地球物理反演中病态矩阵方程正则化解算方法研究_精品文档资料下载.pdf
《地球物理反演中病态矩阵方程正则化解算方法研究_精品文档资料下载.pdf》由会员分享,可在线阅读,更多相关《地球物理反演中病态矩阵方程正则化解算方法研究_精品文档资料下载.pdf(86页珍藏版)》请在冰豆网上搜索。
成都理工大学博士学位论文地球物理反演中病态矩阵方程正则化解算方法研究姓名:
@#@王文娟申请学位级别:
@#@博士专业:
@#@地球探测与信息技术指导教师:
@#@曹俊兴;@#@谭永基20100501摘要I地球物理反演中病态矩阵方程正则化解算方法研究地球物理反演中病态矩阵方程正则化解算方法研究作者简介:
@#@王文娟,女,1970年05月生,师从成都理工大学曹俊兴教授,复旦大学谭永基教授,2010年06月毕业于成都理工大学地球探测与信息技术专业,获得工学博士学位摘摘要要地球物理反演是地球物理探测数据最重要的解释方法技术求解地球物理反演问题常会涉及到大型病态矩阵方程的求解理论上讲正则化方法是处理病态问题的有效手段,但在实践上正则化参数的选择却是一个困难的问题本文在比较系统地研究了正则化方法的基础上,针对实际计算中常会碰到的问题,将其与Active-Set算法、差分进化算法等相结合,发展了一些新的病态矩阵方程正则化解算方法论文的主要内容包括:
@#@1研究了在实际反演中遇到参数有非负要求特性时的反演计算方法将原问题转化为一个带非负约束的阻尼最小二乘问题,并用Active-Set算法求解通过对理论模型进行数值模拟计算,验证了将Tikhonov正则化方法与Active-Set算法相结合的A-TR算法的有效性应用到实际双频电导率成像反演,也取得了满意的结果2研究了差分进化算法在地球物理反演中的几种应用为加速差分进化算法的收敛速度,提出了将种群熵的自适应差分进化(ARDE)算法以及粒子群差分进化混合(PSODE)算法分别与Tikhonov正则化方法结合在大型反演计算中,这两种方法可以在不影响反演效果的前提下,不同程度地提高收敛速度,降低时间成本同时结合LSQR和差分进化算法的优点,提出了基于LSQR算法的差分混合(HDE)算法,避开了Tikhonov和TSVD等直接正则化算法在选取正则化参数上的困难,同时具有数值稳定性好、不依赖于初值、不易陷入局部极值和收敛速度快等优点,适宜于在正则化参数选取困难情况时的地球物理反演问题的求解3提出了一种双参数混合正则化方法引入了带有二阶正则算子的正则化项,并应用L-曲线法、偏差原理和广义交叉校验准则的优化组合确定了最佳正成都理工大学博士学位论文II则化参数数值模拟实验和实际数据处理实验结果表明了该方法的可行性这是一种将高阶正则化算子应用于实际反演计算的新的尝试基于数值模拟实验和实际数据处理实验,认为研究发展的A-TR算法、HDE等算法各有其不同的适用条件,A-TR算法适用于求解反演参数有非负约束的情况,而当正则化参数选取困难时,可采用HDE算法针对本文所考察的双频电导率反演问题,由于电导率的非负性,采用A-TR方法可得到更加精细可靠的重建图像关键词:
@#@地球物理反演正则化方法差分进化算法正则化参数AbstractIIIRegularizationAlgorithmsforSolvingIll-posedMatrixEquationsArisingfromGeophysicalInversionsIntroductionoftheauthor:
@#@Wangwenjuan,female,wasborninMay,1970whosetutorwasProfessorCaoJunxingandProfessorTanYongji.ShegraduatedfromChengduUniversityofTechnologyinEarthExploration&@#@InformationTechniquesmajorandwasgrantedtheDoctorDegreeinJune,2010.AbstractGeophysicalinversionisakeytechniqueingeophysicalexploration.Geophysicalinversionoftenrelatestosolvingill-posed,largematrixequations.Theoretically,regularizationisaneffectivemethodindealingwithill-posedproblems.Itsapplicationinpracticalgeophysicalinversionproblems,however,stillhasmanydifficultiesinselectingregularizationparameters.Basedonsystematicstudyofregularization,thisdissertationpresentsafewnewlydevelopedregularizationmethodsthatapplyActive-Setalgorithm,DifferentialEvolutionalgorithm,andafewothers.Theseregularizationmethodsfocusonsolvingill-posedmatrixequationsarisingfrompracticalproblems.Mainresultsofthedissertationinclude:
@#@TikhonovregularizationandActive-Setalgorithmareappliedtogethertothegeophysicalinversionproblemssothattheproblemswithnon-negativeparametersareconvertedintoproblemsofnon-negativedampedleastsquarealgorithm,whichcanbefurthersolvedbytheActive-Setalgorithm.Theimprovedrecursivealgorithmisfurtherverifiedbynumericalsimulation.Satisfactoryresultsareobtainedbyapplyingthisalgorithmtoelectricalconductivityimageryinversion.Furthermore,differentialevolutionalgorithmsarealsostudied.ToimprovetherateofconvergenceofDifferentialEvolutionalgorithms,twonewTikhonovregularizationalgorithmsareproposedthatrespectivelyemployAdaptiveRecursiveDifferentialEvolution(ARDE)algorithmbasedonpopulationentropyandParticleSwarmOptimizationandDifferentialEvolution(PSODE)algorithm.Withoutanycompromiseineffectiveness,thesetwoalgorithmsbothimprovetheconvergencespeedandthusreducecomputationcost.Stillfurther,anewDEalgorithmbasedonLSQS,whichinheritstheadvantagesofbothLSQRandDE,isdesigned.ThisnewalgorithmavoidsthecommondifficultyofregularizationparameterselectioninTikhonovandTSVDalgorithms.Italsodisplayssuperiorstability,independenceoninitialvalues,unlikelihoodoflocalextrema,andfasterconvergence.Thisalgorithmisparticularlysuitabletosolvetheproblemofselectionofregularizationparametersinthestudyofgeophysicalinversion.Finally,theselectionmethodofregularizationtermisalsostudied.Aregularizationtermwithasecondorderregularizationoperatorisintroducedto成都理工大学博士学位论文IVproposeamixedregularizationmethodwithtwoparameters.TheL-curvecriterion,discrepancyprinciple,andgeneralizedcross-validationareappliedtodeterminetheoptimalvalueoftheregularizationparameter.Thevalidityandsuperiorityoftheproposedmethodisverifiedbynumericalsimulationofthetheoreticalmodel.Basednumericsimulationsandresultsfromactualdataprocessing,itisfoundthatthenewlydevelopedA-TRregularizationalgorithm,HDEregularizationalgorithmareeffectiveincertainconditions.TheA-TRregularizationalgorithmisapplicabletoinversionproblemsthatrequirenon-negativeparameters,whereasHDEregularizationalgorithmisapplicabletothoseinversionproblemswhoseselectionofregularizationparametersisdifficult.Becauseofnon-negativeconductivity,A-TRalgorithmwillgetmoredetailedandreliableimageryfordual-frequencyconductivityinversionproblemsinvestigatedbythispaper.Keywords:
@#@Geophysicalinversion;@#@Regularizationmethod;@#@Differentialevolutionalgorithm;@#@Regularizationparameter.独创性声