Զѧ
ACTA AUTOMATICA SINICA
1997 23 3 vol.23 No.3 1997



ջϵͳʶƫС˷1
šӱ봿
ժҪоɫŶ·δ֪ıջϵͳƫʶ⣬һƫС˷.Ӧַδģ£ɻǰͨжģͺͷͨпģͲĽƫ.
ؼʡջʶС˷һ¹.
BIAS-COMPENSATED LEAST-SQUARES METHOD FOR
IDENTIFICATION OF CLOSED-LOOP SYSTEMS
ZHANG YING FENG CHUNBO
(Research Institute of Automation,Southeast University,Nanjing 210096)
Abstract This paper studies the problem of indentifying closed-loop systems in the presence of colored noises.A bias-compensated least-squares method is developed,which is able to estimate consistently the transfer function of the plant and controller in the closed-loop system,even if the models of the noises are not known
Key words Consistent estimation,least-squares method,closed-loop identification.
1
ڹ̵ҪջϵͳĹһֱܵǵĹ㷺ӣ1.Ŀǰмջʶֱӱʶӱʶ.Щľʶģ.(IV)ڲģʵֲһ¹ƣڱջϵͳҪҵһʵĸʮѵģ2.
ʵУջϵͳǰͨͷͨж᲻ɱܵδ֪ɫĸţЩģֲܾȷ֪.ʱʶͬʱȡڶͿһ¹.ľһּӱʶջϵͳƫС˷.ͨС˹ֵĽԿⲿο롢ϵͳɲ£ϵͳֻݺзӵĲƴƫ.ˣӦͨС˷öģͲƫƣȻЩƵĹԼӱջʽС˹ȡƫõһ¹.ɱջʽĲƺͶģ͵ĲƣõģͲһ¹.
2
ͼ1ʾĵ롢ջϵͳ.ͼr(k),u(k)y(k)ֱⲿοźšϵͳǾɲ.v(k)w(k)ֱǰͨͷͨеɫ.ģGp(z-1)ͿģGc(z-1)ֱȡΪ½ṹ
(1)
(2)
ϵͳ¼裺
1)r(k)ƽ̣
2)v(k)w(k)ǾܶȵƽУr(k)ͳƲأ
3)ջʽA(z-1)C(z-1)+B(z-1)D(z-1)ȶA(z-1)B(z-1),C(z-1)D(z-1)ֱǻʵȶʽԣ
4)ϵͳṹģͲn,m֪ϵͳɱʶn>m.
ʶǣù۲ZN={r(k),y(k),u(k)}Nk=1(1)(2)ʽжʽA(z-1),B(z-1),C(z-1)D(z-1)еϵϣõǵһ¹.

ͼ1뵥ջϵͳ
ͼ1֪ջϵͳ롢ⲿοr(k)Ĺϵ£
(z-1)u(k)=(z-1)r(k)+(k),(3)
(z-1)y(k)=(z-1)r(k)+(k).(4)
С(z-1)=A(z-1)C(z-1)+B(z-1)D(z-1)
=1+1z-1++m+nz-(m+n);(5)
(z-1)=A(z-1)C(z-1)
=1+1z-1++m+nz-(m+n);(6)
z(-1)=B(z-1)C(z-1)
=1z-1+2z-2++m+nz-(m+n);(7)
(k)=-A(z-1)D(z-1)v(k)-A(z-1)C(z-1)w(k);(8)
(k)=A(z-1)C(z-1)v(k)-B(z-1)C(z-1w(k).(9)
ɼ1)3)(10),(11)ʽ֪(k)ͦ(k)ΪƽУr(k)ͳƲ.

T1=ۦT;T]=1,2,,m+n;1,2,,m+n],(10)
T2=ۦT;T]=1,2,,m+n;1,2,,m+n],(11)
T1(k)=-u(k-1),,-u(k-m-n);r(k-1),,r(k-m-n)],(12)
T2(k)=-y(k-1),,-y(k-m-n);r(k-1),,r(k-m-n)],(13)
(3)(4)ʽд»عʽ
u(k)-r(k)=T1(k)1+(k),(14)
y(k)=T2(k)2+(k).(15)
ӦͨС˷Ʀ1ͦ2
(16)
(17)
1)4)ɵ3
(18)
(19)


ɦ(k),(k)r(k)ؿ֪
r1=-ru(1),,-ru(m+n),0,,0]TR2(m+n)1,(20)
r2=-ry(1),,-ry(m+n),0,,0]TR2(m+n)1.(21)
ru(i)ry(i)źŵغ.
ˣ(10)(11)(18)(19)ʽ֪ӦС˷ֱӻùڦºͦõƫ.(6)(7)ʽͼ3)ͨӵķLS(N)LS(N)ʽA(z-1),B(z-1)C(z-1)ϵƫ.ʽĹӻƹӵķܶ࣬Ʃ糣÷MATLABоֳɵ㷨˴׸.
ͬ(18)(21)ʽ֪ڴʱӦͨС˷ֻܵõƫĦĹLS(N).ˣҪùD(z-1)ϵƫƣ跨æƫ.潫.
3ƫƷ
(19)(21)ʽֻ֪ҪƳغry(i)ɰʽæ2ƫ
(22)
2(N)r2ǵһ¹.
ۻr2һ¹Ƶķ.ȸһõ֤ģ4.
1.mrάGȵģmlάȾHʹHTG=0,lm-r.
Σ
T=c1,c2,,cm;d0,d1,,dm].(23)
(5)(7)ʽƳ2֮ͦĹϵ£
2=M+p.(24)

(25)
(26)
(27)
ˣ1ͼ4)֪һ2(m+n)(m+n)άľH
HTM=0.(28)
ǣ(24)ʽͬʱHT
HT2=HTp.(29)
HT(19)ʽ߲(29)ʽ
(30)
ʽṩ(m+n)Լǿr2е(m+n)δ֪ry(i),((21)ʽr2ǵıʽ).
ǡrTy=-ry(1),,-ry(m+n)],(31)
KT=In+m;O]R(m+n)(2m+2n),(32)

(33)
Ϸ֪ӦͨС˷ɻai,bi(i=1,2,,n)һ¹ƣЩֵ뵽MpУɵõr2ǵĹ
(34)

(35),(36)
Һai,biС˹LSi(N),LSi(N)(25)(27)ʽоMpĽ.
(22)ʽɵæ2ĹΪ
(37)
½ڽ֤BELS2(N)Ǧ2һ¹.
˦2,ai,bi,cjһ¹ƣ(5)ʽڶʽD(z-1)ϵdj(j=1,,m)һ¹.
4һԷ
Ϸ֪Ϊ˷ñķõĲƵһԣֻҪ֤BELS2(N)Ǧ2һ¹Ƽ.Ϊˣ¶.
1.ݳNʱ(37)(35)(36)ʽõĹBELS2(N)ʵ1ֵ2
(38)
֤.ɵڶڷ֪ͨС˷õĲai,bi(i=1,2,,n)ĹLSi(N),LSi(N)һµģ

ˣ(36)ʽеҲMһ¹ƣ.ͬ(37)ʽеҲpһ¹ƣ.ԣվѡNʱʱļΪޣӦTM=0.
(28)(30)ʽ
(39)

(40)
ۺ(37)(34)(22)(19)(40)ʽ
֤.
ɦ2,ai,bicjƵһ֪(5)ʽdjĹҲһµ.
5
ͼ1ʾıջϵͳ

ɫv(k)w(k)ֱ¶MAģ⣬v(k)=e(k)+0.5e(k-1),w(k)=e(k)+0.7e(k-1)+0.8e(k-2).e(k)ֵλɫ.
ⲿr(k)ȡΪֵλźţӦñķ10ʵ.1˹۲ݳȷֱΪN=200,N=500N=1000ʱֵľֵͷ.ݱ÷ɫ»ø߾ȵıջϵͳĹ.
1

Nture value2005001000
a1-1.425-1.4750.089-1.3980.063-1.3930.041
a20.4960.4830.0780.5140.0550.5110.052
b10.1730.1600.0740.1690.0670.6170.039
b20.1020.0980.0930.1050.0570.1070.043
c1-1.5-1.5260.097-1.5190.078-1.5150.054
c20.60.5870.0780.5830.0630.5780.035
d03.4853.4230.0833.4320.0573.4370.048
d1-5.422-5.5180.089-5.5140.071-5.5150.064
d22.152.1080.0772.0540.0602.0760.043

6
ӦļӱʶջϵͳƫС˷ϵͳⲿźſɲ£ԻöģͺͿģͲƫƣǰͨͷͨܵɫŽнģ.
1)Ȼѧص.
߼飺šӱ1967.1989ҵڶϴѧԶϵ1995ڸУԶоʿѧλ.ҪоΪϵͳʶźŴӦƵ.
봿19941.
ߵλϴѧԶоϾ210096
ο
1Gustavsson I,Ljung L.,Soderstrom T.Identification of Processes in closed-loop:Identifiablity and accuracy aspects.Automatica,1977,13(1):59-76.
2Soderstrom T,Stoica P G.Instrumental methods for system indentification. Berlin:Springer-Verlag,1983.
3Davis M H A,Vinter R B.Stochastic modelling and control.London:Chapman and Hall,1985.
4ݺ壬ιڣɿ.ϵͳ۵Ĵѧ磬1986.
ոڡ1994-06-23
