Զѧ
ACTA AUTOMATICA SINICA
1999 25 6 Vol.25 No.6 1999



һֻԭģģ
ʶ1)

ժҪһµĻLRģģģͽṹչTSKģ(ETSKģ).LRģͼεĺETSKģ͵ʽ֤ETSKģȨTSKģ͵ĵȼ۹ϵͬʱһֶETSKģ͹Ĳʶ.ʶʵETSKģ͵ıʶЧԤȸ.
ؼʡģʶLMģͣTSKģͣETSKģ.
AN EXTENSION-PRINCIPLE-BASED FUZZY MODEL
AND ITS IDENTIFICATION ALGORITHM
CHEN YixinXIAO Deyun
(Department of Automation,Tsinghua University,Beijing100084)
AbstractAn extended TSK modelETSK model,based on LR type fuzzy numbers and their extended operations,is proposed in this paper.The input-output analytic expression is proved and a parameter identification algorithm is also constructed to identify the areas and gravity centers of the membership functions of LR type fuzzy numbers.A kind of variable weight TSK model which is equivalent to the ETSK model is also deduced.Simulation shows that the ETSK model can give out more accurate long-range predictions than an LM model and a TSK model.
Key wordsFuzzy identification,LM models,TSK models,ETSK models.
1
ģģģģ߼ϵͳģеһӦã1ݣΪһЩӡȷԵϵͳṩһЧĽģ.ģͽṹ˵ģģͿɷֳࣺڹģģͺͲڹģģ.ǰҪģ(Linguistic Models,LM)2ݺTakagi-Sugeno-Kangģ(TSKģ)2ݣ߰ģعģ(Fuzzy Regression Models)3ݺģģ(Fuzzy Neuor-Network Models)4.
LMģһģνʹɵIF-THENһϵͳģĺǰģԲȷԵȽǿTSKģҲֲعģ(Local Regressive Model),ĺһΪĳֺʽȷ.TSKģͽLMģ;ȸ.
ĻLRģ5ݼչһµģģͽṹȾLMģͶԲȶԵ־TSKģ͵ľȣģΪETSKģģ(Extended TSK model),LMģͺTSKģ͵ۺ.
2չTSKģ͡ETSKģ
2.1ETSKģ͵Ļ
ƹµģģ
(1a)
(1b)
ΪչTSKģͣETSKģ.xjR1ΪETSKģ͵YΪETSKģ͵ģAijΪϵģaijΪLRģ(i=1,m;j=1,,n),ֱʾԭģͼӷ.
ʽϿETSKģTSKģ͵ֱ䣬ѹԺеϵչģӶʹÿĺΪģԺ.һʵԵı仯ʹETSKģͿԼLMģͺTSKģ͵ŵ.
ģ(1)ĹģARMAɵõµETSKģͣһ̬ϵͳ.
(2a)
(2b)
ʽy(k-1),,y(k-p),u(k),,u(k-q)R1ΪETSKģ͵YΪģETSKģ͵ģAijBitΪϵģaijbitΪLRģ(i=1,,m;j=1,,p,t=0,,q);ֱΪģͼӷ.
2.2ETSKģ͵Ľ
ʽ(1)ETSKģ֤ͣ¶.
1.һģϵͳFRp+q+1R1ʽ(2).еģaijbit(i=1,m;j=1,,p;t=0,,q)ΪLRģķʽΪ
(3)
ʽy(k)ΪkʱֵģYģh=y(k-1),,y(k-p),u(k),u(k-1),,u(k-q)TΪh=ۣy(k-1),,y(k-p),u(k),u(k-q)TҶ

ʽΪӦģΧͼε(),ΪӦͼεĵĺ().
1.saij=sbit=const0(i=1,,m;j=1,,p;t=0,,q),

(4)
ʱETSKģ͵ȼTSKģͣTSKģETSKģ͵һ.
2.3ȨTSKģ
ʽϿETSKģLRģ㣬չTSKģ.ͨ2ɿչ൱ڶTSKģ͵ĹмȨETSKģ͵ȼһֱȨTSKģͣģ£
(5a)
(5b)
ʽy(k)ΪģAijBitΪϵģi(h)R1ΪiȨֵhy(k-j),u(k-t),gaij,gbitR1,i=1,,m,j=1,,p,t=0,,q.
2.һģϵͳFRp+q+1R1,ʽ(5).

ʽ(2)ETSKģ;ͬʽ
(6)
Цi,h,h,sigiĶ붨1ͬ.
ɶ2ԿETSKģͶTSKģ͵չ൱ڶTSKģ͵ĹмȨ.ȨֵhsiȷֻhĲgi.2ETSKģеģչתΪͨʵ.
2.4ETSKģص
1)1ԿETSKTSKģ͵䣬TSKģ͵ŵ.LRģͬʱȥĲTSKģеϵʱTESKģ;TSKģ.ģԺɻΪģͨʵʱETSKģ;ͳΪLMģͻõĵ㼯ģ.
2)ԹĺʾLRģģԺʹETSKģΪģLMģ͵ص㣬TSKģͶ.
3)ڹLRģ㣬ƺ˼.ץסģľ״ӻʮ.
4)2ETSKģ͵ģչ㣬ָETSKģͶTSKģչıʡLRģ൱ڶTSKģ͹ж̬Ȩ.ȨֵhsiȷhĲgi.
3ETSKģ͵ıʶ
ETSKģ͵ıʶΪṹʶͲʶнṹʶָģ֣ʶǶԹLRģ(ģsiĲgi)бʶ.ṹʶͨģȷʵ.ֻ򻮷ȷǰ£Ĳʶ.
3.1Ĳgiıʶ
ĻѾȷԶhݶ1ЦiĶi.si䣬ʽ(3)д
(7)
.ʽ(7)ΪС˸ʽy(k)HΪɲСʶ㷨ɶԲGбʶ.
3.2siıʶ
h,gi䣬ʽ(3)д
(8)
H1=d1hT,d2hT,,dmhTT,di=ihTg1,H2=ۦ1hT,2hT,,mhTT,S=sT1,sT2,sTmT.ʽ(8)y(k),H1H2ΪɲݶУɶԲSбʶ.
ETSKģ͵Ĳʶ㷨ͼͼ1ʾ.


ͼ1ETSKģͲʶ㷨ͼ
4о
·ϵͳģ
y(k)=0.3y(k-1)u(k-2)-0.7y(k-2)+0.4u(k-1)cos(0.7y(k-1))+
0.3u3(k-1)+0.3y3(k-1),
u(k)Ϊ-11ݾȷֲźţʼֵy(0)=y(-1)=u(0)=u(-1)=0.ǰ300ݽģͱʶú300ݽģͼ.y(k-1),y(k-2),u(k-1)u(k-2)Ӧ򻮷ͼ2ʾ.


ͼ2y(k-1),y(k-2),u(k-1)u(k-2)򻮷
ʶõLMģ36(Ϊ㼯)
IF y(k-1) is  AND y(k-2) is AND u(k-1) isAND u(k-2) isTHEN y(k)=0.5109
IF y(k-1) is  AND y(k-2) is AND u(k-1) isAND u(k-2) isTHEN y(k)=-0.8717
ALSO  ALSO
IF y(k-1) is  AND y(k-2) is AND u(k-1) isAND u(k-2) isTHEN y(k)=0.1102.
õTSKģ3
IF y(k-1) isTHEN y(k)=0.2413y(k-1)-0.6724y(k-2)+0.5107u(k-1)-0.2376u(k-2) ALSO,
IF y(k-1) isTHEN y(k)=-0.1279y(k-1)-0.7159y(k-2)+0.6112u(k-1)-0.0294 u(k-2) ALSO,
IF y(k-1) isTHEN y(k)=0.3263y(k-1)-0.6679y(k-2)+0.4439u(k-1)+0.3145u(k-2).
õ3ڵıʶ300ʶõ3ETSKģͣ



(*,*)Ϊģļ򻯼Ƿǰ߱ʾģ߱ʾģ.ͼ3LMģ͡TSKģͺETSKģ͵ıʶȽ.


ͼ3LMģ͡TSKģͺETSKģ͵ıʶȽ
ͼ3ɼETSKģ͵ıʶãԤľֵEe(k)ͷVare(k)ΪС.滹֤ʵͬԤҲETSKģ.
5
ETSKģģLMģͺTSKģ͵ͷչ.LRģͬʱȥĲΪTSKģеϵ.ʱETSKģģ;TSKģģ.ģԺɻΪģͨʵʱTESKģģLMģͻõĵ㼯ģ.ETSKģ͵ĹLRģģԺһ̶ۺLMģͺTSKģ͵ŵ㣬ģģͶԲȷϵͳ.ETSKģģΪģLMģ͵ص㣬TSKģͶ.ETSKģ͵LRģйأTSKģиǿı.ץסģ뿼ľ״ʹETSKģ͵ļӲ.ETSKģ͵ıʶ㷨Ҳʵ.
1)Ȼѧ(No.6947012)йоԺѧ(No.960643)͸ߵѧУʿѧƵרл(No.97000339)Ŀ.
߼飺1995ҵڱҵѧԶϵ1998廪ѧԶϵ˶ʿѧλʿѧλ.Ҫоģģͽṹģģ͵ıʶͿƷ.
ơ1970ҵ廪ѧΪ廪ѧԶϵڡʿʦ. Ҫо̿ϵͳģʶģ̹ϡҵCIMS.
ߵλ廪ѧԶϵ100084
ο
1Pedrycz W.Fuzzy Modelling:Paradigms and Practice.USA Boston: Kluwer Academic Publishers,1996.320
2Sugeno M,Yasukawa T.A fuzzy-logic-based approach to qualitative modeling.IEEE Transactions on Fuzzy Systems,1993,1(1):731
3Tanaka H,Uejima S,Asai K.Linear regression analysis with fuzzy model.IEEE Transactions on Systems,Man and Cybernetics,1982,12(6):903907
4Pedrycz W.Fuzzy Sets Engineering.Boca Raton:CRC Press,1995,2970
5Dubois D,Prade H.Fuzzy Sets and Systems:Theory and Applications.New York:Academic Press,Inc.,1980,3666
6Mizumoto M.Min-max-gravity method versus product-sum-gravity method for fuzzy controls.In:Proceedings fourth IFSA Congress,Brussels,1991,127130
ոڡ1997-07-14޸ĸڡ1998-09-28
