Page 125 - 2025年第56卷第2期
P. 125
culttoachieveintelligentwaterdistributiontasksincanalsystems.Therefore,asetofmeasurement - controlgatee
quipmentthatcanaccuratelymeasuretheoutletflowofthefinalchannelisdeveloped.Toachieveaccuratemonito
ringoftheflowthroughthegatestructure ,thispaperanalyzesthehydrauliccharacteristicsofthemeasurement -
controlgateandidentifiestheflowpatternatdifferentopenings (e)ofthegate.Then,basedontheBuckingham
theorem andtheleastsquaremethod ,theflowmeasurementmodelofthegatestructureisestablished.Finally,the
accuracyofthemodelisanalyzedandevaluated.Theresultsshowthatwhentherelativeopeningofthemeasure
ment - controlgateisrelativelyhigh,theflowthroughthegateisweirflow,andviceversa,itisorificeflow.The
criticalrelativeopening(e?H) c ofthecriticalrelativeopeningofthetwoflowpatternsis1.033.Theaveragerelative
errorofthedischargecalculationmodelconstructedfordifferentgateopeningsis4.7%,whichis26.3% moreaccu
ratethantheflowmeasurementequationforrectangularthin - plateweirs.Theerroroftheorificeflowmodelisless
than2.9%.Whene?H c >0.14,theflowisclassifiedasalargeorificeflowwithadischargecoefficientof0.7,oth
erwise,itisasmallorificeflowwithadischargecoefficientof0.846.Theresearchresultscanprovideareference
fortheaccuracyimprovementoftheintelligentgateflowmonitoringtechnologyoftheirrigationchannel.
Keywords:measurement - controlsluicegate;dischargecoefficient;dimensionalanalysis;precisionevaluation;
terminalirrigationchannelflowmeasurement
(责任编辑:王 婧)
(上接第 252页)
Researchonstreamflowintervalpredictionbasedondeeplearning
ensembleoptimizationmodel
1
2
1
HUANGJinghan,WANGZhaocai,WUJunhao,YAOZhiyuan 1
(1.CollegeofInformation,ShanghaiOceanUniversity,Shanghai 201306,China;
2.StateKeyLaboratoryofEstuarineCoastalScience,EastChinaNormalUniversity,Shanghai 200241,China)
Abstract:Duetotheincreasingoccurrenceofextremeweathereventsandthecomplexityofstreamflowvaria
tions ,itischallengingtorealizeaccuratestreamflowprediction,andpreviousstudiesaremostlybasedonpoint
predictionofdeterminatevalues,whichishardtotakeintoaccounttheeffectofuncertaintyandleadtothelack
ofpracticalapplicabilityofthepredictionresults.Inthisstudy ,adeeplearningensemblemodelforstreamflow
intervalpredictionbasedonmeteorologicalandhydrologicalvariablesisdeveloped.Themodelfirstfiltersoutthe
keydrivingvariablesaffectingstreamflowthroughthePearsoncorrelationcoefficient (PCC).Thentherawdata
aredecomposedintointrinsicmodefunctions(IMFs)byvariationalmodaldecomposition(VMD).Thecompo
nentsare then quadratically decomposed using complementary ensemble empiricalmodaldecomposition
(CEEMD)tocapturemoredetailsofthedata.Thepointpredictionresultsofstreamflowareobtainedbyagated
recurrentunit (GRU)incorporatinganattentionmechanism (AM),andanimprovedsparrowsearchalgorithm
(ISSA)isusedtooptimizehyperparameterssuchasthelearningrateoftheGRUandthenumberofhiddenlayer
dimensionstoenhancethemodelperformance.Finally,nonparametrickerneldensityestimation(NKDE)isin
troducedforintervalprediction.ThecombinedmodelVMD - CEEMD - ISSA - AM- GRU(VCIAG)constructedin
thisstudyperformsadvancemulti - periodpredictionforninehydrologicalstationsintheJialingRiverbasin.The
resultsofthestudyshowedthatthemodelperformswellintheshortterm,withNashefficiencycoefficients
(NSE)closeto1forseveralstations.Forfloodforecasting,themodel’sNSEforDongjintuo,Wusheng,and
Jinxistationsare0.73,0.92,and0.92,respectively.Inaddition,theeffectsoftheinputvariablesonrunoff
arequantifiedbytheShapleyvaluemethod(Shapley).TheVCIAGmodelproposedinthisstudynotonlyper
formswellinstreamflowpredictionaccuracy,butalsohassignificantadvantagesinintervalpredictionofuncer
tainty ,whichcanprovidemanagerswithmoreaccurateandreliablerunoffinformation,andthusbettersupport
runoffriskassessmentandscientificdecision - makingprogramsinpractice.
Keywords:deeplearningensemblemodels;streamflow intervalprediction;modaldecomposition; improved
sparrowsearchalgorithm;attentionmechanism
(责任编辑:韩 昆)
— 2 6 5 —