1、論文題目：基于測井數(shù)據(jù)的巖性識別方法研究專業(yè)：應用數(shù)學碩士生：李保霖(簽名)指導教師：龍熙華(簽名)摘要隨著測井技術(shù)的發(fā)展，綜合應用各種測井數(shù)據(jù)，成為更快速，更準確的獲得巖性信息最主要的途徑。巖性識別是測井數(shù)據(jù)解釋中最關鍵的一環(huán)。傳統(tǒng)的識別方法效率慢、精度低、人為因素大、不利于實際工程的應用。因此，構(gòu)造一種速度快，識別率高，泛化能力好的巖性自動識別法成為測井數(shù)據(jù)解釋領域迫切的需求。圍繞巖性識別的目的，本文主要完成三方面內(nèi)容：測井數(shù)據(jù)預處

2、理，測井相分析，巖性識別方法研究。測井數(shù)據(jù)預處理是在測井數(shù)據(jù)解釋之前，為消除非地層因素的影響而必要完成的基礎性工作；測井相分析是建立在多源信息融合的測井曲線自動分層基礎上，為構(gòu)建全區(qū)域測井相巖性數(shù)據(jù)庫而必須執(zhí)行的過程，它為巖性識別提供樣本選取的標準和依據(jù)；巖性識別方法研究針對三種神經(jīng)網(wǎng)絡：遺傳BP網(wǎng)絡，F(xiàn)CM_PNN網(wǎng)絡，遺傳優(yōu)化RBPNN網(wǎng)絡，進行了算法構(gòu)造和巖性識別應用對比。結(jié)果表明：三種網(wǎng)絡對巖性識別的應用均是可行的，但遺傳BP速

3、度較慢，F(xiàn)CM_PNN識別正確率相對較低，都不能完全達到實際工程的應用規(guī)范。為此，構(gòu)造全結(jié)構(gòu)遺傳優(yōu)化的RBPNN網(wǎng)絡模型。它通過采用遺傳算法搜索使得RBPNN訓練法誤差最小的最優(yōu)隱中心矢量，以及相匹配的核函數(shù)控制參數(shù)，優(yōu)化網(wǎng)絡結(jié)構(gòu)，增加網(wǎng)絡穩(wěn)定性。相對遺傳BP和FCM_PNN，遺傳優(yōu)化RBPNN在保證收斂速度和識別精度的同時能夠?qū)崿F(xiàn)巖性信息更快速和更準確的識別，且網(wǎng)絡結(jié)構(gòu)簡單，推廣泛化能力強，能為實際測井數(shù)據(jù)解釋領域所應用?；跍y井數(shù)據(jù)

4、的巖性識別方法研究為測井數(shù)據(jù)解釋提供了科學的方法支持和理論依靠，對整個油田地質(zhì)勘探領域具有重要的實際意義。關鍵詞：巖性識別；測井數(shù)據(jù)；BP神經(jīng)網(wǎng)絡；FCM_PNN神經(jīng)網(wǎng)絡；RBPNN神經(jīng)網(wǎng)絡研究類型：應用研究Subject:TheMethodsResearchofLithologyIdentificationBasedonLoggingDataSpecialty:ApplicationMathematicsName:LIBaolin(S

5、ignature)Instruct:LONGXihua(Signature)ABSTRACTWiththedevelopmentoftheloggingtechnologythecomprehensiveapplicationofallstsofloggingdatahasbecamethemainwayfgettinginfmationoflithologyrapidlyaccurately.Lithologyidentificati

6、onisthemostkeyringofloggingdatainterpretation.Thetraditionalmethodshaveslowefficiencylowaccuracyhumanfactssothatnotconducivetotheactualengineeringapplication.Therefeconstructingaspeedhighrecognitionratestronggeneralizati

7、onabilitylithologyidentificationmethodhasbeenthepressingneedsininterpretationfieldofloggingdatas.Aroundthepurposeoflithologyidentificationthispapermainlycompletesthreeareas:loggingdatapretreatmentlogfaciesanalysisthemeth

8、odsresearchoflithologyidentification.loggingdatapretreatmentisabasicwkbefecompletingofloggingdatainterpretationisfeliminatingtheinfluencesofunfmationfacts.Logfaciesanalysisbasedonloggingcurveslayeringofthemultipleinfmati

9、onfusionisanecessaryprocessfconstructinglogfacieslithologydatabaseinthewholeregional.ThemethodsresearchoflithologyidentificationcompletealgithmconstructionapplicationcontrastofthreeneuralwkswhichcontainGABPneuralwkFCM_PN

10、NneuralwkgeicoptimizedRBPNNneuralwk.Theresultsshow:theapplicationsofthreemethodsinlithologyidentificationareaarefeasiblebutthereexitslowspeedtoGABPrelativelylowrecognitionaccuracytoFCM_PNN.Thustheycannotachievethestardof

11、practicalengineeringapplication.SostructuringgeicoptimizedRBPNNwkmodel.ItissearchedftheoptimizedhiddencentervectsmatchingkernelfunctioncontrolparametersoftheRBPNNbyusingthegeicalgithmwhichmustbesatisfiedminimumerrofRBPNN

12、training.RelativetotheBPwkFCM_PNNwkThismethodcannotnolyguaranteeconvergencespeedtheaccuracybutalsocanidentifylithologyinfmationsmequicklymeaccuratly.ithassimplewkstructurestrongpromotiongeneralizationability.Thatmakeitca