Page 64 - 2025年第56卷第5期
P. 64
2022,51(10):122 - 129.
[10] YANGY,LIUA,XINH,etal.Faultearlywarningofwindturbinegearboxbasedonmulti - inputsupportvector
regressionandimprovedantlionoptimization [J].WindEnergy,2021,24(8):812 - 832.
[11] ZHANGR,ZHOUQ,TIANL,etal.Anoveloutlierdetectionmodelforvibrationsignalsusingtransformernet
works[J].IEEEAccess,2022,10:57234 - 57241.
[12] 杨茂,张书天,王勃,等.基于门控循环加权共形分位数回归的风电功率短期区间预测[J?OL].中国电机工
程学报,2024.(2024 - 09 - 11)[2024 - 12 - 17].http:??kns.cnki.net?kcms?detail?11.2107.tm.20240910.1548.004.html.
[13] MATHONSITHABANG,VANZYL.Multivariateanomalydetectionbasedonpredictionintervalsconstructedusing
deeplearning [J?OL].NeuralComputingandApplications,2022.(2022 - 01 - 03)[2024 - 05 - 12].https:??link.
springer.com?article?10.1007?s00521 - 021 - 06697 - x.
[14] LIY.Anomalydetectioninwirelesssensornetworksbasedontimefactor[J].JournalofIntelligent& FuzzySys
tems,2019,37(4):4639 - 4645.
[15] PANGJ,LIUD,PENGY,etal.Optimizethecoverageprobabilityofpredictionintervalforanomalydetectionof
sensor - basedmonitoringseries [J].Sensors,2018,18(4):967.
[16] 余宇峰,朱跃 龙,万 定 生,等.基 于 滑 动 窗 口 预 测 的 水 文 时 间 序 列 异 常 检 测 [J].计 算 机 应 用,2014,
34(8):2217 - 2220,2226.
[17] DEWOLFN,BAETSBD,WAEGEMANW.Validpredictionintervalsforregressionproblems[J].ArtificialIn
telligenceReview ,2023,56(1):577 - 613.
[18] STEINWARTI,CHRISTMANN A.Estimating conditionalquantileswith the help ofthe pinballloss[J].
Bernoulli ,2011,17(1):211 - 225.
[19] CHENX,ZHAN Y.Multi - scaleanomalydetectionalgorithm basedoninfrequentpatternoftimeseries[J].
JournalofComputationalandAppliedMathematics,2008,214(1):227 - 237.
[20] 王彦兵,王聪,赵亚丽,等.基于 ROC曲线的永久散射体识别最佳阈值定量筛选 [J].遥感学报,2021,
25(10):2083 - 2094.
[21] 曲力涛,潘罗平,曹登峰,等.基于振动能量趋势预测和 K均值聚类的水电机组故障预警方法研究 [J].
水力发电,2019,45(5):98 - 102.
Adynamicintervalcoveragemethodfordetectingabnormalvibrationofhydroelectricunits
1
1
1
1
1
WANGKun,WANGShaoqiang,XIONGXing,LIAng,QIUXudi,WANGBin 1,2
(1.CollegeofWaterResourcesandArchitecturalEngineering,NorthwestA&FUniversity,Yangling 712100,China;
2.KeyLaboratoryofAgriculturalSoilandWaterEngineeringinAridandSemiaridAreas,
MinistryofEducation,NortheastA&FUniversity,Yangling 712100,China)
Abstract:Abnormaldetectioninhydropowerunitsiscrucialfortheirstableoperation.Existinganomalydetection
methodsbasedonpointpredictionsoftenfailtoadequatelyaddressdatainstabilityandmodellimitations.Thispaper
proposesadynamicintervalcoveragemethodfordetectingvibrationanomaliesinhydroelectricunits.Initially ,an
intervalpredictionmodelisconstructedusingdirectintervalestimationandquantilelossasthelossfunctiontotrain
themodel.Adynamicintervalcoveragerateisthendevelopedbyintegratingaslidingwindowwithintervalcoverage
rateofthemodelevaluationindex.Subsequently ,thereceiveroperatingcharacteristiccurveisutilizedformodele
valuationandoptimalthresholddetermination.Thismethodallowsforflexibleadjustmentofdetectionintensity
throughconfidencelevel ,slidingwindowwidth,andstepsize.Tovalidatetheeffectivenessoftheproposedmeth
od ,simulationexperimentsareconductedusingbothsimulatedandrealanomalydataofhydroelectricunits.The
experimentalresultsindicatethatthismethodachievestruepositiveratesandfalsepositiveratesof1.0and0.015,
respectively,onsimulateddataset,and0.979and0.053onanother,accuratelyidentifyingvibrationanomaliesin
hydropowerunits.Thisapproachprovidesanewtechnologicalmeansforanomalydetectioninhydropowerunits.
Keywords:hydroelectricunits; abnormaldetection; intervalprediction; slidingwindow; receiveroperating
characteristiccurve
(责任编辑:王 婧)
— 6 1 —
0