Զѧ
ACTA AUTOMATICA SINICA
1997ꡡ236ڡVol.23No.61997




Hޱʶ㷨һŵϽ

ժҪñƽеn-ȺBernstainʽһԵĴϵΪ³ʶеHޱʶ㷨һŵϽ纯. úǴϵ͹ڼⴰϵӦıʶϽ磬ΪŻѡHޱʶ㷨Ĵϵṩ˿;. 
ؼʡ³ʶ µȷͱʶ Hޱʶ Hޱʶ㷨.
A QUASI-OPTIMAL UPPER ERROR BOUND FOR TWO-STAGE H
IDENTIFICATION ALGORITHMS
WANG SHUNING
(Dept. of Automation, Tsinghua University, Beijing 100084)
AbstractBy making use of n-width in approximation theory and Bernstain's inequality, a quasi-optimal upper error bound for two-stage H identification algorithms of robust identification is established in this paper, which is an explict function of general window coefficients. The function is convex with respect to window coefficients. It not only can be used for computation of upper error bounds corresponding to any concrete window coefficients, but also supplies a feasible way for choosing window coefficients of two-stage H identification algorithms with optimization techniques.
Key wordsRobust identification, worst-case/deterministic identification, H identificatioin, two-stage H identification algorithms.
1
Hޱʶ㷨ף1ݣף2,3ݵķչΪƵH³ʶһ㷨.񣬶Ը㷨о϶ڶԾĴϵƵʶϽ磬֤Ӧ㷨. ЩϽһȽϱأ2ݣܾСȽϲͬϵ.ΪˣһԴϵΪһŵϽ纯.ΪⴰϵӦıʶϽṩ˱ ֱŻѡ񴰿ϵ. 
Ĳ·ԼCNʾʼ±Ϊ0ȫNάļϣLޱʾ㡬hL:=h(z)޵ĸ亯ɵĸռ䣻QN:={h,hLޣh(z)=hkzk};Hޱʾ㡬hH:=h(z)޵Ľɵĸռ䣻PN:={h,hHޣh(z)=hkzk};ڡhH=hL,hH,ͳһáh:=h(ejw)ʾL޺H޵ķjʾxCNxkܱʾhHޣhkܱʾTaylorϵ.
2Ԥ֪ʶ
2.1Hޱʶ
豻ʶϵͳΪɢȶʱĵϵͳ. ٶ1)ϵͳӦ{hk}k=0ɵĴݺh(z):=hkzkģͼH(M,):={h,hHޣh(z)M,z}M0ͦѣ12) N0ɵõȾƵ۲EN(h,)CNENt(h,)=h()+t,0tN-1ЦǡBN():={,ǡCNtܦ,0tN-1}δ֪ţ0. Ҫ1)㷨:CNHޣʹ(EN(h,))ƽh; 2) ȷ¿ܲıʶeN():=sup{(EN(h,))-hޣs.t.hH(M,),ǡ
BN()}Ͻ磬eN()ɷNͦ㣬ʵ㷨Ϊ㷨. 
2.2Hޱʶ㷨
FN:CNCN ʾɢFourier任FNk(y):=0kN-1yCN. m0kN-1涨FNk+mN(y)=FNk(y).ӳnx1:CNL޺*:LH£

nΪ{xk}n-1k=-(n-1)ΪĴϵ. νHޱʶ㷨ӳnx2(EN(h,)):=*(nx1(EN(h,)))ģ. ֤3
eN(nx2)2eN(nx1).(1)
⣬*ȫNehariȷ4.ɼnx2ʵֻȡڴϵxѡ. 
2.3
1.5M0,ѣ1NhH(M,)

ЦN:HH޶Ϊ(N(h))(z)=(1-2(k-N))hkzk.
2.aCNbCN

HN(M,):={h,hPN(Nk)2hk2M2},Nk:=k(1-2(k-N))-1,0KN-1.
֤.Cauchyʽɵ

(2)
⣬ttֱʾatbtĹ. ȡh0PNΪ֤
h0HN(M,),0BN(),(3)
(4)
(2)(4)ʽ֪. 
3.hQNn4(N-1)n

֤.벻ʽȻ.ֻ֤Ұ벻ʽ.ѿ
h(ej)=f(),0ܦء2.(5)
0ln-1f()ʵ鲿ڴһTaylorչBernstainʽ6ݣ

Ƶ

ʽ߶lע⵽,ɵ
 (6)
󣬽(5)ʽ(6)ʽɵ֤֮ʽ. 
3Ҫ
Ϊ㣬Hޱʶ㷨еnȡΪNn=mN, m1ƴ. 
1.㷨

ЦǡBN(+M-N)}.
֤.hH(M,)ͦǡBN() 0tN-1. 1֪
h-N(h)ޡM-N, dBN(+M-N).(7)
⣬Parsevalʽɵ

ԣN(h)HN(M,).ע⵽EN(h,)=EN(N(h),d), (7)ʽƵ

2.m4У

Ц0(x,):=x0k(x,):=xk+xk-Ne-jN, 1kN-1.
֤.hHN(M,)ͦǡBN(+M-N)

Եõ

ע⵽FNk(EN(h,))k0kNʱĹ涨ɵ

ʽ2ֱӵõ
(8)
һ棬Nx1(EN(h,))-hQN 3ɵ
(9)
ע⵽,أ(9)ʽhͦȷ磬(8)ʽɵõĽ. 
϶(1)ʽɵ. 
1.һ鴰ϵx
 
mΪm4е. 
4
Ϊ1ȡֱصĴһ1h(z)βתΪһH޷M-Nδ֪ڶøhtĲʽԼ2tht2M2治״ԭʼԼhttejtأM,.һCauchyƦthtM,t.CauchyƿԴӦ2tht2ܦ2tht2M2ƳCauchyƳֻܵõβϽhtejtأܣhtM-N(1-p-1)-1. ˣĽϽӦӽȷ. 
⣬x͹͹滮һЧ㷨. ˣ1Ͻ粻Ϊһ봰ϵӦıʶϽṩ˱ΪŻѡ񴰿ϵṩ˿;. 
1) õȻѧͽίʿ.
ߵλ廪ѧԶϵ100084
ο
[1]Helmicki A J, Jacobson C A, Nett C N. Control oriented system identification: a worst-case/deterministic approach in H. IEEE Trans. Automatic Control, 1991, 36(10):1163-11276.
[2]Partington J R. Robust identification and interpolation in H. Int.J. Control.1991,54(5):1281-1290.
[3]Gu G,Khargonekar P P. A class of algorithms for identificaiton in H. Automatica, 1992, 28(2):299-312.
[4]Young N J. An Introduction to Hilbert Space, New York: Cambridge University Press, 1988.
[5] Gu G. Suboptimal algorithms for worst case identification in H and model validation, IEEE Trans. Automatic Control,1994. 39(8):1657-1661.
[6]Beckenbach E F, Bellman R. Inequalities, Berlin: Springer-Verlag, 1983.
ո 1995-09-06
