Զѧ
ACTA AUTOMATICA SINICA
1998 24 4 Vol.24 No.4 1998




ϵͳϵͳƷ1)
С봿
ժҪϵͳϵͳƷԷԺıƽϵͳ̬Եɻ칤Ͽʵֵԭϵͳϵͳ(ϵͳֱԭϵͳӣɦ׻αԸϵͳ)Ԧ׻αԸϵͳԿʵֶԭϵͳЧ.÷ڽһԡϵͳҽṹ򵥣ڹʵ.
ؼʡϵͳϵͳ磬ʶ.
NEURAL NETWORK THORDER INVERSESYSTEM CONTROL
METHOD FOR NONLINEAR CONTINUOUS SYSTEMS
DAI XIANZHONG LIU JUN FENG CH UNBO
(Automatic Control Eng.Dept.of Southeast Univ.,Nanjing 210096)
Abstract Based on the design process of continuous systems,t his paper focuses on the NN thorder inversesystem of nonlinear continu ous system.The NN thorder inversesystem is constructed by combining th e NN approximating nonlinear function of the system with integrators denoting dy namic characteristics.Furthermore,a linear closedloop controller for the compo site system is designed to complete effective control of the original system.Sim u lation results show that this method can apply to general linear or nonlinear co ntinuous systems,and that the controller can be implemented easily.
Key words Nonlinear system,inversesystem,neural networks(NN),identification ,control.
1
˹ԡϵͳıʶȡ˽ϴĽչ1-6.Էϵͳ(NN)ƵоĿǰɢʱϵͳ(ʹе״ϵͳҲֻǽϵͳɢȻɢϵͳƷɢĽģͽбʶƣʵǽɢϵͳķӵϵͳ).ҪNNɢϵͳĿȴϵͳڷۣڹʵ֣ڷ.ʵʵıϵͳϵͳоϵͳҪ.
һϵͳȫϵͳƷоϵͳ (ϵͳֱԭϵͳӣɦ׻αԸϵͳ)ĽṹʶӦڷϵͳĿƣγֱϵͳƷ.
2ϵͳϵͳ
ϵͳϵͳĲף7е¶壺
һϵͳ(Ի)Ϊu(t),Ϊy(t),ʼ״̬ x(t0)=x0.ӳϵΪ:uy,
y(.)=ȣx0,u(.)ݻдΪy=u(1)
覰ΪһϵͳʾӳϵΪydud,ʽ
yd=ud=yd,(2)
ϵͳΪϵͳĵλϵͳ.ӦأϵͳΪԭϵͳ.
覰Ϊһϵͳʾ䴫ݹϵΪud, ڦ=y()d(t),ydĦ׵Ӧ
Ȧ=Ȧ(y()d)=ud=yd,(3)
ϵͳΪϵͳĦ׻ϵͳ.λϵͳǦ=0ʱϵͳ.
ڸϵͳϵͳ򦰦( ϵͳ໥תһϵͳڵλϵͳʹڦϵͳǵȼ۵ģ7)ϵͳΪ ϵͳ .ϵͳѧģ(ѧʽ)֪ʱݸѧģͣжϵͳǷ7 .ϵͳѧģδ֪δȫ֪ʱж.Ϊϵķ㣬ڱ۵ķ ϵͳٶµϵͳ.
3ϵͳĽṹʶ
ϵͳĹ˼ǣδ֪ϵͳþ̬ɻ(΢)ʵ֦ϵͳĹܣо̬ƽ̬Ժ(΢)ӳϵͳĶ̬.־̬붯̬໥ֿ˾ְĽṹڼڲṹٶ.
ԭϵͳ(SISOҲԺʱt)ѧģ÷΢ַ
fy(n),y(n-1),y(1),y,u(m),u(m-1),u (1),u=0(4)
ʾ.ʽu(t)Ϊ룬y(t)Ϊm,n(mn)ֱΪ롢Ľ״Σy(n)Ϊy n׵u(m)Ϊum׵.
ԭϵͳ뼰׵֪.ٶδ֪gʹ
u=gy(n),y(n-1),y(1),y,u(m),u(m-1),u (1)ݡ(5)
.ʱʵֽṹܼ򵥵ϵͳͼ1(a)ʾYʾy,y(-1),y(+1),,y(n),U ʾu(1),u(m)ݣƽʽ(5)еδ֪Ժ g.ԭϵͳ뼰׵֪ۦ0n֮ȡֵYU Ϊ֪Կʵ׵ϵͳ.ϵͳ߱ ʶṹҲܼ򵥣ͼ1(b)ʾY1ʾy(1), , y(n),Uͬͼ1(a)NNźȫԭϵͳṩҪκλ ֻ΢.


ͼ1ԭϵͳĸ׵֪ʱNNϵͳṹ(a)ʶṹ(b)
ʵӦУҪԭϵͳĸ׵ȫ֪(ɲ).£ϵͳyuܼϵͳĽnm֪⣬ϵͳ һ δ֪.ʱԭϵͳĸ׵ֻ֪ͨ״̬лֻ΢.ʵʼϵͳУֻ΢ɼֵֵֻͨ΢ʵֵ֣ʵ־ȸߣֵ΢־Ȳһֵ.
ԭϵͳĸ׵δ֪ʱͬٶδ֪gn
u(m)=gny(n),y(n-1),y(1),y,u(m-1),u(1),u =0(6)
.ʱûNNɹnϵͳͼ2(a)ʾƽʽ(6)еδ֪Ժgn.Ӧϵͳʶṹͼ2(b)ʾҪmn΢߱ʶ(עڷʵʱ߱ʶУͨy,uĸٲþܶݺ͸߾ȵֵ΢߱ʶ).


ͼ2ԭϵͳ״̬ȫδ֪ʱnϵͳṹ(a)ʶṹ(b)
(Dʾ΢֣D-1ʾ)
nϵͳԭϵͳɵαԸϵͳn׻ֵʣnϸʱĸϵͳȶԲ.˴ʵӦýǶȣϣϵͳĽԽԽ.絥λϵͳģԭϵͳĸ׵δ֪ʱ絥λϵͳṹҪõ=0ʱͼ1(a)еYUҪ϶΢ʵ־Ⱥܲ.ԭϵͳĸ׵δ֪ʱʹ΢ֶʹ ûʵֵϵͳͽΪ=n-mԭϵͳԽΪʹ΢ʱҪ(ʵ)ͱϵͳ״β״Σϵͳʽ(6)Уy΢ֽ״αu΢ֽ״θߦֻѡyĦ׵߽׵Ϊϵͳ룬ʹϵͳ״βuĽ״ΣӶ(ʵ)ʱɵϵͳΪ׻߽׵ϵͳ.(עϵͳԽΪ0)
ϵͳԭϵͳ΢ַ̱ʾʽ(4)ֿɱʾΪ
y(n-m)y,y(-1),y(-m),u,u(-1),u(-m)=0.(7)
ʽy(-m)Ϊym׻֣u(-m)Ϊum׻֣ʽ(4)һδ֪Ժʹ
u=y(n-m)y,y(-1),y(-m),u(-1),u(-m)ݡ(8)
.ʱʵͼ3ʾϵͳ(ṹμͼ2(a)ͼЦΪԭϵͳԽyʾy(-1) ,y,y(-1),y(-m),uʾu(-1), u(-m),ƽʽ(8)еδ֪Ժ.


ͼ3ϵͳṹ
ڷϵͳԭϵͳ΢ַʽ(4)ʽ(7)ʾͬʵ ϵͳΪԭϵͳԽ.
ϵͳʹõѡһ־̬磬ֻҪܱƽԺ .ʶʹõ㷨õƥһ㷨.ڷоУѡýһĶ֪ϵͳĽnԽȷһΪ2n+1-ڻԴΪ1.ʶ㷨óõĸӳĸĽBP㷨.򵥵ṹѧϰ㷨˵ϵͳṹ򵥡ʵ.
4NNϵͳڿеֱӦ--ϵͳƷ
Ѿʶõϵͳֱԭϵͳ͹һαϵ ͳЦ׻.һӦϵͳеĸֿƷPIDơУ(ͼ4ʾ)ȵȣƳϵͳҪĿͬϵͳһ𹹳ɸϿʵֶԭϵͳЧ.


ͼ42αԸϵͳüʾͼ
5NNϵͳƷʵ
׷ϵͳΪ7
+3y2+y3=u.(9)
źud(t)=sin(2t/2.5)ѧϰud,yֵͨ (ʵʵϵͳ)ȡ(Ϊ0.001s2500ݶ)ֵ΢ֵãٴеȼȡ25ݶԹѵud,y, .
ʶʱԭϵͳδ֪.ʶṹͼ2ʾõNN2ϵͳ(341)ԭϵͳ2׻αԸϵͳ󣬲üùͼ4ʾĿϵͳ(K=4,f1=1,f2=0.8).ϵͳڷ(ֱֵΪ0.6͡0.4)Ӧͼ5ʾʵΪϿϵͳӦΪźţ㻮ΪԭϵͳԸźŵӦ.


ͼ5NN2ϵͳϿϵͳ(ͼ4)ԷӦ
6ϵͳƷƷıȽ
(1)ͬ˼ƷֱȣĽܵķ3 ص㣺
1)NNƽԺ֡΢ֱ̬ԣֹȷNNṹ򵥣
2)ɵϵͳΪNNϵͳNNλϵͳʵ֣
3)NNϵͳΪһ֣ϵͳƵԿһ֣ԿƼ򵥣ϵͳ.
(2)ͨµϵͳ7ȣĽܵķиӦ÷ΧͨµϵͳҪ֪ԭϵͳ׼ȷģͺͱܽʽ(4)ϵͳӦʽʾͨµϵͳʵӦеƿ⣬ʵӦ.ĽܵķҪ.
7
(1)ĽܵӻɵNNϵͳٸԷƵȳƷȫϵͳƷΪϵͳNNṩһֽһĿɹʵֵķصǣ
1)ʽ(4)ʾһϵͳ(ٶϵͳǿ).
2)֪ԭϵͳѧģ(֪ϵͳĽԽ).
3)ṹ.
(2)ԭϵͳԽΪ롢ĸ׵δ֪ʱʵʹ̿ʵֵһ㲻ǵλϵͳǦϵͳ.
1)ȻѧĿ.
߼飺СУ19545.1986ҵ廪ѧʿѧλΪϴѧԶϵΣڣʿʦ.ǰоΪ硢ԿơϵͳơƵ.
У19729.199119941997ڶϴѧֱѧʿ˶ʿʿѧλ.ǰоΪ硢硢ϵͳơƵ.
봿鼰Ƭ221ڣ199511µѡΪйѧԺԺʿ.
ߵλϴѧԶϵϾ210096
ο
1 Hunt K J,Sbarbaro D et al.Neural networks for control systems-- a survey.Automatica,1992,28:1083-1112
2Nguyen D H,Widrow B.Neural networks for selflearning control systems.IEEE Contr.Syst.Mag.,1990,10(2):18-23
3Chen Fuchuang,Liu Chenchung.Adaptively controlling nonlinear continuoustime systems using multilayer neural networks.IEEE Trans.AutoContr., 1994,AC39:1306-1310
4Ahmed M S.Block partial derivative and its application to neuralnetbase d directmodelreference adaptive control.IEE Proc.D,1994, 141(5):305-314
5Hunt K J,Sbarbaro D.Neural networks for nonlinear internal model control.IEE Proc.D,1991,138(5):431-438
6Nielsen R H.Theory of the backpropagation neural network,I JCNN 89,1989,I59 3-I605
7ģԪ.ԿƵϵͳ.廪ѧ磬1991
ոڡ1996-03-26
