Page 25 - 2023年第54卷第8期
P. 25
NewunderstandingofThoma’sformulaforcriticalstablesectionof
surgechamberinhydropowerstation
2
2
2
1,2
ZHANGJian ,YAOTianyu,WANGQinyi,QIUWeixin,
2
1,2
CHENLong,CHENSheng
(1.TheNationalKeyLaboratoryofWaterDisasterPrevention,Nanjing210098,China;
2.CollegeofWaterConservancyandHydropowerEngineeringHohaiUniversity,Nanjing210098,China)
Abstract:Thoma’sformulaforcriticalstablesectionisgenerallyregardedasasignificantbasisforjudgingtheop
erationalstabilityofthewaterconveyanceandpowergenerationsystem ofhydropowerstationwithsurgechamber.
Thispresentstudyfirstlytheoreticallydemonstratesthesystemcannotbestablewithboththerigidbodyassumption
andconstantoutputregulation.TheThoma ’sformulaforcriticalstablesectioncouldbeobtainedbecausethewater
inertiaofthepenstocksandthetailracetunnelisneglectedinthederivation,whichisthekeyfactortothesystem
instability.Furthermore ,itisprovedtheoreticallythatthehydraulicsystemmaynotrunstablyundertheconstant
outputregulationmodefortheactualelasticwater.Meanwhile ,thestabilityofthewaterconveyanceandpower
generationsystemofthehydropowerstationwithsurgechambermainlydependsonthewaterhammerreflectioncoef
ficientoftheupstreamanddownstreamsideoftheturbine.Thefrictionofthediversiontunnel,penstockandtail
racetunnelareallbeneficialtostability ,whilethesizeofthesurgechamberhasrelativelylittleeffectonstability.
Underconstantoutputregulationmode ,thewaterconveyanceandpowergenerationsystemofthehydropowerstation
withsurgechambermustbeunstablenomatteradoptingtherigidwaterhypothesisoractualelasticwater.Anunstable
systemcouldnotexiststablesection ,andThoma’sformulaforcriticalstablesectionisnotvalidintheory.
Keywords:hydropowerstation;surgechamber;stability;Thoma’sformula;waterhammerreflectioncoefficient
(责任编辑:李福田)
(上接第 897页)
Probabilisticforecastingoffloodprocessesbasedonhybriddeeplearningmodels
1
1
1
1,2
2
CUIZhen,GUOShenglian,WANGJun ,ZHANGJun,ZHOUYanlai
(1.StateKeyLaboratoryofWaterResourcesandHydropowerEngineeringScience,WuhanUniversity,Wuhan 430072,China;
2.HydrologyBureau,YangtzeRiverWaterResourcesCommission,Wuhan 430010,China)
Abstract:Thetraditionalartificialneuralnetworkmodelcannotquantifytheuncertaintyoffloodforecastingand
isunabletoconsiderthetemporalcorrelationoffloodprocessforecastinginmulti - timecontinuousforecasting.In
thispaper ,aXAJ - LSTM- EDE - MDNmodelisconstructedbyfusingtheXinanjiang(XAJ)model,thelong
short - termmemory(XAJ - LSTM- EDE)neuralnetworkbasedontheexogenousinputencoder - decoderstructure,
andthemixturedensitynetwork(MDN)toachieveprobabilisticforecastingofthefloodprocess.Themodeltrans
formsthepointestimatesgeneratedbythedecodingprocessintotheestimatesofconditionalprobabilitydistribu
tionswhileconsideringthetemporalcorrelationoftheforecastedflood.Thelossfunctionisfurtherestablishedu
singthemaximumlikelihoodestimationmethod ,andthemodelparametersaretrainedbyanadaptivemomentes
timationalgorithm.ThestudyresultsinthetworiverbasinsofLushuiandJianxishowthatthemodelcaneffective
lyreflecttheuncertaintyoftheforecastfloodwithoutreducingtheforecastaccuracyoftheXAJ - LSTM- EDEmod
el ,andobtainreasonableandreliableconfidenceintervalsandexcellentprobabilisticforecastperformance.It
providesmoreriskinformationfordecision - makingsuchasreservoirfloodcontrolandscheduling ,andalsopro
videsareferenceforstudyingtheapplicationofdeeplearninginprobabilisticfloodforecasting.
Keywords: probabilisticforecasting; uncertaintyanalysis; longand short -term memory neuralnetworks;
encoder - decoderstructure;mixturedensitynetworks
(责任编辑:耿庆斋)
— 9 0 9 —