the effectiveness of the proposed algorithm.

关 键 词：维度簇 最大熵原理 数据流 信息熵 异常检测 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]