Page 85 - 2024年第55卷第5期
P. 85
[23] ZHANGX,ZHENGZ,WANGK.PredictionofrunoffintheupperyangtzeriverbasedonCEEMDAN - NARmod
el [J].WaterSupply,2021,21(7):3307 - 3318.
[24] SIBTAINM,LIX,NABIG,etal.Developmentofathree - stagehybridmodelbyutilizingatwo - stagesignalde
compositionmethodologyandmachinelearningapproachtopredictmonthlyrunoffatswatriverbasin,Pakistan[J].
DiscreteDynamicsinNatureandSociety ,2020,2020:7345676.
[25] WANGH,SUNJQ.VariabilityofNortheastChinariverbreak - update[J].AdvancesinAtmosphericSciences,
2009,26(4):701 - 706.
[26] WUZH,HUANGNE.Ensembleempiricalmodedecomposition:Anoiseassisteddataanalysismethod[J].Ad
vancesinAdaptiveDataAnalysis ,2009,1(1):1 - 41.
[27] COLOMINASM A,SCHLOTTHAUERG,TORRESM E,etal.Noise - assistedEMDmethodsinaction[J].Ad
vancesinAdaptiveDataAnalysis,2012,4(4):1250025.
[28] COLOMINASM A,SCHLOTTHAUERG,TORRESM E,etal.ImprovedcompleteensembleEMD:asuitable
toolforbiomedicalsignalprocessing[J].BiomedicalSignalProcessingandControl,2014,14:19 - 29.
[29] LIANGH,BRESSLERSL,DESIMONER,etal.Empiricalmodedecomposition:amethodforanalyzingneural
data[J].Neurocomputing,2004,65:801 - 807.
[30] BENGIOY,SIMARDP,FRASCONIP.Learninglong - term dependencieswithgradientdescentisdifficult[J].
IEEETransactionsonNeuralNetworks,1994,5(2):157 - 166.
[31] GERSFA,SCHMIDHUBERJ,CUMMINSFA.Learningtoforget:continualpredictionwithLSTM[J].Neural
Computation ,2000,12(10):2451 - 2471.
[32] GRAVESA,SCHMIDHUBERJ.FramewisephonemeclassificationwithbidirectionalLSTM andotherneuralnet
workarchitectures [J].NeuralNetworks,2005,18(5?6):602 - 610.
[33] BOXGEP,JENKINS G M, REINSEL C.Time SeriesAnalysis: Forecasting and Control[M].London:
PalgraveMacmillan,2013.
[34] WANGJ,LUOY,TANGL,etal.AnewweightedCEEMDAN - basedpredictionmodel:Anexperimentalinvesti
gationofdecompositionandnon - decompositionapproaches[J].Knowledge - BasedSystems,2018,160:188 - 199.
Forecastingbreak - updateofrivericeinnorthernChinabasedon
deeplearningandsignaldecompositiontechnology
2
1
1
1
1
DINGHong,WANGWeize,YANGZefan,LIUHuan,HUPeng
(1.StateKeyLaboratoryofSimulationandRegulationofWaterCycleinRiverBasin,ChinaInstituteof
WaterResourcesandHydropowerResearch,Beijing 100038,China;
2.SchoolofCivilEngineeringandArchitecture,Xi’anUniversityofTechnology,Xi’an 710048,China)
Abstract:IcefloodsoccasionallyoccurduringrivericebreakingupinnorthernChinainspring,threateningthe
safetyofhydraulicstructures.Forecastingthebreak - updateofriverice (BUDRI)accuratelyisanimportantrefer
enceforanti - floodingcommandanddispatchingdecision - makingduringicebreakingperiod.Forforecastingthe
BUDRIinnorthernChina ,theobservedbreak - updateseriesofrivericeof5representativehydrologicalstationsin
HeilongjiangprovincelocatedinnorthernChinawasselected,andtheComplementaryEnsembleEmpiricalMode
DecompositionwithAdaptiveNoisetechnologyanddeeplearningmodelLongShortTerm Memory (CEEMDAN -
LSTM)wasusedtoforecasttheBUDRI.TheresultsshowthattheforecastaccuracyofCEEMDAN - LSTM,com
paredwithLSTM ,hadbeensignificantlyimprovedwiththemeanabsoluteerrorreducedfrom2.51dto1.20d,the
qualificationrateincreasedfrom91.59% to100% inthetrainingperiod.andthemeanabsoluteerrorreducedfrom
3.85dto1.65d ,thequalificationrateincreasedfrom 88% to96% inthevalidationperiod.TheCEEMDAN -
LSTM performedwellinforecastingtheBUDRIinnorthernChina ,whichcanprovideimportantinformationfor
command ,dispatch,anddecision - makingoficefloodcontrol.
Keywords:break - updateofriverice;signaldecompositiontechnology;deeplearning;forecastingmethod;
northernChina
(责任编辑:王 婧)
— 5 8 5 —