啤酒发酵 外文翻译原文

Proceedingsofthe2001IEEEInternationalSymposiumonIntelligentControlSeptember5-7,2001MexicoCity,MexicoGeneticAlgorithmsforOptimalControlofBeerFermentationG.E.Carrillo-Ureta",P.D.Roberts*andV.M.Becema**g.camllo@city.ac.uk,p.d.roberts@city.ac.ukandv.m.becerra@reading.ac.uk*CityUniversity,ControlEngineeringResearchCentre,NorthamptonSquare,LondonEClVOHB,U.K.**UniversityofReading,DepartmentofCybernetics,ReadingRG66AY,U.KAbstract-ThispaperusesgeneticalgorithmstooptimisetheobtainedwiththeGeneticAlgorithmoptimisationhasalsomathematicalmodelofabeerfermentationprocessthatoperatesbeenincludedtoachieveimplementableresults.inbatchmode.Theoptimisationisbasedinadjustingthetemperatureprofileofthemixtureduringafixedperiodoftime11.DESCRIPTIONOFTHEPROCESSinordertoreachtherequiredethanollevelsbutconsideringcertainoperationalandqualityrestrictions.Fermentationhascometohavedifferentmeaningstobiochemistsandtoindustrialmicrobiologists.ItsbiochemicalIndexTerms-batchfermentation,beerfermentationmeaningrelatestothegenerationofenergybythecatabolismmodelling,geneticalgorithms,optimalcontroloforganiccompounds,whereasitsmeaninginindustrialmicrobiologytendstobemuchbroader.I.INTRODUCTIONBatchfermentationreferstoapartiallyclosedsysteminThemodellingoffermentationprocessesisabasicpartofwhichmostofthematerialsrequiredareloadedontotheanyresearchinfermentationprocesscontrol.Sinceallthefermentor,decontaminatedbeforetheprocessstartsandthenoptimisationworktobedoneisbasedonthereliabilityoftheremovedattheend.Conditionsarecontinuouslychangingmodelequations,theyareimportantfortherightdesign.withtime,andthefermentorisanunsteady-statesystem,Theseequationsaregenerallynon-linear.althoughinawell-mixedreactor,conditionsaresupposedtoInbatchorfedbatchfermentationprocesses,thereisnobeuniformthroughoutthereactoratanyinstantoftime[5].steadystate.ThecontrolofafermentationprocessisbasedonAlcoholicbreweryfermentationisthemainobjectiveofthisthemeasurementofphysical,chemicalorbiochemicalwork.Brewingandtheproductionoforganicsolventsmaybepropertiesofthefermentationbrothandthemanipulationofdescribedasfermentationinbothsensesofthewordbutthephysicalandchemicalenvironmentalparameters[8],[l11.descriptionofanaerobicprocessasfermentationisobviouslyTheheuristicmethodoftrialanderror,whichisusedtofindusingtheterminthemicrobiologicalcontext.Theanoptimalorpseudo-optimaloperatingregimebymicroorganismsorbiomassconcentrationsarethecentralmanipulatingtheprocesstechnologicalparameters,isoneoffeatureoffermentationaffectingtheratesofgrowth,substratetheoldestoptimisationmethods.GeneticAlgorithmsareconsumptionandproductformation.Growthandproductrandomsearchmethodsbasedonthemechanicsofnaturalformationratesvarywithtimeduetoadependenceontheselection,andnaturalgenetics.InordertouseGeneticpresentstateofthebatch;characterisedbybiomass,substrateAlgorithms,asolutiontotheproblemasagenome(orandproductconcentrations,dissolvedoxygentension,nutrientchromosome)mustberepresented.Thegeneticalgorithmfeedratesandalsoontheconditionoftheculture[7],[lo].thencreatesapopulationofsolutionsandappliesgeneticoperatorssuchasmutationandcrossovertoevolvethe111.MATHEMATICALMODELsolutionsinordertofindthebestone(s).Infermentation,anaccuratemathematicalmodelisAppropriateimplementationofGeneticAlgorithmsincludesindispensableforthecontrol,optimisationandthesimulationthefollowingthreeaspects:definitionoftheobjectiveofaprocess.Modelsusedforon-linecontrolandthoseusedfunction,definitionandimplementationofthegeneticforsimulationwillnotgenerallybethesame(eveniftheyrepresentation,anddefinitionandimplementationofthepertaintothesameprocess)becausetheyareusedforgeneticoperators.Thesimulationoftheselectedmodelhasdifferentpurposes;nomodelcanbesaidtobethebest.ThebeenaccomplishedwiththehelpofSIMULINK(Version2.2)modelisnotexpectedtobeareconstructionoftheprocess,underMATLAB(version5.2)environmentasamodemandratheritisintendedtoserveasasetofoperatorsontheimprovedwayforprocesssimulationandpossiblecontrol.identifiedsetofinputs,producingsimilaroutputasexpectedTheoptimisationoftheprocesshavebeenaccomplishedwithfromtheprocess.theSHEFFIELDMATLABGENETICALGORITHMTheproblemisthatinreallifetheprocessoutputisusuallyTOOLBOXVersion1.2whichisanovelinstrumentforcontaminatedwithnoiseandotherdisturbances,whereasimplementinggeneticalgorithmmethodsasscriptfilesthatideallythemodelshouldfollowthetrueoutputofthecanbechangedaccordingtotheproblemrequirements[3].Aunderlyingrepresentativeprocess,whichisunknown.refiningprocedureforsmoothingthetemperatureprofileEstimationalgorithms,ifproperlychosen,yieldtheparameter0-78034722-7/01/$10.0002001IEEE391valuesafterprocessingofdatacomingfrommeasurementsonthesystem.Forthepurposeofthiswork,akineticmodelhasbeenchosentobepartofthesimulationandoptimisation.Thismodelwasdevelopedandpublishedin[2].ThemodelwasobtainedfrommanyexperimentalstudiesatlaboratoryscalewiththenecessaryequipmentascanbeseeninFigure1.Themodelhasshowngoodresults,anditshouldbenotedthatittakesintoaccountrealisticaspectsoftheprocess,suchasthecharacteristicsofwortandyeast.Fig.1ExperimentalSet-up[2]ds-=-PsXuctivedtThismodeltakesintoaccountthreecomponentsofthebiomass:lag,activeanddeadcellsandconsiderstheactivede-=Pofiuctivecellsastheonlyfermentationagent.Italsoincludessugaranddtethanolconcentrationsandtwoimportantby-productsofthefermentation:ethylacetateanddiacetyl,bothofthemfactorsthatdegradebeerquality.AschemeoftheprocesswithitsmainvariablesispresentedinFigure2.--d(diac)-kdcSXoctive-kdm(diuc)edtMostoftheparametersofthemodelhavebeencalculatedasArrheniusfunctionsoftemperature(assumingtobeaffectedbythetemperatureandrepresentedasexponentialequations).TheconstantvaluesOfkdcandkdmwerecalculatedwiththeexperimentaldataforthediacetylconcentration’sbehaviour:31934.0938313108.3I--130.16--T+273.15Pxo=ekm=eT+273.152658910033.2889.92--33.82--T+273.15T+273.15Pea=ePDO=e11654.641267.24-41.92+-3.27--seffig-Pso=eT+273.15Puo=eT+273.15Fig.2Processschemeforthekineticmodel[2]9501.5434203.9530.72---119.63+-T+273.15ModelequationsandparameterswereadjustedwithPlog=eks=eT+273.15experimentaldatavaluesbynon-linearregression;withthese,kdc=0.000127672kdm=0.00113864goodaccuracyinthepredictionandfitnessofthemodelwasachieved.Themodelisdescribedbythefollowingequations:Table1describesthenomenclatureused:392TABLE1NOMENCLATUREUSEDJ=J,+J,+J,+J,+J,(18)ParameterDescriptionUnitv.USEOFGENETICALGORITHMSFOROPTIMISATION,uoEthanolproductionrateh‘GeneticAlgorithmsaresearchalgorithmsbasedonthemechanicsofnaturalselectionandnaturalgenetics[6].Theyp~Specificyeastsettlingdownrateg/lefficientlyexploithistoricalinformationtospeculateonnew,Ueou,,,Ethylacetatecoefficientrated1hogSpecificrateoflatentformationh-’searchpointswithexpectedimprovedperformance.GAShaveSubstrateconsumptionrateh-’beenusedinavarietyofoptimisationtasks,includingASpecificyeastgrowthrateh-’numericaloptimisationandcombinatorialoptimisationmetEthylacetateconcentrationPPmproblemsascircuitlayoutandjob-shopscheduling[9].dimDiacetylconcentrationPPmConventionalsearchtechniquesareoftenincapableofeEthanolconcentrationg/Ioptimisingnon-linearmulti-modalfunctions.Insuchcases,afFermentationinhibitorfactord1randomsearchmethodmightberequired.GASdonotusek&Diacetylappearanceratemuchknowledgeabouttheproblemtobeoptimisedanddokd,Diacetylreductionratenotdealdirectlywiththeparametersoftheproblem.Theyk,Yeastgrowthinhibitionparameterg/1workwithcodes,whichrepresentparameters.Theparametersk,Sugarinhibitionparameterg/ItobeoptimisedareusuallyrepresentedinastringformsinceSConcentrationofsugarg/lgeneticoperatorsaresuitableforthistypeofrepresentationsiInitialconcentrationofsugarg/ltTimeh(binaryorintegerrepresentationmethod).TTemperature“Cx,,IiveSuspendedactivebiomassg/lVI.IMPLEMENTATIONxboIIn,,,SuspendeddeadbiomassdlASIMULINKmodeloftheIndustryBeerProcesshasbeenxlonSuspendedlatentbiomassg/1created.Themodelincludesthedifferentialequations,parametersandinitialvaluesinanS-Function,togetherwithIv.DESCRIPTIONOFTHEOPTIMISATIONPROBLEMthenecessarytermsoftheobjectivefunction.Figure3showsFermentationcanbeacceleratedwithanincreaseofthismodelundertheMATLABenvironment.temperaturebutsomecontaminationrisksandundesirableby-productsyields(diacetyl,ethylacetate,etc.)couldappear.Theobjectivefunctionwasoriginallydefinedinordertoacceleratetheindustrialfermentationreachingtherequiredethanollevelinlesstime,withoutqualitylossorcontaminationrisks[l].Somemodificationstotheoriginalsub-objectiveparametershavebeenperformedinordertoachieveequivalentweightvaluesinthecostfunction.Thefollowingtermshavebeendefined:Fig.3SIMULNKModeloftheBeerProcess1601Ti+1-T;1J,=-cInordertooptimisetheSIMULINKbeerfermentationi=lAtprocessmodelled,aMATLABscriptcontainingthenecessaryThegoalofeachsub-objectiveintheobjectivefunctionisinstructionsforthegeneticalgorithmtoolboxhasbeenasfollows:J,measuresthefinalethanolproduction,J2limitscreated.Inthisscriptsomeinitialparametersneededinthelevelofethylacetateattheend,J3limitsdiacetylGAStoolboxhavebeendefined,e.g.:numberofindividualsconcentrationattheend,J4temperaturelimitalongthepersubpopulation,maximalnumberofgenerations,boundsonprocess,andJ5penalisesbriskchangesintemperature.decisionvariables,crossoverandmutationrate,etc.SomeThesetermshavebeenjoinedtoobtainanoverallcostspecificroutinescanalsobechosenandorchangedhere,suchfunctionoftheprocess:as:selection,recombinationandmutationfunctionforindividuals[4].393InordertoimplementtheGeneticAlgorithm,theinputItisclearthatwithmoreindividualsfortheoptimisation,temperatureprofilehastobeparameterisedintermsofsmootherresultscanbeobtained;thus,moretimewillbechromosomes.Integernumbersforthetemperaturevalues(inrequiredbythealgorithminordertoobtainanimplementabledegreesCelsius)havebeenusedtorepresentthisprincipletemperatureprofile.Thetemperatureprofileobtainedwiththeinsteadofabinaryprofile.TheinputprofilehasbeenGAToolboxfortheselectedparametersinthispaperisshowndistributedeveryonehouralongtheentireprocess.inFigure5.Consequently,everychromosomeorinputtemperatureprofileconsistsof160digits(decisionvariables)thatcorrespondtoTemperatureProfileonedifferentobjectivefunctionvalueeach.ThisistheJvalue16Xtobemaximisedwiththealgorithminordertomeasurethefitnessofeachchromosome.Initialtestsweredonewithalownumberofindividualsandgenerations,100individualsand100generations,generationgapof50%(howmanynewindividualsarecreatedineverygeneration),crossoverrateof80%(recombination)andmutationratesetat1/160(dependingonthenumberofdecisionvariables).Afterdifferentruns,theparametersandfunctionshavebeensettothefollowingvalues:numberofindividuals=1500;maximumnumberofgenerations=250;generationgap=80%;crossoverrate=80%;mutationrate=1/160;selectionfunction=RouletteWheel;and020406080100120140160recombinationfunction=Doublepointcrossover.HoursThetemperaturerangeforthedecisionvariableshavebeenFig.5Non-SmoothedTemperaturevs.Timesetfrom8to16degreescentigradetakingintoaccountequipmentrequirementsandpreviousprofilesusedinthebeerGiventheseresults,adifferentsmoothingprocessshouldbefermentationindustry(alsorestrictingthealgorithmtoobtainusedtoimproveandmakethemsuitableforimplementation.fasterresults).Withtheseparameters,apromisingThishasbeendonebymeansofaveragecalculationsforeverytemperatureprofilehasbeenachieved.40hoursoftheinitialtemperatureprofileobtained(fourintotal).VII.RESULTSThishasbeendonebymeansofaveragecalculationsforFirsttestshadbeendonewithalownumberofindividualsevery40hoursoftheinitialtemperatureprofileobtained(fourandgenerations;generationgap,mutationandcrossoverratesintotal).Withtheseresults,fournewtemperaturevaluesarehavealsobeenmodified.Differentselectionandobtainedandplacedinthecentrepointofeveryaveragerange.recombinationfunctionsforthegeneticalgorithmhavebeenHorizontallinesfromthepreviouspointandstraightlinestoapplied.Thishasbeendoneinordertoachieveagoodvaluethenextonehavebeenlinkedtoobtainapiece-wiseoftheobjectivefunctionandasmoothtemperatureprofile.linearisation.AnotherMATLABscripthasbeendevelopedtoFigure4showshowtheobjectivevalueismaximisedwithachievethispurpose.Figure6showsthefinaltemperatureeverygeneration.profileobtainedaftersmoothingtheoriginalfromtheoptimisationprocess.TemperatureProfile14.5~9P300s200Ozm1150.I-O50IM,150200250Generation0HoursFig.4ObjectiveValuevs.GenerationNumberFig.6SmoothedTemperature(“C)vs.Time(Hours)394ThisimprovedprofilecannowbeimplementedforByproductsConcentrationindustrialapplication.Thebehaviourofthekineticvariables1.61Diacetyl4ofthemodelcanbeshowninFigures7,8and9;thesewere1.4-obtainedusingthesmoothedtemperatureprofileappliedtothesimulatedmodeloftheprocess.Figure7illustratesthetotal1.2-suspendedbiomassanditscomponents:active,latentand1-dead.Figure8enclosesthesugarandethanolconcentrationEcurves.Figure9includestheby-productsconcentration.20.8-0.6-SuspendedBiomass0.4-IO,I0.2-91I81/Total0M406080100I20140160HoursFig.9ByproductsbehaviourVIII.ANALYSISOFRESULTSSeveralrunshavebeenmadechangingtheinitialparametersrelatedtothegeneticalgorithm.Forinstance:numberofindividuals(startingwith100upto1500),numberofgenerations(from100to350),generationgap(0.5to0.8),etc.Afterreviewingtheseresultsthebesttemperatureprofilewithmaximumobjectivevalue(andwithconvenientoptimisationtime)hasbeenselectedandsmoothedforfeasibleHoursimplementationbymeansofanaveragelinearapproximation.Fig.7SuspendedbiomassbehaviourWiththisnewtemperatureprofilethefinalobjectivefunctionvaluereaches585.5214,whichcomparedwiththeoriginalvalueof541.5504,thatcanbeobtainedusingthetemperatureprofilementionedby[2]forindustryapplication;givesaconsiderableimprovementinthevalueofthecostfunction.Thisvaluehasbeenobtainednotjustbymaximisingthefinalethanolproductionbutalsominimisingtheby-productsconcentrationattheendofthefermentationprocess(DiacetylandEthylAcetate)makingsurethatnounwantedfeaturesaregoingtobepresentinthebeer.0020406080la,1201401601x.CONCLUSIONSANDFURTHERWORKEthamlQncentraticmGeneticAlgorithmshaveprovedtobesuitableintheoptimisationoffermentationprocessesandnopreviousknowledge,suchaninitialtemperatureprofile,hasbeennecessarytoobtainasatisfactoryresult.Previousworkdoneinordertoachieveabetterprofileforimplementationhavebeenacceptableandencouraging[11.TheSIMULINKimplementationdescribedhereappearstobeaflexiblerepresentationofthemodelthatwaseasyto020406080m120140160interfacewiththeGeneticAlgorithmtoolbox.Inaddition,aHllssuperiorcost-valuefunctionhasbeenobtainedbymeansoftheGeneticAlgorithmToolboxfortheoptimisationoftheFig.8Sugarandethanolconcentrationbeerprocess.Alsoasofterprofilebyparameterisingandcalculatingaveragetemperaturesmaderesultssuitableforimplementation.395Furtherworkcouldbedonetoinvestigatethebenefitsofparameterisingthetemperatureprofileusingafewlinearsegments.Inthisway,asmoothprofilecanbeobtaineddirectlyandthereisnoneedforanadditionalsmoothingstep.REFERENCES[I]Andres-Toro,B.;Gir6n-Sierra,J.M.;Lopez-Orozco,J.A.andFemandez-Conde,C.“OptimizationofaBatchFermentationProcessbyGeneticAlgorithms.”IFACADCHEM,InternationalSymposiumonAdvanceControlofChemicalProcesses.pp.183-188.June1997.[2]Andres-Toro,B.;Girh-Sierra,J.M.;Lopez-Orozco,J.A.;Fernandez-Conde,C.;Peinado,J.M.andGarcia-Ochoa,F.“AKineticModelforBeerProductionunderIndustrialOperationalConditions”.MathematicsandComputersinSimulation.ElsevierScienceB.V.Vol.48,pp.65-74.1998.[3]Chipperfield,A,Fleming,P.;Pohleim,H.andFonseca,C.GeneticAlgorithmToolbox.ForUsewithMATLAB.User’sGuide.Versionl.2.Dept.ofAutomaticControlandSystemsEngineering,UniversityofShefield.U.K.1994.[4]Chishimba,L.K.DecompositionandParallelProcessingAspectsofMultiple-ObjectiveGeneticAlgorithmsAppliedtotheOptimisationofIndustrialProcesses.M.Phil.toPhDTransferReportCERC/LKC/I67CityUniversity,London.1998.[SIFerreira,Gisela.Reviewonfed-batchfermentations:mathematicalmodelling,parametersandcontrol.http://www.gl.umbc.edu/-gferrelloutline.htm1,1994.[6]Goldberg,DavidE.GeneticAlgorithmsinSearch,Optimisation,andMachineLearning.Addison-WesleyPublishingCompany,inc.1989.USA.[7]Johnson,A.“TheControlofFed-batchFermentationProcesses-ASurvey”.Autornatica.Vol.23,No.6pp.691-705.1987.GreatBritain.[SILeigh,J.R.OptimalControlinFermentationProcesses.PeterPengrinus,1986.[9]Michalewicz,Zbigniew.GeneticAlgorithms+DataStructures=EvolutionPrograms.Springer-Verlag.1992.USA.[IO]Stanbury,P.F.;Whitaker,A.AndHall,S.J.PrinciplesofFermentationTechnology.Pergamon-ElsevierScience.1995.[IllVolesky,B.andVotruba,J.ModelingandOptimisationofFermentationsProcesses.ElsevierSciencePublishersB.V.Amsterdam,Netherlands.1992.396