V01.14No.4Trans.NonferrousMet.Soc.ChinaAug.2004ArticleID:1003—6326(2004)04—0807—04
Intelligenttemperaturecontrolsystemofquenchfurnace①
HUYan—yu(胡燕瑜)1,GUIWei—hua(桂卫华)1,TANGZhao—hui(JNiNN)1,TANGLing(f蕾玲)2(1.SchoolofInformationScienceandEngineering,CentralsouthUniversity,Changsha410083,China;
2.InstitutionofProduceCommodityQualitySupervisionandTesting,Changsha410087,China)
Abstract:Afuzzy-neuralnetworksintelligenttemperaturecontrolsystemofquenchfurnacewaspresented.Com—binedgeneticalgorithmwithback—propagationalgorithm,theweightvaluesofneuralnetworks,parametersoffuzzymembershipfunctionsandinferencerulescanbeadjustedautomatically,whichrealizestheoptimalcontroloftern—perature.Theresultsshowthatthiscontr01systemcanruneffectivelywithsatisfiedtemperatureprecision:in
tern—peratureuprisingstage,overshot
oftemperatureisunder4℃;instablestage,thescopeoftemperaturechangeiscontrolledwithin士2℃,whichmeetstheneedofcontrolveracityoftemperature.
Keywords:fuzzy-neuralnetworks;MIMOsystem;geneticalgorithm
CLCnumber:TP183Documentcode:A
lINTRoDUCTIoN
Quenchfurnaceisakindofkeyequipment
usedtoheattreatlargecomplexcomponentsofair—
craftsorrockets,withheightof24m,containing
twoheatingroomsandaworkroom,withsteel
boardssetbetweentherooms.Bothheatingrooms
areseparatedintosixsections,eachofwhichcon—
tainsaheatingresistancecontrolledbyapowerad—
juster.Workpiecesareputinworkroom.Acen—
trifugalelectromotorisusedtoblowhotairinto
workroomfromheatingroom.Thedemandofal—
loyintemperature—risingrateandstableveracityat
fixedtemperaturepointishigh,risingtimeneeds
tobeshort,overshoottemperatureshouldkeep
under6℃,atthestablestage,thescopeoftem—
peraturechangeisnotmorethan土3℃.These
twofactorsaffectthestructureandqualityof
workpieces,especiallythelatter.Thestructureof
thequenchfurnaceisshowninFig.1.
Temperaturecontrolsystemofquenchfurnace
isnonlinear,strong—coupling,time—varymg,long—
timedelay,whichmakesconstructionofaccuracy
systemmodelimpossible.Mathematicalmodelwas
discussedinRef.[1],temperaturepredictionwas
givenbytheacquiredmathematicmodelinRef.
[2],becausetheexistenceofundeterminablefac—
tors,theveracityofmodelandpredictionwere
constrained.Multi—sectiontemperaturecontrolwas
discussedinRef.E3],whichwasalsobasedonme—
chanicmathematicalmodel.
Whenthesystemmodeliscomplicatedorhard
toconstruct'engineersexpertknowledgeandmethods,suchasfuzzyoftendescribetheplantby
controlitbyintelligent
orexpert
method.Iffuzzy
1\//////5
2~么厂\/,K
/
/
/6
/
差\
/薹
/
/
——
乡
/8
,9
/
3~
墨
1
厚
/
r]
7
§
4一一——
么至
7董
厂]至
I
V//////A
Fig.1
Structureofquenchfurnace
1一shell:2一Steelboard;3--Thermocouplein
heatingroom;4--Thermocoupleinworkroom;
5--Directionofwind;6一Heatingroom;7~Workroom;
8--Heatingresistance;9--Eleetromotor
controlischosen,theknowledgebase,rulebase,
type
ofmembershipfunctions,defuzzification
methodsandsoon,willallaffectthecontrol
effect.
2FUZZY—NEURoSTRUCTUREoFCoNTRoL
SYSTEM
Manyresearchresultshavebeenacquiredin
generatingfuzzyrulesandadjustingfuzzymember
functionsbyneuralnetworksorgeneticalgo—
rithmsc4—8|。thecombinationofneuralnetworksand
fuzzyinferencehastheadvantagesofself—learning
andinference.Ithasbeenwidelyappliedinsolving
engineeringproblems[9—11|.
Amongtheseresearches,mostofthemare
simpleinputandsimpleoutputsystem(SISO)Is-7],
ormulti—inputandsimpleoutputsystem
①FoundationitemlProject(2002CB312200)supportedbytheNationalBasicResearchProgramofChina
Receiveddate:2003一10—30lAccepteddate:2004—02—10
CorrespondencetHUYan-yu,PhDcandidate,Tel:-I-86—731—8876262;E-mail:resun_O@163.com
.808.Trans.NonferrousMet.Soc.ChinaAug.2004_[4'8t9’12’131.thenregardingthemulti—inputand
multi—outputsystem(MIMo)asthecombinationof
MISOsystem。inthemeantime,numberofthe
membershipfunctionsofinputandoutputisregar—
dedtobesimilar.Researchonabstractingrulesby
GAcanbefoundinRefs.E6,141,butonlyMISO
systemwasdiscussed.Inthispaper,MIMOfuzzy-
neurosystemwithdifferentsemantictermisdis—
cussed,GAoptimalmethodisalsoemployedtoat—
taingeneralMIM0method,eventually,thismeth-
odisusedintemperaturecontrolofquenchfur-
naCe.
2.1
MIMO
fuzzy-neurosystem
Supposecontrolsystemhasminputsand咒
outputs,eachinputhasSf(i一1,2,…,仇)member—shipfunctions,eachoutputhasTjmembershipfunction(J=1,2,…,咒),thenthetotalruleisq一ⅡSi(1)i=1
Eqn.(1)showsthattherulenumberincreasesinindexlawwiththevariablesandthemembershipfunctionsofvariables.Structureoffuzzy—neurosystemisshowninFig.2.
Yn
x1bLayer5
H“5)
Layer4
H一4)
matrixisrepresentedby∥孙,numberofrowand
lineisidentical,soitisadiagonalmatrix.
W(2)。
'.,(3)一
1
all
0
O
0
O1
1O
O1
…
O
1
0
0
O
O
1
口棚m
0…01
Thethirdlayerdescribestheantecedentpartofrules.Atthebeginning,neuronnodenumberis
themaximumrules,theweightsmatrixofneural
networksw(3’isusedtodescribetheantecedent
part,eachrowofthematrixrepresentstheante—
cedentofarule.theelementsofmatrixconsistof1
and0,where‘1’denotesthatthecorresDondent
antecedentexists,‘0’isontheversus。thetotal
numberof1islessthan7n.Thenodefunctioncal—
culatesthefiringstrength,multipliermethodisa—
dopted,sothejthoutputnodehasthefollowing
form:
Therowandlineofweightmatrixw‘3’IIS;
i=1
Layer3”
讯3)and∑Si,respectively,theelements1inthefirst
Layer2
H“2)
Layerl
wO)
Fig.2Structureoffuzzy-neurosystem
Thefirstlayerisinputlayerwhichiscrisp,thenodefunctionsperformthetransferofinputtothenext1ayer,equationsbetweentheinputsandtheoutputscanbewrittenas
qfl)一∞譬’z:D—wn’z;”
Thesecondlayerperformsfuzzificationofthe
inputvariables,theneuronnodefunctionstransferthecrispinputintosemanticterm。iftheGaussac—
tivationfunctionisemployed,wehavethefollow—ingequation:
q2,一Hi—exp[一(挚)z]孝)-掣,
“ii
(2)wheretheweightsandbiasesofneuralnetworksrepresent
thewidths(n』f)andthemeans(Cji)ofinputmembershipfunctionsrespectively.Numberofneuronsatthislayerisn2=∑Sf.Weights
i雀1rowandthesecondcolumn,thefirstrowandthelastcolumnrepresentthefollowingantecedent:ru[e2:IFx1ISP12AND…ANDx。IS弘。MTHEN…
Theothersrepresenttheantecedentofrulesinthesimilarmanner.
Thefourthlayerconsistsofoutputs.Neuron
outputinthislayeristhesumoffiringstrengthofruleswithsimilarconsequence,thenumberofneu一
"
ronisn4一yTj,and丁iisthenumberofmem一篇。
bershipfunctionofjthoutput.Weightmatrixcanbeadjustedinthetrainingprocess,everyelement
inwhichconsistsof0or1,eachrow
representstheoutputmembershipfunctiontermintheconse—quencepartofarulewhichisdefinedbycolumn
number.Sincethere
are72outputs,nseparateblockmatrixexistsrepresentingeveryoutput.
m
ROWofeachblockmatrixisL,columnis口s{,
i=1
theformofwhichissimilartothe
W(2),onlythenumberofrowandcolumnandthenumberof1inthemisdifferent.Forexample,positionof1inthefirstrowandthesecondcolumnofithblockmatrixrowshowthatthesecondrulehastheithmember—shipfunctiontermintheconsequencepart,asin
0
.
O
一2
o
一吼o
S”
。Ⅱ㈦o
2
l
一
”
“q
一
\,
纷
,L
m
m
一
谚
V01.14№.4Intelligenttemperaturecontrolsystemofquenchfurnace?809?
thefollowing:
R2:IF…THEN…YiIST_
Several‘1’mayexistineachrowofw‘”be—
causeconsequenceofseveralrulesmayhavesimilar
outputsemanticterm,however,arulecanhave
onlyoneoutputterm,numberof1inacolumnis
notmorethan
w“’一[硼1
r0
one,then
w2|¨?1w。
1
…1
w‘4’isas
|¨?l叫。]T
驴雌0?.0000。卜2,…川
毗一||o‘.1严l'Z,…川
…
Thefifthlayerisoutputlayer,defuzzificationisperformedinthislayer,ifcenterofgravitymethodisemployed,outputcanbewrittenas
3,l
2∑m刍站口,/∑懿a,(4)』=1J=1
wheremSand毽aremeanandwidthofthejthout—putmembershipfunctionfortheithsystemoutput.Weightmatrixissimilartothesecondlayer,rowandcolumnareequal,itconsistsofadiagonalmatrix.2.2Fuzzy-neurostructureofquenchfurnaceBecausethetwoheatingroomshavethesamestructure,hereonlyoneisdiscussed,thequench
furnacecanbetreated
asa6inputs
and3outputscontrolsystem.Theinputscontroltheonandofftimeoftheresistance,outputisthetemperatureofworkroom.Intheheatingroom,hotairrises,sothebottomresistanceneedstoproducemoreheattomaintaintheuniformityofthetemperatureintheworkingroom.Intheworkingroom,thehotairi8fromtoptobottom,10adingworkpiecesisthroughbottom,whichcausesthelowerpartwith10wtemperature.Inordertosimplifytheproblem,temperaturevaluemeasuredinupandmiddlepartisaveraged,treatedasoneoutput.
Thedifferenceoftemperature(e)andthechangeofdifference(Ae)aretreatedasfuzzy-neuroinputs,thedifferenceisdescribedby5se—manticterms,thechangeofdifferenceisdescribedby3semanticterms.Sothefuzzy—neuronetworkshas4nodesinthefirstlayer.16and225nodesinthesecondandthirdlayer,respectively.Liketheoutputoftemperature,thebottomheatingroomisdescribedby5,theothersused3topresent,sothefourthlayerandthefifthlayerhas20and6nodes,respectively,andthusthefuzzy-neuronet—worksisdetermined,theweightmatrixisadjustedbyGAandBP,opticalrealizationofthesystemareacquiredbytheopticalparameters.
3GAANDBPTRAININGALGoRITHMSInodertoacquireanoptimalfuzzy-neurostructureofthesystem,trainingisperformedbyexperimentdata,byadjustingtheparametersandrules.Beforetrainingprocess.theinitialparame—tersandrulesarechosenatrandomorbyexpertexperience.ThestructureframeisshowninFig.3.
Fig.3Structureofcontrolsystem
Fromtheaboveanalysis,weknowthatthefuzzyrulesaredefinedinlayers3and4.TheweightmatrixiscodedasthechromosomesofGA。whichhasthestructureasfollows:eachchromo—somehasnparts,eachpartiscalledasub—chromo—some,thenumberofgenesineachsub-chromo—someisthesumofrulesQ,theantecedentandconclusionmatrixisencodedbyaninteger,whichbelongtoasetG∈[o,1,…,t](‘『一1,2,…,Q).Intheinitialphase,thelengthofeach
partisequaltothenumberofmembershipfunction,thevalueofCidenotestherowpositionof1ineachcolumnoftheithconclusionblockmatrix:ifthereisnomembershipfunctionofcertainoutput,Cjiszero,thestructureofchromosomecanbedepictedinFig.4.
Fig.4
Formofchromosome
GAactsoneachsub-chromosometofindtheopti—
realfuzzyrules.Calculatingthesumofsquare
errorofdifferenceforeachindividualinthepopulation.sub—
tractingit
fromthelargest
errorvalueofthatgenera—tionmakesthefitnessfunctionofeachindividual。ifthereareenoughinput/outputexperimentdata,GAcanbeappliedtofindtheoptimalsetoffuzzyrules.Initializationcontentscontainthefollowing:numberofpopulation,timesofiteration,permittederrors,selec—tionprobability,crossprobability,mutationprobabili—tyandSOon,thechromosomesarechosenrandomlyatthebeginning,denotingtheweightmatrixoflayers3and4.
BecausetheinitialparametersofGaussfunc—
tionischosenatrandomorbyexpert
experience,BPalgorithmisemployedtooptimizetheweightmatrixparametersoflayers2and5.Theobject
functionistheminimumofsquare
errorsum.
1』L
E一专≥:(y^一y埘)2(5)
●E=1
Theoptimalweightsandthresholds
areattainedbyBPalgorithm,whicharetheoptimaImembershipfunctionparametersofinputsandoutputs.TwoinputsM噼havethesimilarparameters.andtheupper5out—
?810?
Trans.NonferrousMet.Soc.ChinaAug.2004
puts
havethesimilarMFparameters.
4RESULTS
Thetemperatureismeasuredevery400milli—second,averagevalueofevery5measurevaluesasonemeasurevalue,averagevalueofevery4meas—urevalueasonesamplevalue.Thesettemperaturevalueofupandmiddlepartinworkroomis470,468℃atbottompart,respectively.Sampleinter—nalis8s,thesampledataisalsothefeedbackvalueofcontrolsystem.InGA,initialpopulationis50,iterativetimesis200,permittederroris0.3℃,selectionprobabilityis0.5,crossprobabilityis0.3,mutationprobabilityis0.05,aftertheGAandBP,thedatainTables1and2areacquired.ExperimentshowsthatbettercontroIeffectisat—tainedbytheseparameters,thebiggestrelativeer—roris0.89%.
Table1MFParameterof3inputs
8AeClassClass
口fnCNB90182N3438
NS85262Z3868
Z71338P4298
PS58396
PB52445
Table2MFParametersof6outputsClass占mClass占mS0.240.32VS0.200.25
M0.160.68SO.13O.43
B0.220.87M0.140.56
B0.230.71
VBO.30O.86
5CoNCLUSIoNS
1)TheMIMOsystemmodelofquenchfur—naceisconstructedaccordingtothedemandoftemperature.Combininggeneticalgorithmwithback—propagationalgorithm,theweightvaluesofneuralnetworks,parametersoffuzzymembershipfunctionsandinferencerulesareadjustedautomat—icallytorealizetheoptimalcontroloftemperature.2)Thiscontr01systemcanruneffectivelvwithsatisfactorytemperatureprecision:inuprisingstage,overshotoftemperatureis3℃,instablestage,thescopeoftemperaturechangeiscon—trolledwithin±2℃.Ithasbeenusedinafactorysuccessfully.
[11
[23
[3]
[41
[51
[6]
[73
[8]
[91
[10]
[11]
[121
[13]
[14]
REFERENCES
LIUGui—qing,HakamadaH.Modelanalysisforcorn—
bustioncharacteristicsofRDFpellet[J_:].TransNon—ferrousMetSocChina,2002,12(3):534—539.
ZHENQiang,BA0Hong,WANGFu—ming,eta1.
Predictionofthermodynamicpropertiesformulticom—
ponentsystemwithChoumodel[J1.TransNonferrous
MetSocChina,2000,10(5):642—644.
CHENGXiao—ruing.Researchonsubsectiontempera—turecontrolsystem[J].TheChineseJournalofNon—ferrousMetals,1998,8(S3):103—105.(inChinese)
WUShi—qian,JooErM,YANGGao.Afastapproach
forautomaticgenerationoffuzzyrulesbygeneralizeddynamicfuzzyneuralnetworks[J1.IEEETransactiononFuzzySystems,2001(4):578—593.
CheongF,LaiR.Constrainingtheoptimizationofafuzzylogiccontrollerusinganenhancedgeneticalgo—
rithm[J].IEEETransactiononSystems,Man,and
Cybernetics--PartB,2000(1):31—45.
WANGJing,LIYu-lan,CAIZi—xin,eta1.Optimaldesignmethodoffuzzysystembasedongas[J].Con—
trolTheoryandApplication,1999(5):728—734.(inChinese)
ZHANGYun,MAOZhong—yuan.Ageneralizedfuzz—
y-ANNnetworksanditslearningalgorithm[J].Con—
trolTheoryandApplication,1998(1):148—151.(inChinese)
WANGyin,RoNGGang.Aself-organizingneural-network—basedfuzzysystem[J1.ControlTheoryandApplication,1999(6):455—457.(inChinese)
WANGYao-nan,ZHANGChang—fan.Genetic.based
neurofuzzynetworkcontr01forcomplicatedindustrial
process[J].ControlTheoryandApplication,1999(6):886—891.(inChinese)
ZHANGYi,LIRen-hou.Designofmulti—variance
fuzzycontrollerbasedongas[J].ControlTheoryand
Application,1996(5):409—416.(inChinese)
YA0Yi—wu,XIoNGJing-tao,MA0Zhong-yuan.
Self-organizedfuzzycontroller[J].ControlTheory
andApplication,1996(6):738—743.(inChinese)
CHENJian-qin,CHENLai-jiu.Learningoffuzzy
rulesbasedonself-adaptiveneurons[J1.Controland
Decision,1994(5):350—355.(inChinese)
SantosM,DexterAL.Temperaturecontr01in1iauid
helium
cryostatusingself-learningneurofuzzycon—troller[J].IEEEProc—ControlTheoryAppl,2001
(3):233—238.
SetteS,BoullartL.Animplementationofgenetical—
gorithmsforrulebasedmachinelearning[J].Engi—
neeringApplicationofArtificialIntelligence,2000,
13:381—390.
(EditedbyLONGHuai—zhong)
Intelligent temperature control system of quench
furnace
作者:胡燕瑜, 桂卫华, 唐朝晖, 唐玲
作者单位:胡燕瑜,桂卫华,唐朝晖(School of Information Science and Engineering,Central south University,Changsha 410083,China), 唐玲(Institution of Produce Commodity
Quality Supervision and Testing,Changsha 410087,China)
刊名:
中国有色金属学会会刊(英文版)
英文刊名:TRANSACTIONS OF NONFERROUS METALS SOCIETY OF CHINA
年,卷(期):2004,14(4)
被引用次数:1次
参考文献(14条)
1.LIU Gui-qing.Hakamada H Model analysis for combustion characteristics of RDF pellet 2002(03)
2.ZHEN Qiang.BAO Hong.WANG Fu-ming Prediction of thermodynamic properties for multicomponent system with Chou model[期刊论文]-Transactions of Nonferrous Metals Society of China 2000(05)
3.CHENG Xiao ming Research on subsection temperature control system 1998(zk)
4.WU Shi-qian.Joo Er M.YANG Gao A fast approach for automatic generation of fuzzy rules by generalized dynamic fuzzy neural networks 2001(04)
5.Cheong https://www.wendangku.net/doc/178143874.html,i R Constraining the optimization of a fuzzy logic controller using an enhanced genetic algorithm 2000(01)
6.Wang Jing.LI Yu-lan.CAI Zi-xin Optimal design method of fuzzy system based on gas 1999(05)
7.Zhang Yun.MAO Zhong-yuan A generalized fuzzy-ANN networks and its learning algorithm 1998(01)
8.Wang Yin.RONG Gang A self-organizing neuralnetwork-based fuzzy system 1999(06)
9.Wang Yao-nan.ZHANG Chang-fan Genetic-based neurofuzzy network control for complicated industrial process 1999(06)
10.Zhang Yi.LI Ren-hou Design of multi-variance fuzzy controller based on gas 1996(05)
11.YAO Yi-wu.XIONG Jing-tao.MAO Zhong-yuan Self-organized fuzzy controller 1996(06)
12.CHEN Jian-qin.CHEN Lai-jiu Learning of fuzzy rules based on self-adaptive neurons 1994(05)
13.Santos M.Dexter A L Temperature control in liquid helium cryostat using self-learning neurofuzzy controller 2001(03)
14.Sette S.Boullart L An implementation of genetic algorithms for rule based machine learning 2000
相似文献(2条)
1.外文期刊HU Yan-yu.GUI Wei-hua.TANG Zhao-hui Intelligent temperature control system of quench
furnace
A fuzzy-neural networks intelligent temperature control system of quench furnace was presented. Combined genetic algorithm with back-propagation algorithm, the weight values of neural networks, parameters of fuzzy membership functions and inference rules can be adjusted automatically, which realizes the optimal control of temperature. The results show that this control system can run effectively with satisfied temperature precision; in temperature uprising stage, overshot of temperature is under 4 deg C ; in stable stage, the scope of temperature change is controlled within + - 2 deg C , which meets the need of control veracity of temperature.
2.外文会议Shuqing Wang.Hui Liu.Zipeng Zhang.Suyi Liu Research on the Intelligent Control Strategy
Based on FNNC and GAs for Hydraulic Turbine Generating Units
It is difficult to gain better control performance using general control strategy to control Hydraulic turbine generating units system because it is a complicated non-linear MIMO system. In this study, a new control technique, which efficiently get optimal control parameters for fuzzy neural network controller through the training of neural network and Genetic algorithms, was proposed and then applied to control turbine generating unit system. In the designed control system, fuzzy reasoning rules, member function
error is less. The improved genetic algorithms, which overcomes general genetic algorithms' disadvantage, has quick training speed and gives whole optimized parameters for fuzzy neural network controller. RBF neural network is employed to identify and predict the relation between input and output of hydroelectric generating units system. Simulation experiment results show that the designed controller can control hydraulic turbine generating units efficaciously and has quick controlling speed and less controlling max-error. So it provides a good control strategy for hydraulic turbine generating units system.
引证文献(1条)
1.姚玲英多参数解析描述的模糊控制规则及其优化方法[期刊论文]-广东教育学院学报 2005(3)
本文链接:https://www.wendangku.net/doc/178143874.html,/Periodical_zgysjsxb-e200404034.aspx
授权使用:河南工程学院(hngcxy),授权号:89a855b5-fd10-4ab3-bc46-9e9900a22268
下载时间:2011年3月1日