Page 128 - 水利学报2025年第56卷第3期
P. 128
2022,46(1):15 - 24.
[16] LIKERA,BARTAZ.Theeffectsofdominanceonsocialforagingtacticuseinhousesparrows[J].Behaviour,
2002,139(8):1061 - 1076.
[17] 李楠,薛建凯,舒慧生.基于自适应 t分布变异麻雀搜索算法的无人机航迹规划[J].东华大学学报(自然
科学版),2022,48(3):69 - 74.
[18] 周广得,吕小龙,黄丹,等.基于 Kriging代理模型的迭代更新高效反演方法 [J].计算力学学报,2023,
40(4):602 - 607.
[19] 刘瑞江,张业旺,闻 崇 炜,等.正 交 试 验 设 计 和 分 析 方 法 研 究 [J].实 验 技 术 与 管 理,2010,27(9):
52 - 55.
[20] DRAGONMIRETSKIYK,ZOSSOD.Variationalmodedecomposition[J].IEEETransactionsSignalProcessing,
2014,62(3):531 - 544.
[21] SANGS,SHIK,LIK,etal.Short - termwindpowerpredictionbasedonEVMD - SSA - LSSVM Model[C]??2023
3rdNewEnergyandEnergyStorageSystem ControlSummitForum (NEESSC).IEEE,2023.
Studyontheseepagepredictionmodelofearth - rockdamsbasedon
mechanism- datafusionandresidualcorrection
1,2
1,2
1
1,2,3
HUANGHaoran,GUYanchang ,CHENSiyu ,WANGShijun ,HUANGHaibing 1,2
(1.DamSafetyandManagementDepartment,NanjingHydraulicResearchInstitute,Nanjing 210029,China;
2.DamSafetyManagementCenteroftheMinistryofWaterResources,Nanjing 210029,China;
3.KeyLaboratoryofFlood&DroughtDisasterDefenseoftheMinistryofWaterResources,Nanjing 210029,China)
Abstract:Themechanisticmodelscanpredictandevaluatetheseepagesafetystateofearth - rockdams,whichof
ferclearphysicalmeaningandgoodinterpretations ,buttheirpredictionaccuracyfluctuatesgreatly.Toenhance
thisaccuracy ,afusionmodelthatincorporatesadata - drivendeeplearningapproachwasintroduceinthisstudy,
andtheSparrowSearchAlgorithm(SSA)andRadialBasisFunction(RBF)wereemployedtoinvertthepermea
bilitycoefficient.ThisprocessconstructsanSSA - RBFsurrogatemodelforpredictingseepagepressure,yielding
boththemodel ’ spredictivevaluesandaresidualsequence.Then,theresidualsequencewasdecomposedbyu
singVariationalModeDecomposition (VMD),trainingaLongShort - TermMemory(LSTM)neuralnetworktoob
tainamodelforcorrectingtheresidualsequence.Byoverlayingthemechanisticmodelwiththedata - drivenmodel,
anSSA - RBF - VMD - LSTMfusionmodelwasconstructed,whichenablesaccuratepredictionsofseepagewaterlev
els.Theengineeringcasedemonstratesthatthemodelproposedinthispaperpossesseshighpredictiveaccuracy,
withimprovementsof89.64%,69.59%,and60.45% inpredictionaccuracycomparedtostatisticalmodels,
LSTM models,andSSA - RBF - LSTM models,respectively.Notably,evenwhentheseepageprocesslineunder
goessignificantfluctuations ,themodelisstillcapableofprovidingtimelyandaccuratepredictions,showcasing
goodstabilityandextrapolationcapabilities.Theseattributesmakethemodelworthyofpracticalapplicationand
dissemination.
Keywords:earth - rockdam;surrogatemodels; sparrow searchalgorithm; VariationalModalDecomposition;
LSTM neuralnetworks;mechanism - data - drivenfusion
(责任编辑:李 娜)
— 4 1 —
0