基于模糊聚类的CO_2数据流时空异常模式的研究  

Fuzzy cluster based approach for spatial-temporal anomaly detection over CO_2 data streams

作　　者：刘莘[1,2] 张赛男[3] 

机构地区：[1]徐州医科大学医学信息学院,江苏徐州221004 [2]中国矿业大学环境与测绘学院,江苏徐州221008 [3]中国人民解放军理工大学指挥信息系统学院,南京210007

出　　处：《计算机应用研究》2016年第8期2353-2357,共5页Application Research of Computers

基　　金："十二五"科技支撑计划项目(2011BAC08B03);江苏高校优势学科建设工程资助项目(SZBF2011-6-B35)

摘　　要：传统的异常检测算法不能区分CO_2数据流的异常类型,为了有效识别因泄漏造成CO_2数据流的异常,提出了基于模糊聚类的CO_2数据流时空异常模式检测算法。该算法首先利用3σ规则实现自适应阈值的异常点检测,其次提取待检测滑动窗口的特征值(均值),构建指定区间内邻居节点间的时空关系矩阵,采用模糊聚类分析相邻节点特征值的时空相关性并对其进行分类,根据分类结果确定泄漏异常概率,最后利用真实观测数据对算法进行验证并对参数的选取进行分析。实验结果表明该算法能有效地识别因泄漏造成的事件异常,具有较高的检测率和较低的误警率。In view of the traditional anomaly detection algorithms could not identify CCS abnormal monitoring data types, so in order to identify the CO2 data streams anomalies caused by leakage, this paper proposed a fuzzy clustering based approach for spatial-temporal anomaly detection over CO2 data streams. Firstly, it used the σr rules as an adaptive threshold to realize the outlier detection ; Secondly, extracted the average of detecting sliding window as its characteristic value, and then built a spa- tial-temporal matrix between neighbor nodes in specified interval, analyzed the spatial-temporal correlation of adjacent nodes characteristic value based on fuzzy clustering, and classified the results. The algorithm could identify the abnormal leakage probability according to the results of the classification. Finally, this paper evaluated the algorithm by real observation data, and analyzed the selection of parameters. The results show that the proposed algorithm can recognize the anomaly caused by leakage, and has higher detection rate and lower false rate.

关 键 词：模糊聚类 时空异常 CO2 数据流 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]