Զѧ
ACTA AUTOMATICA SINICA
1997ꡡ236ڡVol.23No.61997




һŶķϵͳ
Ĳ
ر÷־
ժҪоһжʱŶϵͳƵĶʽƽ.һļ¸ʱŶŶʽŶĿԤ.˵˸÷Ч.
ؼʡɢϵͳNarmaxģͣŶʶ.
PARAMETER ESTIMATION FOR NON-LINEAR SYSTEMS
WITH A KIND OF INPUT PERTURBATION
QIN BINYANG YANMEI HAN ZHIGANG
(The Institute of Applied Mathematic, Heilongjiang University, Harbin150080)
AbstractThe approximation method based solution to parameter estimation for nonlinear systems with independent input perturbation is studied. The form and the predicability of output perturbation caused by time-variant independent input perturbation are presented under some hypotheses. Simulation examples demonstrate the effectivence of this method. 
Key wordsDiscrete-time nonlinear systems, Narmax model, input perturbation, parameter estimate.
1
ϵͳʶУŶ()ձڣҪԴ
1)źʱƯƣ
2)ܿص룬ϵͳܵźžΪʹ.
ָŶΪڶ.Ŷʵձڵ.ģ1ٵϵͳִĿΪͶǣִʻжǽܵˮ˵ĸ.ƵľжŶĿϵͳܶ࣬¯ƹϵͳѹܱڸֵʱ;Ŷ绯̵ķӦϵͳѹܵӰ죬ӰҲɿŶ.ϵͳУŶԵӵΪΪϵͳʹĵǱȽϼ򵥵ģ2.ڷϵͳȻṹӣŶϲŶʽͱȽϸ.ĲöʽƽķŶõʽһַϵͳȷԲֵĲŶֲͬʱʶķ.ƵķڽϵͳĲʶʱõ34ݣǿǵһвع۲ϵͳıʶ.ȻŶصģģ34ݵĽԱĵ.
2
ģ5ָķϵͳNarmaxģ
1)нϵͳнģ
2)ƽ㸽ϵͳƽ.
NarmaxģΪ
y(t)=Fy(t-1),,y(t-Ny),u(t-1),,u(t-Nu)+e(t). (1)
y(t)Rn,u(t)Rm,e(t)RnֱΪϵͳͰNy,NuΪF . δ֪ķԺ.
ٵʵԼϵͳŶʽ
u(t)=(t)u(t-1), (2)
Ц(t)RmmΪʱתƾ.(1)(2)ʽɵµĴŶNarmaxģ
y(t)=Fy(t-1),,y(t-Ny),u(t-1)+u(t-1),
,u(t-Nu)+u(t-Nu)+e(t). 
Ϊд㣬y(t)дy(t)
y(t)=Fy(t-1),,y(t-Ny),u(t-1)+u(t-1),
,u(t-Nu)+u(t-Nu)+e(t).(3)
Ե뵥ϵͳΪŶʽ.
1.(3)ʽķϵͳµĶʽƽ⾫ȣ
(4)

(5)
֤(ɲμģ56).
1.(3)ʽϵͳЧƽ¿Էֽʽ
y(t)=Fy(t-1),,y(t-Ny),u(t-1),,u(t-Nu)+gy(t-1),,
y(t-Ny), u(t-1),,u(t-Nu), u(t-1), u(t-Nu)+e(t). (6)
g . ΪF . ݾķԺжʽչÿһ
(7)
BL1niNu, LΪʽչĴ.
֤.1֪(3)ʽд(4)(5)ʽĶʽʽ.(4)(5)ʽһչ
 (8)

(9)
 (10)
ʽmΪŲչs+T=q, T0; 1nulNu, 1nuNuඨͬǰ.Ƚ(9)(5)ʽɵ
y(t)=Fy(t-1),,y(t-Ny),u(t-1),,u(t-Nu)+gy(t-1),,
y(t-Ny), u(t-1),,u(t-Nu), u(t-1), u(t-Nu)+e(t). 


֤.
ɶ1֪һŶķϵͳĶʽЧرƽ
(11)
ǵΪʶi.ȡiδ֪ģX2i(t-1)ΪЦu(t-nuh)Ҳδ֪.(11)ʽԽһֽ⣬õ
(12)

һ
y(t)=W(t)G(t)+e(t). (13)
 W(t)=(X11(t-1)X1n(t-1)X31(t-1)X3m(t-1)),G(t)=(12nȡ1(t)ȡ2(t)ȡn(t)).
֪(13)ʽԲС˸ʽ7.Ҫ(13)ʽıʶ.
3ŶķϵͳĲʶ
(12)ʽ
 (14)
(2)ʽ֪ u(t)=(t)u(t-1).

ʽ(14)ʽ
 (15)
(13)(15)ʽ֪еʱȷʱ̣1nΪʱ.ôӵС˷ʶG(t)õ(t)еķ1nǹȫŵ.
ԴˣǲµĸĽ.
1) ôӵС˷(13)ʽвʶõ(t), t=1, N.Ԥ
(tt-1)=y(t)-W(t)(t).
ȡ(tt-1)(ֵ)Ĳʶ
(t), t{ky(k)-W(k)(t)}.
Ȼ1,nƽֵi, i=1,,n.
2) i(13)ʽ
y(t)=W(t)(t)+e(t). (16)
 (t)=(12nȡ1(t)ȡm(t)) .(15)ʽôӵС˷вʶiΪ֪Wi(t)˺Ƶ̵.ֿɵõڦȡiNơi.
3) i(13)ʽɵ
y(t)=W(t)(t)+e(t). (17)
(t)=(12n1(t)m(t)).
ٲòӵС˷(ȡ1)вƣɵõii=1,,n.Ȼ1)ͬİ취i.
4) ظ2)3)ֱ.
ӵС˷Ϊ8


4ʶ㷨ͳ
(12)ʽ
y(t)=WӦ(t)G(t)+(t) .(18)
 (t)=Wb(t)Gb(t)+e(t), G=(12n)\+,
Wa(t)=(X11(t-1)X12(t-1)X1n(t-1)), 
Gb(t)=(ȡ1(t)ȡ2(t)ȡm(t)),Wb(t)=(X31(t-1)X32(t-1)X3m(t-1)).
(18)ʽдɾʽ
Y=WaGa+. (19)
 Y=(Y(1)Y(2)Y(N)), Ga=(12n), =((1)(2)(N)),

(t)=Wb(t)b(t).㷨1)-4)(15)ʽͨʵѡӿɵ.
2.㷨1)-4)õĹ(t) E{-(t)}M+.
йôС˷ʶʱۺܶ࣬нģ178о.
2.ڴŶķϵͳ(3)㷨1)-4)вʱ 
E{a-Ga}R+ .
֤.(19)ʽõGaС˹ֵΪ a=(WaWa)-1Wa(Y-).
(19)ʽʽ
a=Ga+(WaWa)-1Wa(-), a-Ga=(WaWa)-1Wa(-).
ʽȡ
E{a-Ga}=(WaWa)-1WaE(-).
2֪ E{a-Ga}R+ .֤.
5
ŶSISOϵͳƵֵʵ.ϵͳģΪ
y(t)=0.5y(t-1)+0.8y(t-1)u(t-1)-0.4y(t-2)u(t-2)-0.5u2(t-1)+e(t) .
 e(t)Ϊֵ.ŶΪ (t)=sin(t/100)(t-1)u(0)=0.1.ӦĴŶģ͵ķֽʽΪ
y(t)=0.5y(t-1)+0.8y(t-1)u(t-1)-0.4y(t-2)u(t-2)
-0.5u2(t-1)+0.8u(t-1)y(t-1)-0.4u(t-2)y(t-2)
-u(t-1)u(t-1)-0.5u2(t-1)+e(t).
дС˸ʽ
y(t)=Wa(t)Ga(t)+Wb(t)Gb(t)+e(t)=W(t)G(t)+e(t). 

W(t)=(Wa(t)Wb(t))
=(y(t-1)y(t-2)u(t-1)y(t-2)u(t-2)u2(t-1)y(t-1)y(t-2)u(t-1)1),
G(t)=(GaGb(t))
a1a2a3a4
=(0.50.8 -0.4-0.50.8u(t-1)-0.4u(t-2)-u(t-1)-0.5u2(t-1)).
1)-4)㷨Ʋ2)3)ظκõͼ12ʾĽ.

ͼ1ֵͼ2Ԥʵ
ͼ34ΪôС˷ƷϵͳNarmaxģͲĽ.ԿϵͳŶʱƳַɢ.ͼ5ΪϵͳŶ.

ͼ3ֵͼ4ʵʺԤ

ͼ5ϵͳŶ
1) õȻѧйʿ.
߼飺ء1966.1991ۼӦרҵ˶ʿѧλ.1996ڶѧԿϵʿѧλϺͨѧԶϵʿվо.ҪоΪϵͳʶӦƼ³.
÷1966.1991ۼӦרҵ˶ʿѧλ.ΪѧʦѧԶϵʿо.ҪоΪϵͳʶ״̬.
־ա1934.1958ҵڼִѧѧϵ.Ϊѧڡѧְڣʿʦ.ҪоΪϵͳʶӦơݽ׷ģͿ.
ߵλѧӦѧо150080
ο
1ݡLjung L. System identification theory for the user. Englewood Cliffs, New Jersey: Prentice Hall1987.
2ݡء.ƵУ-¸˲.ۼӦã1992767-71.
3ݡBillings S A, Jones G.Orthogonal least-squares parameters estimation algoritms for nonlinear stochastic systems. Int.J.System SCI., 1992, 23:1019-1032.
4ݡJones I J, Billings S A. Mean levers nonlinear analysis and identification. Int.J.Control, 1993, 58:1033-1052.
5ݡChen S, Billings S A. Representations of nonlinear systems:the Narmax model. Int.J.Control, 1989, 49:1013-1032.
6ݡLeontaritis I J, Billings S A. Input-output parameteric models for nonlinear systems, part: deterministic nonlinear systems; part:stochastic nonlinear systems. Int.J.Control, 1985, 41:303-344.
7ݡ־.ݽ׷Ӧ.ѧ磬1989.
8ݡGoodwin G C, Kwai S S.Adaptive filtering, prediction and control. Englewood Cliffs, New Jersey: Prentic Hall, 1984.
ոڡ1995-09-07
