燃煤电站锅炉低NOx燃烧优化运行策略的研究

上海交通大学硕士学位论文燃煤电站锅炉低NOx燃烧优化运行策略的研究姓名：魏辉申请学位级别：硕士专业：热能工程指导教师：陆方;罗永浩20080216INOxNOxNOxNOxNOxNOxNOxNOxNOxNOxDCSNOxNOxNOxNOxNOx300MW300MWL-MBPNOxNOxIIBPNOxBPBP“”NOxBPNOxNOxNOxNOxNOxNOx/BPIIIINVESTIGATIONOFOPERATIONSTRATEGYONLOWNOxCOMBUSTIONOPTIMIZATIONOFCOALFIREDUTILITYBOILERABSTRACTLowNOxcombustionoptimizationofboileristoimprovecombustionefficiencyandreduceNOxemissionsthroughrationalcombustioninthefurnace.ButcontrollingmeasurementsofNOxformationisoppositetosteadycombustionandburnoutoffuelinthefurnace.SohowtogiveattentiontoNOxemissionsandboilerefficiencyandmakecostofdrainingcontaminationandcoalconsumptionlowestistheobjectiveofcombustion.Techinically,advancedcontrolmethodsofhigherefficiencyandlowerNOxemissionscanbeconsideredasconsisitingoftwostages.Inthefirststagesomeformofplantmodeling,whichusesbothhistoricalandcurrentplantdata,attemptstocapturetherelationshipbetweentheplantsmanipulatedinputvariablesandtheNOxoutput.InthesecondstagesomeformofconstrainedoptimizationisusedtomanipulatetheinputsofthemodelinordertominimizetheNOxoutput.Thesevaluesarethenpresentedtotheoperator(open-loopmode)andguidingoperationorinsomecasesusedtoautomaticallyadjusttheinputs(closed-loopmode).Inthispaper,studiesoflowNOxcombustionoptimizationarefirstlydiscussedandsummarized.ItmainlycontainsmechanismsofNOxformationandsometypicalcontrolmethods,andsomedominatingoperationfactorsinfluencingNOxformation,andpredictionmethodsofNOxemissions,andsomeoptimizingalgorithms.ThispaperdevelopsexperiencepredictionmodelsofNOxemissionsandboilerefficiencyversussomeinputmanipulatedoperationvariablesbybpneuralnetworkwithanimprovedBPalgorithmsusingL-MlearningIVfunctionbasedbayesregulation.Theresearchobjectivesofmodelingaretwo300MWtangentiallycoal-firedboilers.Andtrainingdataareresultsofcombustionmodificationexperimentsonthetwoboilers.Afterthetwomodelsaresuccessfullydeveloped,testingexperimentsaboutpredictingabilityofmodelareproceeding.TestingresultsshowthatthemodelcouldquicklyandaccuratelypredictNOxemissionsandboilerefficiencyofreal-timeoperatingconditions.BPneuralnetworkhavebeenwidelyappliedtomodelingandcontrollingofnonlinearsystem.AndithasalsobeenappliedtocontrollingNOxemissionsofcoal-firedutilityboilers.Butbeforeasuccessfullytrainedpredictionmodelisgot,itrequiressupplyingabundantandinformation-richhistoricaldataandpremeditatedlyaddssomereal-timeoperationdatatothemodel.Sothatthemodelcouldbeconstantlyupdatedtofullyreflectdynamicoperationconditionsofboiler.ButBPlearningalgorithmisagradientdescentalgorithmwhichgenerallyhasproblemsoftime-consumingandoverfittingduringneuralnetworktraining.Thusmodel'sabilitiesofupdatingandgeneralizationarelimited.Tosolvetheproblem,thepaperdevelopedpredictionmodelsofNOxemissionsandboilerefficiencybyleastsquaresupportvectormachine,whichhasadvantagesofquickercomputationspeedandbettergeneralizationperformance.AndacomparisonaboutpredictingabilitywasmadebetweenleastsquaresupportvectormachinemodelandBPmodel.ResultsshowedthatformermodelsareabletoaccuratelypredictNOxemissionsunderdifferentoperationconditionsandhaveabettergeneralizationperformance.Comparedtoothermodelingmethods,leastsquaresupportvectormachineismoresuitableforon-linework.AftersuccessfullymodelingNOxemissionsandboilerefficiency,thepapercombinesthedevelopedmodelswithgeneticalgorithmsandthenachievesoptimizationsearchingofmanipulatedinputoperationvariablesfewerthanthreedifferentoptimizationobjectives.ThethreeobjectivesareseparatelygettingthelowestNOxemissionswithunconcernforboilerefficiency,gettingthehighestboilerefficiencywithunconcernforNOxVemissionsandgivingattentiontobothNOxemissionsandboilerefficiency.Theoptimaloperationprogrammegotfromoptimizingsystemhasapracticalfeasibility.Thewholeoptimizingprocesscansupplymodelbasisforclosed-looporopen-loopcontroloflowNOxemissionscombustionoptimizationoperation.KEYWORDS:boiler,NOx,bpneuralnetwork,supportvectormachine,geneticalgorithms,combustionoptimization200848?v“v”20084820084811.1.12020073.12030200457.12.2203073100/2006309028.4%53.46%17.871%12.26%5.32%2007273901.68[1]200820108.870%90%67%70%[2]50%2000188020102020246728702020NOx65%NOx35%NOx199119319952652000469200252020108502SO2NOx20105-10NOxSO2[3,4]199637199711300MWNOx650mg/Nm3NOx1000mg/Nm32003228200092004712004710.60.95NOxNOxNOxNOxNOxNOx515CONOxDCSNOx/3/24NOxNOxNOxNOxNOxSNCRNOxSNCRNOxNOxNOxSNCR1.2NOxBPNOxNOx4NOx/1-11-1Figure1-1.DiagramflowmodelingandoptimizingcombustioncharacteristicofpulverizedcoalfiredutilityboilerNOxNOxNOx5BPBP300MWNOx3NOx3BPNOxNOx62.NOx2.1NOx2.1.1NOxNOxNOx200DrakeBlintNOx21286NOxNOxNOxNOxNOxNOxNOx[5]NOxN2ZeldovichZeldovichN22NONON+⇔+(2-1)2NONOO+⇔+(2-2)NOHNOH+⇔+(2-3)2-12OMOOM+⇔++(2-4)MN2NOxN2O22(/)1/2122[][][]kTNOkeNOt-=(2-5)[]=T=Kt=sk1,k2=2-5N2O27NOxNOxa1NOxCO/H2/CO+H2/a1NOxNOxNOxNO6080%2-1HCNNH2HCNHCNHCN?NCO?NHHCNNHiN2NOx2-1[5]Figure2-1.Diagramofbalancingfuel-N[5]2.1.2NOxNOxNOxNOxNOx8NOx12(1)nrhahah=+-(2-6)a?1NOxr?2NOxNOxNOxHCNNHiNOxNOxNOxNOxNOxN2NOxNOx1NOxNOx2NOxNOxNOxNOxNOx93NOxNOxNOxNOxNOxNOxNOxCONOxNOxCONOxNOx22()0.5(21)(1)NOCNCOCOaaa+-→+-+-(2-7)a2-7NOCO--CN-L