Page 148 - 《水资源与水工程学报》2023年第3期
P. 148
!34 " ! 3 # & ' ( ) & * + , - Vol.34No.3
2023 $ 6 % JournalofWaterResources&WaterEngineering Jun.,2023
DOI:10.11705/j.issn.1672-643X.2023.03.18
'¸GHIJ]KLX1&M¤v,NOP
Qj2ÅA]ef
1 1 2 1 3
²³´ , 7µ¶ , /2· , ¸ ¹ , )º»
(1. zrZ.8 NVNÕ8\, {z z 710048;2. ÞäN&ûM1'%SÏL/ý, Þä
¾ 610072;3. zrZ.8»É-1=r¥, {z z 710048)
< =: ï|²H±ª'ý"zgjÌÍ@, Mi¨w¦±Åýzôõ`¬1, L¾i8himP
LMzAÖ。ÞA/©uºÀÀÏR]=uzÏÒ/ôõ, Ì&, ï*uºÀrè-G9.
uº{Uôzi8h, ¼²H±ª(CFRD) z'ý"nMi¨n«ô, rôSzi¨©P1
[Çx(GPR), -GLM±ª'ý"zh。ïð CFRDznM'ý"Q, Ò Wavelet-GPRh
.'ý©imWLM, ¼W+©ôz GPRh'¿©´。¿Å:Wavelet-GPRhM
UWLMUzgbm^¹, ôS8]Èz´Ëq(RMSE) 8 0.9287mm4u 0.4577mm, 5
q(MAE) 8 0.4850mm4u 0.3306mm; LMÈz RMSE8 1.3089mm4u 0.9176mm, MAE8
0.9263mm4u 0.7303mm; {ôShzqMUi=LMU 95%Hnofz¼´ÏÅý°0。
Ì&, VÒuº{UônM'ýi¨©ýô¥rÚ ýôõHozi¨MUWnUà½zq,
Wavelet-GPRhÒ.LM²H±ªz'ý"ÎÏnÒ1WÀK1。
>?@: ²H±ª"; 'ýLM; uº{Uô; P1[Çx
ABCDE:TV641.4 FGHIJ:A FKLE:1672643X(2023)03014407
ApplicationofGaussianprocessregressionmodelbasedonwavelet
denoisinginsettlementpredictionofCFRD
3
1
2
1
1
HUANGLingzhi,CHENSiqi,LIChengyu,SIZheng,ZHANGFeiyue
(1.FacultyofWaterResourcesandHydroelectricEngineering,Xi’anUniversityofTechnology,Xi’an710048,China;
2.SichuanWaterDevelopmentInvestigation,Design&ResearchCo.,Ltd.,Chengdu610072,China;3.Phase Ⅲ
ConstructionAdministrationOffice,Xi’anUniversityofTechnology,Xi’an710048,China)
Abstract:ThesettlementanddeformationofCFRDarecausedbymultiplecomplicatedenvironmental
factors,andtheobserveddeformationdatashowobviouscharacteristicsofnoiseinterference,whichhasa
greatinfluenceonthefittingandpredictionaccuracyofmathematicalmodels;however,theusefulsignal
andnoiseintheoriginalsignalcanbedecomposedeffectivelybywavelettransformtheory.Therefore,the
wavelettransformtheoryisintroducedtoestablishamathematicalmodelbasedonwaveletthresholdde
noising ,whichisthenappliedtothedenoisingofthemeasuredsettlementdataofaCFRD,andthede
noiseddataisprocessedbyGaussianprocessregression (GPR)toestablishamodelforpredictingsettle
mentdeformationofCFRD.BasedonthemeasuredsettlementdeformationdataofacertainCFRD,the
Wavelet-GPRmodelisusedtofitandpredictthedamsettlement ,andthecalculationresultsarecom
paredwiththoseoftheGPRmodelwithoutdenoising.Theresultsshowthattheresidualdifferencebe
tweenpredictedvaluesandobservedvaluesoftheWavelet-GPRmodelisinaccordwithnormaldistribu
tion.TheRMSEofthelearningsegmentdecreasedfrom0.9287mmto0.4577mm,andtheMAEde
creasedfrom0.4850mmto0.3306mmafterdenoising.Furthermore ,theRMSEofthepredictionseg
!"#$:20220719; %$:20221129
'()*: !"67289:5;(41731289、51879217); lm9:5; Æ ïþÖålm9:5;(2021JLM46)
+,-.: ]³(1982—), î, z&B, CD, kEF, JKLMNZ"aJKE´²z%S。