地下結(jié)構(gòu)可靠度理論與應(yīng)用研究.pdf

1、同濟大學(xué)博士后學(xué)位論文地下結(jié)構(gòu)可靠度理論與應(yīng)用研究姓名：鄧建申請學(xué)位級別：博士后專業(yè)：結(jié)構(gòu)工程指導(dǎo)教師：朱合華20020501AbstractTheaimofthereportistopresentuniversalprobabilistiemodelsofundergroundengineeringstructureparametersandtopresentreliabilityanalysismethodsoHperformanc

2、efunctionwithcomplexityandhigherordernonlinearityingeoteehnicalengineeringThisnowmethodsfmdapplicationinreliabilityanalysisofhighwaytunnelliningstructures11位maincontentsale：1Universalprobabilistiemodelsofundergroundengin

3、eeringslructuroparametersare：(1)Estimationofprobabilisticmodelsusingsimplepolynomials；(b)EstimationofprobabilisticmodelsusingLegendrepolynomials；(曲Estimationofprobabilisticmodelsusingthemaxilnu／nentropyprincipleSamplemom

4、entsabouttheoriginwereusedtodeducetheuniversalprobabilistiodeusityfunctionProbabilityweightedmomentsofsampleswereusedtodeducetheuniversalinversecumulativedensityfunction2Variousreliabilityanalysismethodsareavailableforla

5、rgeandcomplexengineeringstructuresHowever，whoathepⅡ‰ancefunctionsareimplicitespeciallyinouseofthefiniteelementmethodinvolvedtraditionalreliabilityanalysismethodsmaybedifficultThispaperpresentsan01vmethodusingneuralnetwor

6、ks山atcouldbeusedtocalculateroliabilityofcomplexslructures1】vitIIimplicitperformancefunctionsTheproposedneuralnetworkbascdreliabilityanalysismetho正calledFEMANNMCSistocoustngtaneuralnetworkmodeltoapproximatetheimplicitperf

7、ormacefunctionThelearningandgeneralizationcapabilitiesofmultiplelayerfeedforwardnetworksaresobustandastoundingthatitiseasytoobtainthousandsormillionsofperformⅡenfunctionvaluesandthentoperformreliabilityanalysisRigorousth

8、eoreticalverificationofthenewapproachisputforwardandthealgorithmisintroduce也Toinvestigatethesuitahil]ityofthisapproachtheresultsthenewapproacha∞comparedtoonesobtainedbyconventionalreliabilitymethodssuch∞usualMonto—Carlos

9、imulationresponsesurfaceeto3IlIJCmethodthelimitstateequationneedstobesolvedtofindthenowdesignpointThismaybedifficultintheeaseofcomplicatednonlinearperformancefunctionEspeciallyinundergroundengineeringtheperformancemaynot

10、ovenbeavailableinclosedformAnalternativeNewtonRaphsonreenrsivealgorithm，referredtoas月。!剛訂Memod2，isusedtofindthedesignpointItusesthederivativestofindthenextiterationpointMultilayerfeedforwardnetworkswitlI∞fewasasinglehidd

11、enlayerandanappropriatelysmoothhiddenlayeractivationfunctionarecapableofarbitrarilyaccurateapproximationto礬arbitraryfunctionenditsderivativesThisreportistouseaneuralnetworkmodeltoapproximatetheimplicitperformancefunction

12、andthederivativesThenFORMMethod2isusedtodeliabilityanalysisTIIisnewmethodiscalledFE^不rtMv：FD砌f巧DR^Z4MaZaldinghighwaytunnelisatwowayunsymmetricalarcadetunnel，whichisrareinChinaTongiiSugnengsoftware(GeoFBA2D)isappliedtosim

13、ulatetheconstructionprocessThecalculatedresultsandtheinsitumeasureddataagocomparedtotestifythenumericalsimulationAmodelforreliabilityanalysisoftunnellining，whichisfollowedoutthetunnelspecificationporformednowinChinawaspe

14、rformedReliabilityanalysisanddesignoftheliningwerecarriedoutbyuseoftheproposedmethodKeyWardsReliabilityanalysisProbabilityweightedmomentFiniteelementmethod(E歐If)，Artificialneurulnetwork(ANN)，Montocarlosimulation(MCS)tiig