基于VMD-ED-云模型熵的管道运行状态定量识别技术-4cc77b0a.pdfVIP

采输技术

DOI：10.3969/j.issn.1001-2206.2022.04.011

基于VMD-ED-云模型熵的管道运行状态定量

识别技术

常振磊，崔永建，荆琳，王婷婷，王平，杨青霄

中国石油华北油田公司，河北任丘062552

摘要：针对管道不同运行状态下的信号特征在同一尺度上难以全面提取，提出基于变分模态分解（VMD）、

欧氏距离（ED）和云模型熵的管道运行状态定量识别技术。首先，采用VMD算法对负压波信号进行去噪处

理，根据最小巴氏距离法确定分解尺度，并通过计算各固有模态分量（IMF）与原始信号概率密度函数间的欧

氏距离确定有效信号分界点，提取重构信号的云模型熵值作为特征参数，输入支持向量机（SVM）进行工况识

别，并针对实例进行模型对比验证。结果表明，组合模型的总体识别效果较好，可识别正常、敲击、小孔泄

漏、大孔泄漏信号等四种工况，训练集上的分类准确率为100%，测试集上的分类准确率为95%，明显高于其

余对比模型，说明了组合模型具有可行性和合理性。

关键词：VMD；ED；云模型熵；管道；运行状态；识别

Quantitativeidentificationtechnologyforpipelineoperationstatebasedon

VMD-ED-cloudmodelentropy

CHANGZhenlei,CUIYongjian,JINGLin,WANGTingting,WANGPing,YANGQingxiao

HuabeiOilfieldCompany,CNPC,Renqiu062552,China

Abstract：Consideringthatthesignalcharacteristicsofpipelinesunderdifferentoperationstatesaredifficulttofullyextractonthe

samescale,aquantitativeidentificationtechnologyforpipelineoperationstateswasproposedonthebasisofvariationalmode

decomposition(VMD),Euclideandistance(ED),andcloudmodelentropy.Specifically,theVMDalgorithmwasusedtodenoisethe

negativepressurewavesignal,andtheminimumbusdistancemethodwasemployedtodeterminethedecompositionscale.Then,the

effectivesignaldemarcationpointwasdeterminedbycalculatingtheEuclideandistancebetweeneachintrinsicmodefunction(IMF)and

theprobabilitydensityfunctionoftheoriginalsignal.Thecloudmodelentropyofthereconstructedsignalwasextractedasthe

characteristicparameterandinputintothesupportvectormachine(SVM)toidentifytheworkingcondition.Finally,theproposedmodel

wascomparedwithothermodelsandtherebyverifiedthroughanexampleanalysis.Theresultsshowthatthecombinedmodeloffers

favorableoverallidentificationperformanceandcanidentifythesignalsoffourworkingconditions,namelynormalstate,knocking,small

holeleakage,andlargeholeleakage.Moreover,theclassificationaccuracyofthecombinedmodelonthetrainingandtestsets,

respectively100%and95%,aresignificantlyhigherthanthatoftheothermodelsundercomparison,indicating