Page 130 - 2024年第55卷第8期
P. 130
[19] 颜建国,郑书闽,郭鹏程.高热流条件过冷沸腾传热系数试验研究及神经网络预测[J].工程热物理学报,
2022,43(6):1650 - 1659.
[20] 王新,闫文源.基于变分模态分解和 SVM 的滚动轴承故障诊断 [J].振动与冲击,2017,36(18):252 -
256.
[21] LIJ,CHENY,QIANZ,etal.ResearchonVMDbasedadaptivedenoisingmethodappliedtowatersupplypipe
lineleakagelocation[J].Measurement,2020,151:107153.
[22] DRAGOMIRETSKIYK,ZOSSOD.Variationalmodedecomposition[J].IEEETransactionsonSignalProcessing,
2014,62(3):531 - 544.
[23] KIM H,PARKM,KIM CW,etal.Sourcelocalizationforhazardousmaterialreleaseinanoutdoorchemical
plantviaacombinationofLSTM- RNNandCFDsimulation [J].Computers&ChemicalEngineering,2019,125:
476 - 489.
[24] HANL,DENGY,CHENH,etal.ArobustVRFfaultdiagnosismethodbasedonensembleBiLSTM withatten
tionmechanism :Consideringuncertaintiesandgeneralization[J].EnergyandBuildings,2022,269:112243.
[25] ZHANGD,SUNW,DAIY,etal.A hierarchicalearlykickdetectionmethodusingacascadedGRU network
[J].GeoenergyScienceandEngineering,2023,222:211390.
[26] 宁方立,韩鹏程,段爽,等.基 于 改 进 CNN的 阀 门 泄 漏 超 声 信 号 识 别 方 法 [J].北 京 邮 电 大 学 学 报,
2020,43(3):38 - 44.
[27] SPANDONIDISC,THEODOROPOULOSP,GIANNOPOULOSF,etal.Evaluationofdeeplearningapproaches
foroil& gaspipelineleakdetectionusingwirelesssensornetworks [J].EngineeringApplicationsofArtificialIntel
ligence ,2022,113:104890.
Small - scalepipelineleakdetectionbasedonVMDanddeeplearning
3
3
1
1,2
1
1,2
ZHENGShumin,YANJianguo ,GUOPengcheng ,XUYan,LIJiang,LIUZhenxing
(1.SchoolofWaterResourceandHydroelectricEngineering,Xi’anUniversityofTechnology,Xi’an 710048,China;
2.StateKeyLaboratoryofEco - hydraulicsinNorthwestAridRegion,Xi’anUniversityofTechnology,Xi’an 710048,China;
3.XinjiangColdandAridRegionsWaterResourcesandEcologicalWaterEngineeringResearchCenter(AcademicianWorkstation),
Urumqi 830000,China)
Abstract:Toaddressthechallengeofdetectingleakagesignalsundernormalpressureandsmall - scaleleaks,this
paperfocusesonthedetectionofwatersupplypipelineleaks.Theexperimentaldataofleakageundertheconditions
3
of100 - 220kPapressureand40 - 80m ?hvolumeflowwereobtained,andthevariationsinpressuresignalsunder
small - scaleleakconditionswereanalyzed.TheexperimentaldataisdenoisedbyusingVariationalModeDecompo
sition (VMD)toreducenoiseinterferenceandenhanceleaksignalcharacteristics,followedbystandardization
process.Thestudycombinestypicalrecurrentneuralnetworks ,includingLongShort - TermMemory(LSTM),Bi
directionalLongShort - Term Memory (BiLSTM),andGatedRecurrentUnit(GRU),withConvolutionalNeural
Network(CNN)toconstructthreedeeplearningleakagedetectionmodelsCNN - LSTM,CNN - BiLSTM,andCNN -
GRU.Thesemodelswereevaluated fortheirpredictiveperformance , amongthem, theCNN-GRU model
exhibitedthehighestpredictiveaccuracyof99.56% forallexperimentaldata.Theresultsindicatethatthemodels
demonstratehighaccuracyindetectingleaksundernormalpressureandsmall - scaleleakconditions.CNNprovesto
beinstrumentalinextractingpertinentfeaturesefficientlyandaccurately ,therebyimprovingthepredictionaccuracy
oftheleakagedetectionmodel.Theresearchprovidesvaluablesupportfortheintelligentmanagementofpipeline
leakagedetectionsystem.
Keywords:leakdetection;small - scaleleakage;variationalmodedecomposition;deeplearning;watersupplypipe
line
(责任编辑:李福田 鲁 婧)
— 1 0 8 —
0