sensor Based on Neural Networks in Rare-earth Countercurrent Extraction ProcessVol_32.No.4ACTAAUT0MATICASINICAJuly,2006ComponentContentSoft-sensorBasedonNeuralNetworksinRare.earthCountercurrentExtractionProcess1)YANGHuiICHAITianYlOu(SchoolofElectronicsandElectricalEngineering,EastChinaJiaotongUniversity,Nanchang330013)(ResearchCenterofAutomation,NortheasternUniversity,Shenyang110004)(E-mail:yhshuo263.net)AbstractThroughtfusionofthemechanismmodelingandtheneuralnetworksmodeling.acompo-nentcontentsoftsensor.whichiscomposedoftheequilibriumcalculationmodelformulti-componentrareearthextractionandtheerrorcompensationmodeloffuzzysystem,isproposedtosolvetheprob-lemthatthecomponentcontentincountercurrentrare-earthextractionprocessishardlymeasuredon-line.AnindustryexperimentintheextractionyprocessbyHABusingthishybridsoft-sensorprovesitseffectiveness.KeywordsRare-earth,countercurrentextraction,softsensor,equilibriumcalculationmodel,neuralnetworks1IntroductionChinahasthemostabundantrareearthresourceintheworldl.Buttheextractionprocessautomationisstillinthestagethatcomponentcontentismeasuredoff-line,andtheprocessiscontrolledbyexperienceandparametersareregulatedbyhands.Thissituationleadstolowefficientproductionrate,highresourceconsumptionandunstableproductionquality,andlimitstherareearthindustrydevelopmentt1.T0implementautomationintherareearthextractionprocess.theonlinecomponentcontentmeasuringmustbeachievedatfirst.MaincomponentcontentmeasuringmethodsincludetheXrayabsorbspectrometer,Xrayenergyspectrometer,IsotopicX-fluorescenceenergyspectrometer,on-linespectrophotometers.etc,.Theseinstrumentshavenotbeenwidelyusedinindustrybecauseofhighcost,lowreliabilityincontinuousoperation,complexmaintenance,delayinmeasurementandlowaccuracy.Thesoftsensormethod,ontheotherhand,providesanewwaytoon-linemeasurecomponentcontentintherare-earthcountercurrentextractionprocess.Wlewillfurtherourresearchin4byfusionofthemechanismmodelingandtheintelligentmodelingandproposeahybridsoftsensoroftherareearthcomponentcontentwhichcontributestobetterpredictionaccuracyandwiderapplicability.AnindustryexperimentintheextractionYprocessbyHABusingthishybridsoftsensorprovesitseffectiveness.2Descriptionofrare-earthextractionprocessAtwocomponent(AandB)countercurrentextractionprocessisshowninFig.1,whereAistheeasilyextractedcomponentandBisthehardextractedcomponent.Theleftsideistheextractionsectioncomposedof礼stagemixclarifiers.Therightsideisthescrubsectioncomposedofmstagemixclarifiers.InFig.1,u1istheflowofrareearthfeed,u2istheflowofextractionsolvent,u3istheflowofscrubsolvent,u4andu5arethedistributionsofAandBinthefeed,respectively,whereu4+u5=1.pAistheorganicphaseproductpurityofAattheexitandPBistheaqueousphaseproductpurityofBattheexit.PA,kisorganicphasecomponentcontentatthespecifiedsamplingpointinscrubsectionandpB,isaqueousphasecomponentcontentatthespecifiedsamplingpointinextractionsection.Sincethewholeprocessiscomposedofdozensofuptoonehundredstages,theflowregulationofextractionsolvent,scrubsolventandthefeedcannotinfluencetheproductpurityattheexituntilalong-timedelay(oftenhoursevendays).Fortheabovereason,thesamplingpointissetneartheexitandtheexitproductpurity(pA,pB)isguaranteedbymeasuringandcontrolofthecomponentcontents(pA,k,pB,k)atthesamplingpoint.Accordingtocountercurrentextractionprinciple,theparametersp(pA,korPu,k),Ul,u2,u3andu4havethefollowingnonlinearrelationshipP=f(ui,u2,u3,u4,1)Supportedby丽onalNaturalScienceFoundationofP.RChina(50474020,60534010,60504006)ReceivedJanuary24,2005;inrevisedformFebruary16,2006(1)490ACTAAUToMATICASINICAVo1.32where,_isthecomplicatednonlinearfunctionandisthedisturbance.Howtomeasuretheparametersp(pA,korPB,)hasbecomethekeypointofrare-earthprocessautomation.Fig.1Rareearthcountercurrentextractionprocess3MethodofcomponentcontentSOft-sensorintherareearthextractionprocess3.1StrategyofcomponentcontentSOft-sensorinrareearthextractionprocessSincethecountercurrentrareearthextractionprocesshasnonlineardynamicsanduncertainty,itisdifficulttomode1thisprocessbysimplemethodsincludingmechanismmodelingorparametersestimationmethodstTheframeworkofrare-earthextractioncomponentcontentSOft-sensorsystemisdescribedinFig.2.Fig.2FrameworkofrareearthextractioncomponentcontentsoftsensorThewholesystemiscomposedofdatasamplingandpretreatmentsubsystem,extractionequilibriumcalculationmode1anderrorcompensationmode1usingadaptiveneuralfuzzynetworks.ParameteriSthecomponentcontentassayvalue,parameter芦iStheoutputofcountercurrentextractioneqm一1ibriumcalculationmodel,:P一isthemodelingerror,andparameteristheoutputofcomponentcontenterrorcompensationmode1.Apd=一iSusedtotrainthemodelingerrorcompensationmode1.ThentheSOftsensoroutputp=卢+(2)3.2Equilibriumcalculationmodelformulticomponentcountercurrentrare-earthextractionFortherare-earthcountercnrrentextractionprocessshowninFig.1,whenthecomponentAandBrea