Page 88 - 《水资源与水工程学报》2023年第4期
P. 88
!34 " ! 4 # & ' ( ) & * + , - Vol.34No.4
2023 $ 8% JournalofWaterResources&WaterEngineering Aug.,2023
DOI:10.11705/j.issn.1672-643X.2023.04.10
'a ASWPD-BO-GRUdh/>æCZ
mwx, Vyz, ?{|
( 81I& @JK&P ÐàÁ@AK8B.¬6°Í&ѾÛå, /0 01 710054)
H I: ?uXdZ[ÁZ, Äïo"#·z,¨d[qrL¢cnÛù×^, uhÄLso'x
#¬(¨·z(ASWPD) WJJ·qV(BO) ^/Ôáa3(GRU) dZ[qr(ASWPD-BO-GRU)。
ö, J ASWPDo®NdZâfg}¶·z, êcnÛù×^tu¥å 4 {oÇÈ^·zÛf
g, Σ¤[IZ; $ô, J BOq
·zô^Ûfgo¬^ GRUqrõûù; ±®, oÿÛfg}¶[,
[¯{ì&x¥hdZ[¯。uhÄLs^qr¬xÀdeKkÄ@AdZ[¢, Ä
K GRU、BO-GRU、WPD-BO-GRUqr( 'x"#·z¿Do®NdZâfg_V}¶·z^[qr)
^[¯}¶oº。¯ï:ASWPD-BO-GRUqr^Sà. Wù (NSE)? 0.89, ¾«¬¢[ÁZ
±X, A ASWPD-BO-GRUqrA·z^tuþÿûX^[ÁZWkY^~VQi。
JKL: dZ[; #¬]B·zs!; (¨·z; JJ·qV; /Ôáa3
+
+
MNOPQ:TV121 4;P338 .2 7RSTU:A 7VWQ:1672643X(2023)04008408
MonthlyrunoffpredictionmodelbasedonASWPD-BO-GRU
TANGMingze,YANGYinke,ZHANGJingwen
(KeyLaboratoryofSubsurfaceHydrologyandEcologicalEffectsinAridRegion,Schoolof
WaterandEnvironment ,Chang’anUniversity,Xi’an710054,China)
Abstract:Toimprovetheaccuracyofmonthlyrunoffpredictionandtoaddresstheproblemthatthecon
ventionaldecompositionintegratedrunoffpredictionmodelsincorrectlyusesfuturedata,agatedrecurrent
unit (GRU)monthlyrunoffpredictionmodel(ASWPD-BO-GRU)basedonselfadaptationstrategy
waveletpacketdecomposition(ASWPD)andBayesianoptimization(BO)isproposedanddeveloped.
First,inordertoreducethepredictiondifficultytheoriginalmonthlyrunofftimeseriesisdecomposedu
singASWPD ,bywhichfourrelativelyregulardecomposedsubseriesareobtainedwithoutusingfutureda
ta.Then,thehyperparametersoftheGRUmodelcorrespondingtothedecomposedsubseriesareopti
mizedusingBO.Finally ,themonthlyrunoffpredictionresultsareobtainedbypredictingeachsubseries
andsummingandreorganizingthepredictionresults.Theproposedandestablishedmodelisappliedto
thepredictionofmonthlyrunoffatYingluoxiaHydrologicalStationintheHeiheRiverBasin,andthepre
dictionresultsarecomparedwiththoseofGRU,BO-GRU,andWPD-BO-GRUmodels(models
basedontheconventionaldecompositionideawhichdecomposestheoriginalmonthlyrunofftimeseriesas
awhole ).TheresultsshowthattheNash-Sutcliffeefficiencycoefficient(NSE)ofASWPD-BO-GRU
modelis0.89 ,whichhasthehighestpredictionaccuracyintheexampleapplication,indicatingthatthe
ASWPD-BO-GRUmodelhashigherpredictionaccuracyandstrongergeneralizationabilitywithcorrect
decomposition.
Keywords:monthlyrunoffprediction;selfadaptationdecompositionstrategy(AS);waveletpacketde
composition(WPD);Bayesianoptimization(BO);gatedrecurrentunit(GRU)
!"#$:20230301; %$:20230405
'()*: 81I&¢üXÀ'u%;øX)C()*(300102292903); ¢!%&P~pKËâÚÁ¡!"&Ѿ
Ûå±%'()*(SKLLQG1933); /0#$%&'w;<ÌÊ)*(2017JM4018、2021SF497)
+,-.: =ÿ(1998—), ., /0013, ß5;<8, @É@A@CD=Î^;<。
¥¦+,: Q÷%(1977—), ., /0âI3, 45, ª67, @É@A@CD=Î^;<。