Page 173 - 《水资源与水工程学报》2024年第6期
P. 173
!35 "!6 # & ' ( ) & * + , - Vol.35No.6
2024 $ 12 % JournalofWaterResources&WaterEngineering Dec.,2024
DOI:10.11705/j.issn.1672-643X.2024.06.17
'×vxABÕØÙÚÛÜÝÞß¼àáAB
1 1,2 2 2
789 , :;< , 2 = , > ?
(1. M¦%3j& emno&p, !M M¦ 215011;2. yïj& J§J)lP&&p, r 200092)
C D: ìMµß._`KnçÉønÆ, ß #$(l´¬ìÃf÷xRDlƨ(7P。§[â
8P#$½Wî6bßPQR·«7wF。$Ì, Û%&RÙÚs£\ÞFúƨ6u'=b
no?:x[, ̪«1}<÷Ò¥^lo(y$HÚ, ¥ï´。'×#$çÉÛ%&
Rª«D¯jßs9~, ¬ßb#$ê«GRN)*¼s4 6 3+å¼slGÈ, ³´ LightGBM,&
-1},P>Ps9\â8。½¾¿À: éì8~â8¾PQ, ./~l8~
0¯ÇQçs! 0.9, ð12·Ô©µÍ× 4.2%, -.× XGBoost N SVR4¤3,&-u}。F
¾, ³´ SHAP 1}·«4ÓGÈlâ8½¾"¨#5R, ÝûǪ6R。éâ8ÕÏá
º»I_`P>Pâ8, áï´×#$î6b¤3·«nl, ëÄnoï´Å£。
EFG: çÉ#$P; Û%&R;LightGBMu}; ,&-; s9\â8
+
HIJKL:TV554 .14 MNOPQ:A MRSL:1672643X(2024)06016909
Predictionmodelfortunnelliningpressurebasedon
analyticalmodelsandmachinelearning
2
2
1,2
1
XUYujie,WANGHuaning ,HUTao,SONGFei
(1.SchoolofCivilEngineering,SuzhouUniversityofScienceandTechnology,Suzhou215011,China;
2.SchoolofAerospaceEngineeringandAppliedMechanics,TongjiUniversity,Shanghai200092,China)
Abstract:Forhydraulictunnellinginrheologicalsoftrocks,timedependentsupportforcesarelikelyto
occurattherock-supportinterfaceduetotheinteractionsbetweenthesurroundingrocksandthesupport
structures.Abetterunderstandingofthemechanismiscrucialforthedesignofliningstructuresandthe
stabilityanalysisofsurroundingrocks.Numericalsimulationsofviscoelastic-plasticproblemsaretime
consumingduetothelargeconsumptionofcomputationaliterationsandpre/postprocessing ;mean
while ,theanalyticalapproachneedscomplexformuladerivationandprogramimplementation.Therefore,
bothmethodsarefarfrom satisfactionwhenitcomestoengineeringapplications.Inthisstudy,abig
groupofdatasetsareobtainedbasedontheviscoelastic-plasticanalyticalmodelforsupportedtunnels ,
withsixdifferentkeyparametersofmaterialpropertiesandgeometryinformationofthesurroundingrocks
andsupportstructuresasthefeatureparameters.Subsequently,adatadrivenmodelforpredictingstead
ystatesupportforcesisdevelopedbyLightGBM machinelearningmethod.Theresultsindicatethatthe
modeldemonstratesstablepredictiveperformanceonthetestset,withcalibrationdeterminationcoeffi
cientsforboththetrainingandtestsetsexceeding0.9 ,andmeanabsolutepercentageerrorsbelow
4.2%,significantlyoutperformingothermachinelearningalgorithmssuchasXGBoostandSVR.Finally,
SHAPanalysisisappliedtoassesstherelationshipbetweeninputfeaturesandpredictedresults,further
enhancingthemodel ’sinterpretabilityandprovidingdeeperinsightsintofeaturecontributions.Insum
mary ,thedevelopeddatadrivenmodelcanrapidlypredictsteadystatesupportforces,anditcanbefur
therusedinthedesignofsupportstructuresandotherworksofinverseanalyses.
!"#$:20240605; %$:20240726
'()*: !"#$%&'()* (12272274)
+,-.: Â×!(2000—), :, !M0?, @ABCD, BC1éßeno。
/0+,: <"(1975—), V, ñTï0?, YA, \], YAD^_,EFGHßeno.ª«[ªbs£\1
}BC。