M-TAEDA：多变量水质参数时序数据异常事件检测算法

JournalofComputerApplications计算机应用，2017，37(1)：138一l44ISSN1001．9081CODENJYIIDU2017．．01．．10http：／／www．joca．cn文章编号：10(31．9081(2017)Ol一0138-07DOI：10．11772／j．issn．1001-9081．2017．O1．0138M—TAEDA：多变量水质参数时序数据异常事件检测算法毛莺池。，齐海，接青，王龙宝(河海大学计算机与信息学院，南京211100)(通信作者电子邮箱maoyingchi@gmail．corn)摘要：在供水管网中部署传感器网络实时获取多个水质参数时间序列数据，当供水管网发生污染时，高效准确地检测水质异常是一个重要问题。提出多变量水质参数时间异常事件检测算法(M—TAEDA)，利用BP模型分析多变量水质参数的时序数据，确定可能离群点；结合贝叶斯序贯分析独立更新每个参数的事件概率，预测单个传感器节点检测的异常概率；将单变量的事件概率融合为统一多变量事件概率，融合判断异常事件。实验结果表明：BP模型模拟多变量水质参数进行预测可以达到90％精确度；与单变量参数时间异常事件检测算法(S-TAEDA)相比，M-TAEDA可以提高异常检出率约40％，降低误报率约45％。关键词：无线传感器网络；异常事件检测；BP模型；多变量水质参数；时间序列数据中图分类号：TP393；TP18文献标志码：AM—TAEDA：temporalabnormaleventdetectionalgorithmformultivariatetime．seriesdataofwaterqualityMAOYingchi。，QIHal，JIEQing，WANGLongbao(CollegeofComputerandInformation，HohaiUniversity,NangJiangsu211100，China)Abstract：Thereal·-timetime··seriesdataofmultiplewaterparametersareacquiredviathewatersensornetworksdeployedinthewatersupplynetwork．Theaccurateandeficientdetectionandwarningofpollutioneventstopreventpollutionfromspreadingisoneofthemostimportantissueswhenthepollutionoccurs．Inordertocomprehensivelyevaluatetheabnormaleventdetectiontoreducethedetectiondeviation，aTemproalAbnormalEventDetectionAlgorithmforMultivariatetimeseriesdata(M-TAEDA)wasproposed．InM-TAEDA，itcouldanalyzethetime·seriesdataofmultipleparameterswithBP(BackPropagation)modeltodeterminethepossibleoutliers，respectively．M—TAEDAalgorithmcoulddetectthepotentialpollutioneventsthroughBayesiansequentialanalysistoestimatetheprobabilityofanabnormalevent．Finally，itcanmakedecisionthroughthemultipleeventprobabilityfusioninthewatersupplysystems．TheexperimentalresultsindicatethattheproposedM—TAEDAalgorithmcangetthe90％accuracywithBPmodelandimprovetherateofdetectionabout40％andreducethefalsealarmrateabout45％comparedwiththetemporalabnormaleventdetectionofSingle—VariateTemproalAbnormalEventDetectionAlgorithm(S-TAEDA)．Keywords：WirelessSensorNetwork(WSN)；abnormaleventdetection；BackPropagation(BP)model；multivariatewaterqua