Ϣ
Information and Control
1999ꡡ284ڡVol.28No.41999



粹ļù
СУ
塡

ժҪ һģͶԱضбʶٶģͱʶһ вģͺ繲ͬɶıʶģͣһģһ ʽùСУƣ÷ڷԶҾнϸ߾Ⱥͽ ٶȣнǿ³ԣ 
ؼʣ 磬ãСУ
ͼţTP13ױʶ룺B

GENERALIZED POLE PLACEMENT SELF-TUNING CONTROL WITH
NEURAL NETWORK COMPENSATION
JIN Qibing1LI Hongru2GU Shusheng2
(1. Beijing Institute of Petrochemical Technology, Beijing10260 0;
2. School of Information Science & Engineering,Northeastern University 11000 6)
Abstract: The Controlled plant is identified using normal lin ear model, and then the deviation identified by linear model is compensated via a neural network. The identification model is composed of a linear model and a neural network. Based on this model, an explicit generalized pole placement self -tuning control algorithm with neural network compensation is proposed. This al gorithm is suitable for nonlinear system, and has higher precision, faster conve rgent speed and stronger robustness.
Key wordsneural network, pole placement, general minimum va riance self-tuning control

1
ΪܹƷԶӦƵľȺ³ԣ˴粹 Ӧƣ13ݣ[2]˴粹ԤƣԤⲹʱҪõδʱ̵Ŀ룬δʱ̵Ŀδ֪ģͨõĴǽǰһʱ̵ĿΪ֪룮ǰһʱĿвһ׼ȷˣһڲ.[3]˴粹ļУƣʶһֻܴﵽ̬һִ粹ļùСУƣ÷ڷԶҾнϸ߾ȺͽϿٶȣнǿ³ԣ
2ıʶģ
ԿɱʾΪ
y(k)=f[y(k-1),y(k-2),,y(k-n),u(k-d),u(k-d-1),
,u(k-d-m),(k),(k-1),,(k-nc)](1)
Уn,m,ncΪ״ΣdΪʱӳ٣(k)ʾţ³ģͶԱضбʶ
y(k)=-a1y(k-1)-a2y(k-2)--any(k-n)+b0u(k-d)+b1u(k-d-1)
+bmu(k-d-m)+(k)+cl(k-1)++cnc(k-nc)(2)
ʶԺõi(i=1,,n),j(j=1,,m),l(l=1,,nc), ijlͿԶkʱ̵ĶйƣֵΪyL(k),
(3)
ڷԡʱ估δģ̬Ӱ죬yL(k)Ͷʵy(k)y(k)-yL(k)һбƽΪyN(k), 
yN(k)=g1[y(k-1),y(k-2),,y(k-n),u(k-d),,u(k-d-m)](4)
(1),y(k-i)(i=1,2,,d+1)δ(4)ǸŵӰ죬õ
yN(k)=g[y(k-d),y(k-d-1),,y(k-d-n+1),u(k-d-1),,u(k-2d-m+1)](5)
ѵĿΪָ꺯
J1=min|y(k)-yL(k)-yN(k)|
ӶԵõ
y(k)=yL(k)+yN(k)+e(k)(6)
e(k)Ϊıʶ(3)(5)(6)õ

A(z-1)y(k)=z-dB(z-1)u(k)+C(z-1)(k)+yN(k)+e(k)(7)



(7)ΪõıʶģֽͣṹԷԡŵĶкܸߵľȣ[2][3]ҲƵĽṹ
Բǰ磬ҲԲö̬ݹ磬ǰ罫 ϶Ϊ˱ֲСȨֵٶȣԲ[4]еȨֵѵö̬ݹԱ(5)״εӰ죬[5]ѧϰʽз,ǵõ˽ϺõЧ
3粹ļùСУ
ڴ粹ģͽṹԣҪʽӦƣ[6]һּùСУʽ㷨㷨ܹ֤ȫֵԣΪȫֵԣľӦһ㷨˴粹ļùСУƣ
ָ꺯Ϊ
J2=E([P(z-1)y(k+d)-R(z-1w(k)+Q(z-1)u(k)]2}(8)
ʽУw(k)Ϊο룬P(z-1)Q(z-1)R(z-1)Ϊz-1ļȨʽ
Diophantine
P(z-1)=A(z-1)E(z-1)+z-dG(z-1)(9)
C(z-1)E(z-1)=F(z-1)+z-dN(z-1)(10)
degE=d-1, degG=max(n-1,np-d), degF=d-1, degN=nc-1
E=E0+E1z-1++Ed-1z-(d-1)
E(7)ʽ߲(9)(Ϊ˼д½ӱʽڵz-1ԣP(z-1)дP)
Py(k+d)=Gy(k)+BEu(k)+CE(k+d)+Eym(k+d)+Ee(k+d)
(10)ʽõ
Py(k+d)=Gy(k)+BEu(k)+N(k)+EyN(k+d)
+F(k+d)+Ee(k+d)(11)
(11)ʽȺұߵF(k+d)Ee(k+d)Ǻ޹ص, (8)(11)ʽ׵ u(k)·ȷ
Gy(k)+BEu(k)+N(k)+EyN(k+d)=Rw(k)-Qu(k)
(BE+Q)u(k)=Rw(k)-Gy(k)-N(k)-EyN(k+d)(12)
(12)ʽĿ̴(7)ʽõΪ
(13)
ʽyN(k)ҲԿɿɲţ (13)ɼȻ˾зԵв, ϵͳʽΪPB+AQͻԶʱһµģң yN(k)=0e(k)=0,ͿԵõʱĶ̣
ȶջʽT(z-1),õ¼÷
P(z-1)B(z-1)+A(z-1)Q(z-1)=T(z-1)(14)
u(k)(12), kʱ(12)ʽеyN(k+d-i)(i=0,1,,d-1)δ֪Ϊ ȡu(k),Ĳ´
(5)ʽ֪
yN(k+d-i)=g[y(k-i),y(k-i-1),,y(k-n+1-i),u(k-i),u(k-i-1),
,u(k-d-m+1-i)]
yN(k+d-1)yN(k+d-2)yN(k+1)kʱǰ롢뾭ѵ ֱã
һTaylerչ

õ
yN(k+d)=g[y(k),,y(k-n+1),u(k),u(k-1),,u(k-d-m+1)]
g0+g1*[u(k)-u(k-1)](15)
Kʱ̣u(k-1)ȡ뾭ѵ磬Ϊg0g1ȡ ǰɲμ[7], Զ̬ݹɲμ[8]
(15)(12), õ
(16)
(16)ΪĿu(k)ʵȡ̣
ӦƵĲ£
ٸļʽT(z-1)
ڲȡy(k)(2)ʽģͽбʶ
yL(k)y(k)-yL(k), y(k)-yL(k)Ϊ߱ʶ
(14)ʽP(z-1)Q(z-1)
(9)(10)ʽE(z-1),G(z-1),N(z-1)
u(k-1)u(k)뾭ѵõg0ȡg1
(16)ʽu(k)
ཫu(k)ʵʶ磮
k=k+1, תڣ
4о
Ĵʵ֤˱ȷԣһΪ֤

ʶģ͵ԲȡΪ׶һԽǵݹԪΪ10뵥ԪΪ2\ڲ2000ò롢ݶģͽԤѵŵֵΪ 0.2ļʽȡΪ

ԤѵģͲֱƿвУڹ 仯Լδģ̬Ӱ죬ǷɢģüùУƣ 粹ʾͼ1ô粹ļùУƵĽʾ ͼ2

ͼ1ùУƵĽͼ2粹ļùУ
ƵĽ
ɷԿķЧģнϿӦٶȣ
5
۷ͷ˱Ĵ粹ļùСУ ƷЧģȼУƾиߵĿƾȡӦٶȡ ³ԣڷԶ
߼飺,,27,ʿ.оΪ缰ģڶϵͳеӦ.
,,30,ʿ,ʦ.оΪܿƼڶϵͳеӦ.
,,59,ʿ,,ʿʦ,ѧϢѧ빤ѧԺԺ.оΪ ۼӦáܿơϵͳ.
ߵλʯͻѧԺ102600;
塢ѧϢѧ빤ѧԺ 110006
ο
1С,. Ԫǰѧϰ. Ϣ, 1994,23(5):311314
2Զ,,Ԭ. ĹԤ. ۼӦ, 1996, 13(5): 677680
3Fuli Wang, Li Mingzhong, Yang yinghua. A Neural Network-based Adaptive Pole P lacement Controller for Nonlinear System. International Tournal of Systems Scien ce, 1997,28(4):415421
4Robert S S, N Tepedelenliogln. A Fast New Algorithm for Training Feed Forward Neural Networks. IEEE Trans. on Signal Processing, 1992,40(1):2022 10
5ChaoChee K, Y L Kwang. Diagonal Recurrent Neural Networks for Dynamic Syst ems Control. IEEE trans. on Neural Networks, 1994,6(1):144156
6Tianyou Chai. An Indirect Stochastic Adaptive Scheme with On-Line Choice of W eighting Polynomials. IEEE Trans.on Automatic Control, 1990,35(1): 8 288
7̷. BPӦ. ۼӦ,1994,11(1):84 87
8Mingzhong Li, Fuli Wang. Adaptive Control of Black-Box Nonlinear Systems Usi ng Recurrent Neural Networks. 36th IEEE CDC, 1997,December: 10-12, California US A
ո:1998-06-15
