英文資料EVOLUTIONOFASTRATEGYFORSHIPGUIDANCEUSINGTWOIMPLEMENTATIONSOFGENETICPROGRAMMING1THEPRINCIPLEOFGENETICALGORITHMAGENETICALGORITHMGAISASEARCHTECHNIQUEUSEDINCOMPUTINGTOFINDEXACTORAPPROXIMATESOLUTIONSTOOPTIMIZATIONANDSEARCHPROBLEMSGENETICALGORITHMSARECATEGORIZEDASGLOBALSEARCHHEURISTICSGENETICALGORITHMSAREAPARTICULARCLASSOFEVOLUTIONARYALGORITHMSEATHATUSETECHNIQUESINSPIREDBYEVOLUTIONARYBIOLOGYSUCHASINHERITANCE,MUTATION,SELECTION,ANDCROSSOVERALSOCALLEDRECOMBINATIONGENETICALGORITHMSAREIMPLEMENTEDINACOMPUTERSIMULATIONINWHICHAPOPULATIONOFABSTRACTREPRESENTATIONSCALLEDCHROMOSOMESORTHEGENOTYPEOFTHEGENOMEOFCANDIDATESOLUTIONSCALLEDINDIVIDUALS,CREATURES,ORPHENOTYPESTOANOPTIMIZATIONPROBLEMEVOLVESTOWARDBETTERSOLUTIONSTRADITIONALLY,SOLUTIONSAREREPRESENTEDINBINARYASSTRINGSOF0SAND1S,BUTOTHERENCODINGSAREALSOPOSSIBLETHEEVOLUTIONUSUALLYSTARTSFROMAPOPULATIONOFRANDOMLYGENERATEDINDIVIDUALSANDHAPPENSINGENERATIONSINEACHGENERATION,THEFITNESSOFEVERYINDIVIDUALINTHEPOPULATIONISEVALUATED,MULTIPLEINDIVIDUALSARESTOCHASTICALLYSELECTEDFROMTHECURRENTPOPULATIONBASEDONTHEIRFITNESS,ANDMODIFIEDRECOMBINEDANDPOSSIBLYRANDOMLYMUTATEDTOFORMANEWPOPULATIONTHENEWPOPULATIONISTHENUSEDINTHENEXTITERATIONOFTHEALGORITHMCOMMONLY,THEALGORITHMTERMINATESWHENEITHERAMAXIMUMNUMBEROFGENERATIONSHASBEENPRODUCED,ORASATISFACTORYFITNESSLEVELHASBEENREACHEDFORTHEPOPULATIONIFTHEALGORITHMHASTERMINATEDDUETOAMAXIMUMNUMBEROFGENERATIONS,ASATISFACTORYSOLUTIONMAYORMAYNOTHAVEBEENREACHEDGENETICALGORITHMSFINDAPPLICATIONINBIOINFORMATICS,GENETICS,COMPUTATIONALSCIENCE,ENGINEERING,ECONOMICS,CHEMISTRY,MANUFACTURING,MATHEMATICS,PHYSICSANDOTHERFIELDSASTANDARDREPRESENTATIONOFTHESOLUTIONISASANARRAYOFBITSARRAYSOFOTHERTYPESANDSTRUCTURESCANBEUSEDINESSENTIALLYTHESAMEWAYTHEMAINPROPERTYTHATMAKESTHESEGENETICREPRESENTATIONSCONVENIENTISTHATTHEIRPARTSAREEASILYALIGNEDDUETOTHEIRFIXEDSIZE,WHICHFACILITATESSIMPLECROSSOVEROPERATIONSVARIABLELENGTHREPRESENTATIONSMAYALSOBEUSED,BUTCROSSOVERIMPLEMENTATIONISMORECOMPLEXINTHISCASETREELIKEREPRESENTATIONSAREEXPLOREDINGENETICPROGRAMMINGANDGRAPHFORMREPRESENTATIONSAREEXPLOREDINEVOLUTIONARYPROGRAMMINGTHEFITNESSFUNCTIONISDEFINEDOVERTHEGENETICREPRESENTATIONANDMEASURESTHEQUALITYOFTHEREPRESENTEDSOLUTIONTHEFITNESSFUNCTIONISALWAYSPROBLEMDEPENDENTFORINSTANCE,INTHEKNAPSACKPROBLEMONEWANTSTOMAXIMIZETHETOTALVALUEOFOBJECTSTHATCANBEPUTINAKNAPSACKOFSOMEFIXEDCAPACITYAREPRESENTATIONOFASOLUTIONMIGHTBEANARRAYOFBITS,WHEREEACHBITREPRESENTSADIFFERENTOBJECT,ANDTHEVALUEOFTHEBIT0OR1REPRESENTSWHETHERORNOTTHEOBJECTISINTHEKNAPSACKNOTEVERYSUCHREPRESENTATIONISVALID,ASTHESIZEOFOBJECTSMAYEXCEEDTHECAPACITYOFTHEKNAPSACKTHEFITNESSOFTHESOLUTIONISTHESUMOFVALUESOFALLOBJECTSINTHEKNAPSACKIFTHEREPRESENTATIONISVALID,OR0OTHERWISEINSOMEPROBLEMS,ITISHARDOREVENIMPOSSIBLETODEFINETHEFITNESSEXPRESSIONINTHESECASES,INTERACTIVEGENETICALGORITHMSAREUSEDONCEWEHAVETHEGENETICREPRESENTATIONANDTHEFITNESSFUNCTIONDEFINED,GAPROCEEDSTOINITIALIZEAPOPULATIONOFSOLUTIONSRANDOMLY,THENIMPROVEITTHROUGHREPETITIVEAPPLICATIONOFMUTATION,CROSSOVER,INVERSIONANDSELECTIONOPERATORSANTCOLONYOPTIMIZATIONACOUSESMANYANTSORAGENTSTOTRAVERSETHESOLUTIONSPACEANDFINDLOCALLYPRODUCTIVEAREASWHILEUSUALLYINFERIORTOGENETICALGORITHMSANDOTHERFORMSOFLOCALSEARCH,ITISABLETOPRODUCERESULTSINPROBLEMSWHERENOGLOBALORUPTODATEPERSPECTIVECANBEOBTAINED,ANDTHUSTHEOTHERMETHODSCANNOTBEAPPLIEDBACTERIOLOGICALGORITHMSBAINSPIREDBYEVOLUTIONARYECOLOGYAND,MOREPARTICULARLY,BACTERIOLOGICADAPTATIONEVOLUTIONARYECOLOGYISTHESTUDYOFLIVINGORGANISMSINTHECONTEXTOFTHEIRENVIRONMENT,WITHTHEAIMOFDISCOVERINGHOWTHEYADAPTITSBASICCONCEPTISTHATINAHETEROGENEOUSENVIRONMENT,YOUCANTFINDONEINDIVIDUALTHATFITSTHEWHOLEENVIRONMENTSO,YOUNEEDTOREASONATTHEPOPULATIONLEVELBASHAVESHOWNBETTERRESULTSTHANGASONPROBLEMSSUCHASCOMPLEXPOSITIONINGPROBLEMSANTENNASFORCELLPHONES,URBANPLANNING,ANDSOONORDATAMININGCROSSENTROPYMETHODTHECROSSENTROPYCEMETHODGENERATESCANDIDATESSOLUTIONSVIAAPARAMETERIZEDPROBABILITYDISTRIBUTIONTHEPARAMETERSAREUPDATEDVIACROSSENTROPYMINIMIZATION,SOASTOGENERATEBETTERSAMPLESINTHENEXTITERATIONCULTURALALGORITHMCACONSISTSOFTHEPOPULATIONCOMPONENTALMOSTIDENTICALTOTHATOFTHEGENETICALGORITHMAND,INADDITION,AKNOWLEDGECOMPONENTCALLEDTHEBELIEFSPACEEVOLUTIONSTRATEGIESEVOLVEINDIVIDUALSBYMEANSOFMUTATIONANDINTERMEDIATEANDDISCRETERECOMBINATIONESALGORITHMSAREDESIGNEDPARTICULARLYTOSOLVEPROBLEMSINTHEREALVALUEDOMAINTHEYUSESELFADAPTATIONTOADJUSTCONTROLPARAMETERSOFTHESEARCHEVOLUTIONARYPROGRAMMINGEPINVOLVESPOPULATIONSOFSOLUTIONSWITHPRIMARILYMUTATIONANDSELECTIONANDARBITRARYREPRESENTATIONSTHEYUSESELFADAPTATIONTOADJUSTPARAMETERS,ANDCANINCLUDEOTHERVARIATIONOPERATIONSSUCHASCOMBININGINFORMATIONFROMMULTIPLEPARENTSGENETICALGORITHMISPUTFORWARDBYJOHNHOLLAENDERANDHISCOLLEAGUEINMICHIGANUNIVERSITYIN1960SWHENCONDUCTSTHERESEARCHTOTHECELLULARAUTOMATONEXTREMALOPTIMIZATIONEOUNLIKEGAS,WHICHWORKWITHAPOPULATIONOFCANDIDATESOLUTIONS,EOEVOLVESASINGLESOLUTIONANDMAKESLOCALMODIFICATIONSTOTHEWORSTCOMPONENTSTHISREQUIRESTHATASUITABLEREPRESENTATIONBESELECTEDWHICHPERMITSINDIVIDUALSOLUTIONCOMPONENTSTOBEASSIGNEDAQUALITYMEASURE“FITNESS“THEGOVERNINGPRINCIPLEBEHINDTHISALGORITHMISTHATOFEMERGENTIMPROVEMENTTHROUGHSELECTIVELYREMOVINGLOWQUALITYCOMPONENTSANDREPLACINGTHEMWITHARANDOMLYSELECTEDCOMPONENTTHISISDECIDEDLYATODDSWITHAGATHATSELECTSGOODSOLUTIONSINANATTEMPTTOMAKEBETTERSOLUTIONSTHEWEALTHMAGAZINETOP500ENTERPRISESWILLUSEITTOCARRYONTHETIMETABLEARRANGEMENT,THEDATAANALYSIS,THEFUTURETENDENCYPREDICTTHATTHEBUDGET,ASWELLASTHESOLUTIONTOMANYOTHERCOMBINATIONOPTIMIZATIONQUESTION2THEMAINCONTENTSOFTHETHESISINTHISPAPERTHEIMPLEMENTATIONOFGENETICPROGRAMMINGGPTOOPTIMISEACONTROLLERSTRUCTUREFORASUPPLYSHIPISASSESSEDGPISUSEDTOEVOLVECONTROLSTRATEGIESFORMANOEUVRINGTHESHIPTHEOPTIMISEDCONTROLLERSAREEVALUATEDTHROUGHCOMPUTERSIMULATIONSANDREALMANOEUVREABILITYTESTSINAWATERBASINLABORATORYINORDERTODEALWITHTHEISSUEOFGENERATIONOFNUMERICALCONSTANTS,TWOKINDSOFGPALGORITHMSAREIMPLEMENTEDTHEFIRSTONECHOOSESTHECONSTANTSNECESSARYTOCREATETHECONTROLSTRUCTUREBYRANDOMGENERATIONTHESECONDALGORITHMINCLUDESAGENETICALGORITHMGAFORTHEOPTIMISATIONOFSUCHCONSTANTSTHERESULTSOBTAINEDILLUSTRATETHEBENEFITSOFUSINGGPTOOPTIMISEPROPULSIONANDNAVIGATIONCONTROLLERSFORSHIPSINORDERTOENSURETHESAFENAVIGATIONOFSURFACEVESSELSTHEIRMOTIONIENAVIGATIONANDPROPULSIONCAPABILITIESHASTOBECONTROLLEDACCURATELYTHISCANBEACHIEVEDTHROUGHTHEDESIGNANDIMPLEMENTATIONOFAUTOMATICCONTROLSYSTEMSTHEPERFORMANCEOFTHECONTROLTECHNIQUESDEPENDSNOTONLYONTHECONTROLSTRUCTUREBUTALSOONTHEVALUESOFTHECONTROLLERSPARAMETERSCONVENTIONALLY,THESEPARAMETERSAREMANUALLYTUNEDBYTHEDESIGNERTHISRELIESONANADHOCAPPROACHTOTUNING,WHICHDEPENDSONTHEEXPERIENCEOFTHEDESIGNERASOLUTIONTOTHISPROBLEMWIDELYUSEDINTHEFIELDOFCONTROLENGINEERINGISTOUSEEVOLUTIONARYOPTIMISATIONTECHNIQUESSUCHASGENETICALGORITHMSGASTHATTUNESUCHPARAMETERSAUTOMATICALLYHOWEVER,GASAREPARAMETEROPTIMISERSANDINTHEMAJORITYOFCASESDONOTVARYTHESTRUCTUREOFTHEOPTIMISINGSUBJECTINTHECONTEXTOFCONTROLLEROPTIMISATIONTHEYAREPRESENTEDWITHTHESTRUCTUREOFAPARTICULARCONTROLMETHODOLOGYANDVARYTHEASSOCIATEDPARAMETERSTOOBTAINTHEDESIREDPERFORMANCEFORTHESYSTEMGENETICPROGRAMMINGGPEVOLVESCANDIDATESOLUTIONSWITHOUTSPECIFYINGAPRIORITHEIRSIZE,SHAPEORSTRUCTUREBYUSINGGP,THEOPTIMISATIONPROBLEMOFFINDINGANEAROPTIMALCONTROLLERISTAKENASTEPFORWARDINTHATTHESTRUCTUREOFTHEWHOLECONTROLLERISOPTIMISEDANDNOTONLYTHEPARAMETERSTHATDEFINESUCHASTRUCTURETHEPARTICULARAPPLICATIONUSEDINTHISRESEARCHISASCALEMODELOFANOILPLATFORMSUPPLYSHIPCALLEDCYBERSHIPIICS2THEOPTIMISATIONPROBLEMFORTHEGPISTOPROVIDEACONTROLSTRATEGYTHATGOVERNSTHEHEADINGANDPROPULSIONDYNAMICSOFCS2THEGPOPTIMISATIONOFTHECONTROLSTRUCTURESHASBEENCONDUCTEDTHROUGHCOMPUTERSIMULATIONSINMATLABUSINGAMATHEMATICALMODELOFCS2THEOPTIMIZEDCONTROLLERSHAVEBEENIMPLEMENTEDANDTESTEDONTHEPHYSICALMODELOFCS2INORDERTODEALWITHTHEISSUEOFTHEGENERATIONOFNUMERICALCONSTANTS,TWOKINDSOFGPALGORITHMSHAVEBEENIMPLEMENTEDTHEFIRSTONECHOOSESTHECONSTANTSNECESSARYTOCREATETHECONTROLLERSTRUCTUREBYRANDOMGENERATIONGPRGTHESECONDGPALGORITHMINCLUDESAGATECHNIQUEFORTHEOPTIMISATIONOFSUCHCONSTANTSGPGATHERESULTSOBTAINEDFROMBOTHMETHODSAREPRESENTEDANDCOMPARED3CYBERSHIP2THECONTROLSUBJECTUSEDINTHISWORKISCS2,WHICHISASCALEMODELSCALE1/70THAPPROXOFANOILPLATFORMSUPPLYSHIPTHISTESTVESSELHASBEENDEVELOPEDATTHEMARINECYBERNETICSLABMCLABATTHENORWEGIANUNIVERSITYOFSCIENCEANDTECHNOLOGYNTNUINTRONDHEIM,NORWAYTHEMCLABISAPURPOSEBUILTEXPERIMENTALLABORATORYFORTESTINGOFSHIPSANDUNDERWATERVEHICLESPRIORTOTHEREALTESTINGTHENONLINEARHYDRODYNAMICMODELOFTHEVESSELHASBEENUSEDFORTHESIMULATIONSOFTHEDESIGNSTAGETHESEARETHEINPUTSTOCS2THATAREUSEDTOCONTROLITSMOTIONINORDERTOCREATEAMOREREALISTICENVIRONMENT,WINDGENERATEDWAVESARESIMULATEDDURINGTHEMANOEUVRESUSEDTOEVALUATEEACHTREEDURINGTHEOPTIMISATIONTHESEARETHEMOSTRELEVANTDISTURBANCESEXPERIENCEBYSURFACEVESSELSANDTHEYCANBEREALISTICALLYREPRODUCEDINTHEMCLABDURINGTESTSTHEMODELTHATHASBEENUSEDTOSIMULATETHEWAVESACTIONONTHEVESSELDERIVESFROMTHEFORCESANDMOMENTSINDUCEDBYAREGULARSEAONABLOCKSHAPEDSHIP4REALIZINGPROCESSSELECTIONINTHISRESEARCHTOURNAMENTSELECTIONHASBEENUSEDCROSSOVERINTHISWORKSUBTREECROSSOVERHASBEENUSEDANDTHEPROBABILITYOFCROSSOVERHASBEENCHOSENTOBE80DUETOTHESATISFACTORYRESULTSOBTAINEDWITHTHISPROBABILITYINASTUDYCOMPARINGTHEPERFORMANCEOFVARIOUSCROSSOVERPROBABILITIESANDMUTATIONPROBABILITIESPRESENTEDMUTATIONTHETREESTRUCTUREOFGPSOLUTIONSALLOWSAVARIETYOFMUTATIONOPERATORSINTHISSTUDYACOMBINATIONOFTWOMETHODSISEMPLOYEDIESUBTREEMUTATIONANDPOINTMUTATIONMUTATIONOCCURSWITHAPROBABILITYOF01ONCEATREEISCHOSENFORMUTATION,THEPROBABILITYOFUNDERGOINGSUBTREEORPOINTMUTATIONIS05THEGPALGORITHMUSEDINTHISPAPERISCODEDINMATLABTHEDIFFICULTYOFCODINGGPINMATLABLIESINTHELACKOFPOINTERSTHISREQUIRESADIFFERENTCODINGAPPROACHTHEWHOLEPOPULATIONISSTOREDINACELLARRAY,EVERYCELLSTORINGONEINDIVIDUALTHE5TRE7ESTRUCTUREISREPRESENTEDBYAMATRIXSEEFIG2INWHICHTHENUMBEROFROWSISTHENUMBEROFINTERNALNODESANDITISEVOLVEDALONGTHEGPGENERATIONSEVERYINTERNALNODEISENCODEDASA1X5VECTORSEEFIG1NODENUMBERNODETYPE1STARGUMENTFUNCTION2NDARGUMENTFIG1INTERNALNODEREPRESENTATIONTHEFIRSTELEMENTINTHENODEVECTORISTHENODENUMBERTHESECONDELEMENTDISTINGUISHESIFTHEARGUMENTSOFTHEFUNCTIONARETERMINALNODESORINTERNALNODESFOREXAMPLE,AVALUEOF0INDICATESTHATBOTHARGUMENTSARETERMINALNODESANDAVALUEOF1/2INDICATESTHATTHE1ST/2NDARGUMENTISANINTERNALNODEALLTHEINTERNALNODESHAVEARITY1OR2THETHREELASTELEMENTSPROVIDETHEARGUMENTSOFTHEFUNCTIONSINCOLUMNS3AND5ANDTHEFUNCTIONITSELFINCOLUMN4IFANARGUMENTISATERMINAL,ITISINCLUDEDINTHECORRESPONDENTPOSITIONOTHERWISETHENUMERICVALUEREFERSTOTHENUMBEROFTHEINTERNALNODETHATISROOTEDTHEREX265501611X22223FIG2MATRIXREPRESENTATIONOFATREETHETREESTRUCTURENEEDSTOBEFLATTENEDSOEVERYINTERNALNODEISASSIGNEDANUMBERTHENODESOFTHETREEARECOUNTEDFROMLEFTTORIGHT,UPWARDSANODEISNOTCOUNTEDUNTILALLTHENODESOFTHESUBTREESROOTEDINITARECOUNTEDEVERYSOLUTIONTOTHESTATEDCONTROLPROBLEMCONSISTSOFTWOINDEPENDENTTREESONEFORHEADINGCONTROLANDOTHERFORPROPULSIONCONTROLIEDECOUPLEDCONTROLLERSFORTHISAPPLICATIONTHETERMINALSETCONSISTSOF4COMMONTERMSERROR,STATE,REFERENCEANDONENUMERICALCONSTANTASSHOWNINTABLE1TABLE1TERMINALSETSFORPROPULSIONANDHEADINGPROPULSIONHEADINGSARGEERRORHEADINGERRORAWA2W2RRTHEPROBABILITYOFGENERATINGANUMERICALCONSTANTIS05SINCETHENUMBEROFNUMERICALCONSTANTSREQUIREDTOCREATEACONTROLSTRUCTUREISLARGERTHANTHENUMBEROFVARIABLESTABLE2FUNCTIONSETTHEFUNCTIONSETINCLUDESFUNCTION2ARGUMENTFUNCTIONSARG1ARG2ARG1ARG2ARG1ARG2ARG1/ARG2ARG1TANHHXX0/ARG21ARGUMENTFUNCTIONSARGDTARG/DTSINEXPARGSIGNHXX01/2ARGUMENTFUNCTIONSARG0,1,R2PLCETHEPLACECOMMANDRETURNSTHEVALUEK_X,WHEREXISASDEFINEDBEFOREANDKISTHEFEEDBACKVECTOROBTAINEDBYEXECUTINGPLACE0,ARG1,ARG2INTHEHEADINGCONTROLORPLACEARGINTHEPROPULSIONCONTROLINORDERTOENSURETHATTHECLOSUREPROPERTYISMET,THEPOLESTOBEASSIGNEDBYTHEPLACECOMMANDAREALWAYSREALNUMBERSANDSOMEOFTHEFUNCTIONSHAVEAPROTECTIONMECHANISMTHUS,THEHYPERBOLICTANGENTRETURNSARG1WHENARG2IS0ANDTHEPLACECOMMANDRETURNS0IFTHEREISANYERRORFLAGEGIFTHEPOLESARETOOCLOSEINTHEFIRSTIMPLEMENTATIONEVERYTIMETHERANDOMCONSTANTRINTHESETOFTERMINALSISCHOSENARANDOMNUMBERISGENERATEDANDASSOCIATEDWITHTHATTERMINALNODETHEGPSHOULDBEABLETOGENERATEOTHERCONSTANTSNEEDEDBYUSINGARITHMETICOPERATIONSASOPPOSEDTOKOZASGPTHATDOESNOTUSEMUTATION,INTHISWORKPOINTMUTATIONHASBEENINCLUDEDASANOPERATORTHISENABLESTHEGPTOMODIFYTHETERMINALVALUESTHUS,ANUMERICALCONSTANTCANCHANGEITSVALUEANDATERMINALOCCUPIEDBYAVARIABLECANBEMUTATEDINTOANUMERICALCONSTANTVARIOUSAUTHORSHAVEPOINTEDOUTTHATTHERANDOMGENERATIONOFNUMERICALCONSTANTSISNOTAVERYEFFICIENTWAYOFCREATINGNEWCONSTANTS6,15,16THEMAINDRAWBACKOFTHISAPPROACHISTHATTHENUMBEROFCONSTANTSDEPENDSTOTALLYONTHEINITIALISATIONOFTHETREESVARIOUSAPPROACHESCANBEFOUNDINTHELITERATURETHATADDRESSTHISISSUEIN6THEAUTHORSCOMBINEDAGPWITHAGAFORTHETUNINGOFTHENUMERICPARAMETERSINTHEGPTREETHEYASSOCIATEAGALIKEFIXEDLENGTHCHROMOSOMETHATREPRESENTSTHENUMERICALVALUESOFTHESOLUTION,ALTHOUGHTHEYMAYORMAYNOTBEPRESENTINTHETREETHECHROMOSOMEISEVOLVEDTOGETHERWITHTHETREEANDISSUBMITTEDTOCROSSOVERANDMUTATIONTHEMAINPROBLEMOFTHISAPPROACHISTHATTHEFIXEDLENGTHOFTHECHROMOSOMEDETERMINESTHEMAXIMUMNUMBEROFNUMERICALCONSTANTSTHATCANBEFOUNDINTHETREETHISREQUIRESAPRIORIKNOWLEDGEOFTHESOLUTIONALSO,IFTHECHROMOSOMEISMADETOBEVERYLONGJUSTTOACCOUNTFORANYADDITIONALCONSTANT,THELENGTHOFTHECHROMOSOMEHAMPERSTHECORRECTEVOLUTIONOFSOLUTIONSANDINCREASESTHECOMPUTATIONALCOSTINTHISWORK,THESECONDGPALGORITHMTESTEDUSESAGAASAPARAMETRICOPTIMISATIONTECHNIQUETHEAIMISTHATTHEGPGAALGORITHMPROVIDESABETTERPARAMETERTUNINGANDBETTERRESULTSTHEGPGAMETHODUSEDISBASICALLYDIFFERENTFROMTHEMECHANISMPRESENTEDBYINSTEADOFASSOCIATINGAGACHROMOSOMEWITHAGPTREEANDEVOLVINGTHEMTOGETHER,GPGACOMBINESAGPEVOLUTIONPROCESSWITHAGALEARNINGPROCESS,IEEVERYTIMEATREEISEVALUATEDAMINIGAISRUNTOOPTIMIZETHEVALUESOFTHENUMERICALCONSTANTSPRESENTINTHATTREEWITHTHISAPPROACHTHEMAXIMUMNUMBEROFCONSTANTSINTHETREEDOESNOTNEEDTOBEFIXEDANDONLYTHOSECONSTANTSTHATAREINTHETREEAREENCODED,REDUCINGTHESIZEOFTHECHROMOSOMESTHEGPGAHASBEENCODEDSOTHATTHETOTALNUMBEROFTREEEVALUATIONSISTHESAMEASINTHEGPRGCASE,PROVIDINGAGOODBASISFORCOMPARISONTHENUMBEROFTREESINTHEPOPULATIONUSEDINGPRGIS120ANDTHENUMBEROFGENERATIONSIS31INORDERTOGETTHESAMENUMBEROFEVALUATIONSFORTHEGPGAOPTIMISATION,THEGPHASAPOPULATIONOF31INDIVIDUALSANDITRUNSFOR8GENERATIONSEACHGAHASAPOPULATIONOF5INDIVIDUALSANDITRUNSFOR3GENERATIONS5RESULTSANDCONCLUSIONSTHEMANOEUVREUSEDFORTHEGPOPTIMISATIONINTHEEVALUATIONOFTHECANDIDATESOLUTIONSHASBEENADOUBLESTEPMANOEUVREOF45OFORHEADINGWHILEINCREASINGTHESPEEDFROMRESTTO02M/SANDBACKTORESTTHEBESTRESULTSFOUNDINEACHOPTIMISATIONAREVALIDATEDAFTERTHEOPTIMISATIONTHISVALIDATIONTESTISUSEDTOVERIFYTHATTHERESULTINGTREEISACTUALLYPERFORMINGACONTROLTASK,NOTMERELYGENERATINGASIGNALSHAPEDINTHERIGHTWAYFORTHISMANOEUVREBUTTOTALLYWRONGFORANYOTHERTHEMANOEUVREUSEDINTHEVALIDATIONTESTCONSISTSOFTWOTURNINGCIRCLESLINKEDTOGETHERTHISMANOEUVREHASBEENCHOSENFOLLOWINGTHERECOMMENDATIONSOFTHEMARITIMESAFETYCOMMITTEERESOLUTIONMSC13776FORSHIPPERFORMANCETESTINGTHERESULTINGBESTCONTROLLERSHAVEBEENTESTEDONTHEREALVESSELTHERESULTSOBTAINEDINTHEGPOPTIMISATIONSAREVERYSATISFACTORYTHEMANOEUVRINGPERFORMANCEOFTHECONTROLLERSILLUSTRATEDINTHEFIGURESALSOPROVESTHEIRADEQUACYALTHOUGHTHENUMERICALCOSTVALUESOBTAINEDWITHTHEGPRGOPTIMISATIONAREBETTER,BOTHGPIMPLEMENTATIONSHAVECONVERGEDTOTREESTHATPROVIDEVERYSIMILARCONTROLSTRATEGIESTHEBESTRESULTSOBTAINEDINBOTHSETSOFRUNSAREBASEDONAHYPERBOLICTANGENTFUNCTIONPROVIDINGTHEHEADINGCONTROLANDAPROPORTIONALTERMORAHYPERBOLICFUNCTIONACTINGASAPROPORTIONALTERMPROVIDINGTHEPROPULSIONCONTROLTHETERMINALVALUESCHOSENBYTHESEARCHMETHODASARGUMENTSFORTHEHYPERBOLICFUNCTIONSFORTHESEBESTRESULTSMAKETHISFUNCTIONOPERATEINITSPROPORTIONALRANGEINSTEADOFINTHESWITCHINGAREATHUS,INTHECASEOFTHEPROPULSIONCONTROL,SINCETHESUBSYSTEMISOF1STORDER,THEHYPERBOLICTANGENTPROVIDESANOUTCOMEPROPORTIONALTOTHESURGESPEEDERRORIEAPROPORTIONALTERM中文翻譯應(yīng)用兩種遺傳程序?qū)崿F(xiàn)船舶導(dǎo)航策略的演變1遺傳算法的基本原理遺傳算法是計算數(shù)學(xué)中用于解決最優(yōu)化的搜索算法，是進化算法的一種。進化算法最初是借鑒了進化生物學(xué)中的一些現(xiàn)象而發(fā)展起來的，這些現(xiàn)象包括遺傳、突變、自然選擇以及雜交等。遺傳算法通常實現(xiàn)為一種計算機模擬。對于一個最優(yōu)化問題，一定數(shù)量的候選解（稱為個體）的抽象表示（稱為染色體）的種群向更好的解進化。傳統(tǒng)上，解用二進制表示（即0和1的串），但也可以用其他表示方法。進化從完全隨機個體的種群開始，之后一代一代發(fā)生。在每一代中，整個種群的適應(yīng)度被評價，從當(dāng)前種群中隨機地選擇多個個體（基于它們的適應(yīng)度），通過自然選擇和突變產(chǎn)生新的生命種群，該種群在算法的下一次迭代中成為當(dāng)前種群。在遺傳算法里，優(yōu)化問題的解被稱為個體，它表示為一個參數(shù)列表，叫做染色體或者基因串。染色體一般被表達為簡單的字符串或數(shù)字串，不過也有其他的表示方法適用，這一過程稱為編碼。一開始，算法隨機生成一定數(shù)量的個體，有時候操作者也可以對這個隨機產(chǎn)生過程進行干預(yù)，播下已經(jīng)部分優(yōu)化的種子。在每一代中，每一個個體都被評價，并通過計算適應(yīng)度函數(shù)得到一個適應(yīng)度數(shù)值。種群中的個體被按照適應(yīng)度排序，適應(yīng)度高的在前面。這里的“高”是相對于初始的種群的低適應(yīng)度來說的。下一步是產(chǎn)生下一代個體并組成種群。這個過程是通過選擇和繁殖完成的，其中繁殖包括交配CROSSOVER和突變MUTATION。選擇則是根據(jù)新個體的適應(yīng)度進行的，適應(yīng)度越高，被選擇的機會越高，而適應(yīng)度低的，被選擇的機會就低。初始的數(shù)據(jù)可以通過這樣的選擇過程組成一個相對優(yōu)化的群體。之后，被選擇的個體進入交配過程。一般的遺傳算法都有一個交配概率，范圍一般是061，這個交配概率反映兩個被選中的個體進行交配的概率。例如，交配概率為08，則80的“夫妻”會生育后代。每兩個個體通過交配產(chǎn)生兩個新個體，代替原來的“老”個體，而不交配的個體則保持不變。交配父母的染色體相互交換，從而產(chǎn)生兩個新的染色體，第一個個體前半段是父親的染色體，后半段是母親的，第二個個體則正好相反。不過這里的半段并不是真正的一半，這個位置叫做交配點，也是隨機產(chǎn)生的，可以是染色體的任意位置。再下一步是突變，通過突變產(chǎn)生新的“子”個體。一般遺傳算法都有一個固定的突變常數(shù)，通常是01或者更小，這代表變異發(fā)生的概率。根據(jù)這個概率，新個體的染色體隨機的突變，通常就是改變?nèi)旧w的一個字節(jié)（0變到1，或者1變到0）。經(jīng)過這一系列的過程（選擇、交配和突變），產(chǎn)生的新一代個體不同于初始的一代，并一代一代向增加整體適應(yīng)度的方向發(fā)展，因為最好的個體總是更多的被選擇去產(chǎn)生下一代，而適應(yīng)度低的個體逐漸被淘汰掉。這樣的過程不斷的重復(fù)每個個體被評價，計算出適應(yīng)度，兩個個體交配，然后突變，產(chǎn)生第三代。周而復(fù)始，直到終止條件滿足為止。遺傳程序是約翰KOZA與遺傳算法相關(guān)的一個技術(shù)，在遺傳程序中，并不是參數(shù)優(yōu)化，而是計算機程序優(yōu)化。遺傳程序一般采用樹型結(jié)構(gòu)表示計算機程序用于進化，而不是遺傳算法中的列表或者數(shù)組。一般來說，遺傳程序比遺傳算法慢，但同時也可以解決一些遺傳算法解決不了的問題。交互式遺傳算法是利用人工評價進行操作的遺傳算法，一般用于適應(yīng)度函數(shù)無法得到的情況，例如，對于圖像、音樂、藝術(shù)的設(shè)計和“優(yōu)化”，或者對運動員的訓(xùn)練等。模擬退火是解決全局優(yōu)化問題的另一個可能選擇。它是通過一個解在搜索空間的隨機變動尋找最優(yōu)點的方法如果某一階段的隨機變動增加適應(yīng)度，則總是被接受，而降低適應(yīng)度的隨機變動根據(jù)一定的概率被有選擇的接受。這個概率由當(dāng)時的退火溫度和適應(yīng)度惡化的程度決定，而退火溫度按一定速度降低。從模擬退火算法看，最優(yōu)化問題的解是通過尋找最小能量點找到的，而不是尋找最佳適應(yīng)點找到的。模擬退火也可以用于標(biāo)準(zhǔn)遺傳算法里，只要把突變率隨時間逐漸降低就可以了。遺傳算法擅長解決的問題是全局最優(yōu)化問題，例如，解決時間表安排問題就是它的一個特長，很多安排時間表的軟件都使用遺傳算法，遺傳算法還經(jīng)常被用于解決實際工程問題。跟傳統(tǒng)的爬山算法相比，遺傳算法能夠跳出局部最優(yōu)而找到全局最優(yōu)點。而且遺傳算法允許使用非常復(fù)雜的適應(yīng)度函數(shù)（或者叫做目標(biāo)函數(shù)），并對變量的變化范圍可以加以限制。而如果是傳統(tǒng)的爬山算法，對變量范圍進行限制意味著復(fù)雜的多的解決過程，這方面的介紹可以參看受限優(yōu)化問題和非受限優(yōu)化問題。遺傳算法由密歇根大學(xué)的約翰霍蘭德和他的同事于二十世紀(jì)六十年代在對細胞自動機進行研究時率先提出。在二十世紀(jì)八十年代中期之前，對于遺傳算法的研究還僅僅限于理論方面，直到在伊利諾伊大學(xué)召開了第一屆世界遺傳算法大會。隨著計算機計算能力的發(fā)展和實際應(yīng)用需求的增多，遺傳算法逐漸進入實際應(yīng)用階段。1989年，紐約時報作者約翰馬科夫?qū)懥艘黄恼旅枋龅谝粋€商業(yè)用途的遺傳算法進化者（英文EVOLVER）。之后，越來越多種類的遺傳算法出現(xiàn)并被用于許多領(lǐng)域中，財富雜志500強企業(yè)中大多數(shù)都用它進行時間表安排、數(shù)據(jù)分析、未來趨勢預(yù)測、預(yù)算、以及解決很多其他組合優(yōu)化問題。2論文的主要內(nèi)容本文評估了供應(yīng)船的最優(yōu)化控制結(jié)構(gòu)的基因編程實施方法?；蚓幊逃糜诓渴鸫目刂撇呗?。在水水池實驗室中，優(yōu)化控制器通過計算機模擬和真實操縱性測試被評估。為了處理數(shù)字常數(shù)的世代問題，兩種基因編程算法被實施。第一種方法選擇必要的常數(shù)通過隨機世代創(chuàng)造控制結(jié)構(gòu)。第二種算法包括這些常數(shù)的一種最優(yōu)化基因算法。得到的結(jié)果表明使用基因算法可以優(yōu)化船的推進力和航?？刂破鳌榱吮ＷC水面艦船的安全航行，它們的動力即航海和推進力必須被準(zhǔn)確地控制。這可以通過自動控制系統(tǒng)的設(shè)計和實施來達到。控制技術(shù)的性能不僅取決于控制結(jié)構(gòu)，而且取決于控制器的參數(shù)值。按照常規(guī)，這些參數(shù)由設(shè)計者依靠一種特別方法來手動調(diào)整，取決于設(shè)計者的經(jīng)驗。對這個問題的一種解決方法廣泛應(yīng)用在控制工程領(lǐng)域是使用新發(fā)展的優(yōu)化技術(shù)（例如基因算法來自動調(diào)整參數(shù)。盡管如此，基因算法仍然是參量優(yōu)化器，并且在大多數(shù)情況下不會改變優(yōu)化主題的結(jié)構(gòu)。在控制器優(yōu)化過程中，一種特殊的控制結(jié)構(gòu)方法被提出，同時參數(shù)隨之變化從而達到系統(tǒng)的預(yù)