遺傳算法適應(yīng)值曲面及遺傳算法困難度分析.pdf

1、天津大學(xué)博士學(xué)位論文遺傳算法適應(yīng)值曲面及遺傳算法困難度分析姓名：李建武申請學(xué)位級別：博士專業(yè)：管理科學(xué)與工程指導(dǎo)教師：李敏強20021001ABSTRACTGeneticalgorithms(GAs)，asakindofheuristicsearchandoptimizationtechniques，havebeenappliedtomanydomainssuccessfullyHoweverafter“NoFreeLunch”(NFL

2、)theoremswereproposed，applicationsofGAswcrcimpactedTherefore，t11estudyonGAhardnesshasbeenpaidmoreandmoreattentionInaddifiontheconcept，“fitnesslandscapes”，whichcarlaefrombiologyhasbeenappliedtoGAs，andbecameanimportanttool

3、todescribethecharacteristicsofthefitnessspaceandtoanNyzetheperformanceofGASUnderthiscircumstanceswediscussedtheprinciplesoffitnesslandscapesofGAsandanalyzed吐leGAhardnessbasedonfitnesslandscapesFurthermoreweproposedsevera

4、lmethodstomeasureGAhardnessortosolvesomedi佑cultproblemsforGAsThemainContentsofthisdissertationareasfollows：1WereviewedtheoriginanddevelopmentofGAssummarizedcharacteristicsofGAsanalyzedtheresearchstatusabouttheprinciplesa

5、ndapplicationsofGAsandfurtherpointedoutsomeproblems，whichneededtobesolvedcurrentlⅥMeanwhile，weprobedintotheoriginofthenotion“fitnesslandscapes”anddiscusseditsapplicationstatusWealSOanalyzedthebackgrounddevelopmentandcurr

6、entsituationofthestudyonGAhardness2WeinvestigatedtheprinciplesofthefitnesslandscapeanditsapplicationtoGAsindetailFirstlyfromtheviewpointofgraphtheoryweanalyzedfitnesslandscapes，anddescribedtherandomwalkcorrelationfunctio

7、n，thenwededucedtheequationtocomputethecorrelationlengthSecondlyweperformedthetimeseriesanalysisfortherandomwalkmodelonthefitnesslandscapetodiscovermoreinformationaboutthefitnesslandscape，anddemonstratedthemodelingprocess

8、basedontheNK—landscapes111irdlyweproposedthecone(：ptofthefitnesslandscapeofschemata，andimplementedstatisticalanalysisforfitnesslandscapesofschemataLastlythedynamicfitnesslandscapewasanalyzedinbrief3Wjidentifiedsomefactor

9、swhichmaycauseGAhardnessAtthebeginningwediscussedtheNFLtheoremsandtheirsignificanceofthestudyonGAhardnessSubsequentlyweconcludedsomeimportantdefinitionsandtheoremsforGAdeceptiveproblems，andanalyzedschemadeceptiveness’inf

10、luenceonGAhardnessThen，weinvestigatedtheepistasisofGAsusingWaIshSchematransformandfurtherevaluatedtheepistasisorderforthecontinuousfunctionoptimizationproblemsMeanwhileforthestatisticalanalysismodelofepistasisintrodueedb

11、yDavidorwediscussedepistasisvarianceandepistasiscorrelation，andfurtherinducedtwotheorems，ofwhichwegavestrictmathematicproofAdditionallyseveraltestproblemswithepistasisforGAsweredescribedFinallywestudiedsystematicallysome

12、otherreasonsforGA—hardness，suchasmulti—modalfunctions，themggednessoffitnesslandscapes，selectionofgeneticoperators，prematureproblems，controlofgeneticparameters，etc一4WeprobedintosomemethodstomeasureGAhardnessFirstofa11weco