Analysis and synthesis of optimal and robustly optimal control systems under bounded disturbances

Analysis and synthesis of optimal and robustly optimal control systems under bounded disturbances

14th World Congress ofIFAC ANALYSIS AND SYNTHESIS OF OPTIMAL AND ROBUSTLY OPT... F-2d-02-5 Copyright © 1999 IFAC 14th Triennial World Congress~ Bei...

3MB Sizes 0 Downloads 44 Views

14th World Congress ofIFAC

ANALYSIS AND SYNTHESIS OF OPTIMAL AND ROBUSTLY OPT...

F-2d-02-5

Copyright © 1999 IFAC 14th Triennial World Congress~ Beijing, P.R. China

AN ALYSIS AND SYNTHESIS OF OPTIMAL AND ROBUSTLY OPTIMAL CONTROL SYSTEMS UNDER BOUNDED DISTURBANCES Vsevolod IV1. Kuntsevich * Alexei V4 Kuntsevich **1 1

* /,pare Research Institute; 40 Pro.'ipekt Akade.,nika GZnshkova, 252022 Kiev, [Jkru?:ne~ E-rnail: vmkun@d305. icyb ~ kiev. ua ** Iust£tutc fa'" J.~1athB1natir.sJ UniveT.~~dty of GTaz, He1:T1:ch8tr.;S6, A-80l0 Graz, A 1J,Str£a, E-1nail: alex@bedvgm. kfunigraz ~ ac . at

Abst.ract.: Analysis of dynarnic (:loscd-loop control) systcrIls nnder bounded disturbancc~ by calculating HLinirnal invariant sets is di~Cllt1Sed. It i::; suggested to HH8 th~ radiuH of a. 11lillilUal iIlvariant ~et a~ et particular llUIIlCrical nH~RSllr~ of the stability degree of a systenl a.nd ::tpply sHeh a criterion for optilnallincar control synthcsit; by rneans of lninirIli~ing thifi radius. It has been proven that. the radius of a. IIlinitnal invariant ~et rea(:he::-; il.~ lIlininllun at. a. nilpotp.llt paralllctcr rnatrix of a systelu. The sollltion to the optirna,l control synthesis problcrn ean be fOUIld~ therefore, by calculating" the linear feedback pal'arnctcrs providing the nilpotcnt ruatrix of the closed-loop t'iy~tcln. In the caHe of RcaJar controls: t.he problelll has got an analytical solution. (~opyright(g 1999 IFAC I(ey,vords: control synthesi~~ optirnal control~ invariaut, sets, bounded distllrbauce.s~ di::;cl'ctc HyBtclIlS, set val1les.

1.

I~TnODlJCTION

Set-valued rnodels ha.ve been

'\vid(~ly

applied to ~ysteIIl identificat.ion and observation, rolHlst. a.lld adaptive control synthcsi~. Set-lnernberHhip ident.ification for paralnet.er~ and perturbation~, rospectively \VOll:lt-eas<'. anal,ysi~: ha~ beeorne a regular Lool in IlloderIl cont.rol theory and reqnires cornp]et.e]y diffnreut Inethods cOlnpared to average case a.naly~is. No\vadays~ the so-called hard bound approach, \vhich u~e~ only a priori g-iV(~Il hounds on dist11rhance~, nnkno~"ll systerIl pal'alneterH and rnodel errors, attracts Illore and Inure InaJhcluatiCi
1

Supported by the Special R.Bsearch CenteT" F ~OO:l and the

contrarYI considerably less effort~ have been paid

to apply the hard-bound approaeh to a.utornatic control synthesis. The reason for thi,s goes hack to t.he necessity of calculating IniniInal invariant sets. A~ S0011 as an invariant. set ha~ hc{~n dcterrnincd, onc obtains a IIle~llre uf t.he dyualnic pI'p.cision of a 8'ystenl~

which allows

UH

to estiInate the rnaxilnal

deviation of Holl.1tious froIn a lloluinal solution and can be utlcd the sarnc ""Tay a~ a lIlcan sqHar(~ deviat.ion in stochasti(: dynaInics. Actually, t.he f~xistenc(l of a hounded inva.riant set (the property of the nlthnatc boundedne~s or dissil)at,ivit:y) i~ onc of the Ino~L ~ig;nificallt gronp pl'op~rties of tha.t class of eloxed-loop control ~YHtenu·; (dYllalIlic plantH) df~tel"tnilled by noise and/or paraII1et(~rs \vithin the given bounds. Apparently, the problern of calcnlating- an invariant set for a plant under bounded disturbances has been posed for the first tirnc by Bulgakov (1946). Fllrt,her~ Y-oshi~a\Vn (1 g(j{j) and 13arbashin

J F -NBA project No. 6243, Austria

2830

Copyright 1999 IFAC

ISBN: 0 08 043248 4

ANALYSIS AND SYNTHESIS OF OPTIMAL AND ROBUSTLY OPT...

14th World Congress of IFAC

\vhere L == A -+- BC.

(J 970) have been uHing the apparatns of f.i.VaPllILOV functions for calculating stable litnit eyelet;. Tlley ~ucce(lded to obtain 'SOlllC significant result::;) parti(:1 tlarly for lloulillcar sYHtelns~ _HO\VCV(-H' ~ the.'~e reHults revealed H01UC Hcriouo disadvantages of the approach. In fact 1 the final SIlCCCS~ efo:st~IltifLll'y clepCll(h.; on t.he choice of a particular L,yapll110V function. The early results un calculating rnininlal invariRnt sets hy 11Sl~ of an (Ixact alv.;ebraic fOl"111U1<\ hrBt prov(~d ill l(llutscvich alld Pshcnitchnyi (lg96a) can be fOllUd in K llllt~cvich and P.sbcnitchll.vi

(199Ch, 1997). Here Lhe results by K 11l1tsevieh and Pshcllitehuyj (19gGa, 199tib, 1997) arc: (~xtended to'\.vards Syllthet'is of optilnul and robl1~tly opt-inIal control HystCIIlS. \\lith the aiIH to a.pply the apparatus of Ininirnal invariant sets to tll(-~ control synthesis prohlern, the radius of a lninitnal invariant t-:(~t i~ llsed lli"4 a critcrion.'rhc key theorern clnillls that tIle radius of a rnln]rnal invariant. set reachc.s its rniniInurn at the llilpotellt par::nnetcr Inatrix. Hence the pl'obl(nn uf uptiInal control ::;ynthc~ifi takes the [orlIl of calculating the feedback pfLr(=unctcrs providing the llilpotcnt Inatrix of the closed-loop cOlltrol SystCIIl. 'The solutioIl to this pl'ohlenl fol' the case of :-;(~alar conLrols i~ caleulated allalyti<:ally.

III vic\v of the fact tha.t our rnailL g-oa]

i~

to

develop cOlnputational rn(-~thodH, wc present in the follo\.ving the Inaiu (:OUC(~pts for discrete-titHe ~YStCIlt8.

2. i\NALrYSIS OF DYNJ~::VIJC Sy""STENfS UNDER BOUNDED DIS1"1URBANCES 2.1

(~1o.,lr:1J,lat1:·ng

thp nd'll,i'rual

inva.'riani

set of a

linp'fl'" di:..;t'JJ:rberl S?J,lJtp'ln

Consider the cla~H of by the equation X n ---i-l

:=:.

AXt1

lin(~ar

+ BfJn +

control S,vsteIlIR given

P'n.~

n

=:

O~ 1, ... ~

corn pac t set.

[In

1~

i

1 ~ . . . ,.In.

(G)

,ctSSlnnc a systern (1) given by rnatriceH A and KahIHul~S criterion. Hence the ~et of Inatriees C provi(lill~ t.he f\11fi1rnent of the condition (5) is not ernpt.y.

B is controllable according- to

Introd 11C(~ t he set

"'-There the ~ll1n of Hcts is a lvlinko"vski SUII!. R,Peall, ::-let. rrn iH called all invariant Het for the SyStCII1 (12), if for .."Y E 9Jt~ the condition LX + F E 9J1 is fulfilled vvith any F E ~, i.e., L9J1 -+- ~ ~ 9]1".

Defl:nition 1. A cOlnpact set is callc~d a IninhnaJ invariant set of cl, systetIl: if it is invariant and ductS not cont:i.,in an invariant proper subset. The follo\ving t'~vo thcorerlls lul.v(~ b(l{~:n proven by K untsevi(:h a.ud f>sheni tchnyi (1997). Thp-o'rfJn 1. Under the ?.Hs\l1nptioll (5) l the Het 9Jt(L,.;f) given by fOI'nulla (6) is a cOlllpact set

in JRm,. 'Theot'P,"rn $3. 9Jt(L, J) given by forrnula (0) is a rninirual invariant ~et for a SYSt.CIIl (4). If X o 19)1(L,3), theIl for any E > 0J aB the ~t.a.tcs of Hystcrn (/1) ar(~ inside the E-ll(~ighhol'hood of m1(L, 3') ~tarting at a particular discrctc-t.iIIlC lI10lIJent \vhich depends on E.

TheorelllS 1 and 2 deterruine a convex rninilllal invariant set. at-: Hoon a..s J" is a convex ~ct. Note that jf a nl8,trix A is nilpotent, t.he series (6)

finite.

Calculating rninirnal invariant Kets of linear llOl!(closed-loop contrul) sy~Lcln~

~ta.tional"Y dynfI..lnic

under bOllndcd di~tllrbanees are discllssed here by recollecting the resultB by KuntHevich and Psheuitchllyi (1996b, 1997). sy~t.cnl~

it) p.;iven by

the equation

llBCd ,

X n+l systcrn tak(l8

n~O,l, ...

==

\\~e

~ CXn.~

cl()s(~d-loop

Xn+l=:::::LXn+l
<

Assurnc the clas::-; of dynanlic

\Vhcn a lineal' control feedback is

eqnation of a forln

fJ

(1)

(2)

t.lH~

IA,: (L) I ~

i~

\vherc X n and [In cU'C a st.ate J1J,-(hlnel1:::1ional vector ~ r(~.sp. a t:ontrol p-dirnellsional vector, at the nth discrete titne iur;tant, A and B are the lIlatricCR of the rc.spcctive dilucn::..;ion~ and F n is an H1.-(liIuCllSional v(~et()r of l)(nln(l(~d (listllrhant:(~s:

J is a convex

Independent frotH other rnatrix properties v-rhich IIlight be needed later, L is &~sluned to be a SchurC~ohll lllatrix, i.c' l all eigcnvalnes are in the open unit circle:

,

th(~

=::::

A''2.Xn

+ BUn + j-!"'n.,

11,

==:.=

OJ 1~ ... ,

"vhere all the notations arc the H9.l1l(' as in (1), except An is a Inatrix of tirne-'varyiull," pala.nlC1.(~r~: o

(4)

An

=:=:

A

+ bAn~

n == O. 1, ....

2831

Copyright 1999 IFAC

ISBN: 0 08 043248 4

ANALYSIS AND SYNTHESIS OF OPTIMAL AND ROBUSTLY OPT...

14th World Congress of IFAC

2.4 In1)Q,T1:ant set..,' of specia.l tllpe 'fI,on!£ncu','

o

I-icrc A i::-; a constant rnatrix and

bAn

E {;21 =

{8A:

IlfJA~I:;

rr},

n.

=:

O~

,'i1f8t~'fTr.8

Consider t.he class of systelIlt-i given by the equa-

1, ....

tion .A.. HSllnH~ linear fc(~d hack (3) fol' this cl ass of s,ystenlH and cOllt'idcr the clo~('d-loop cOlltrol S:y~tcln:

(7)

(11) vlhcrG ~ (-) ic-; a nonlinear vector function 1ft"n R m . sati,sfyiug the inequality

\vhcrc

(12) o

Ln

=::::

A

+ cn + tAn?

n

=.

and the disturbances F n are hounded hy (2).

D, 1, ....

o

Agaiu) aSSllIllC L --= A + CB is 8. Schul'-Cohn IIlatrix (S{X~ (5)). Introduce th(~ €-ll(~ighborhoodof t.he set ~:

~T here <:3

iH a 'lui t solid Hphcl"c with the center at

tJnder the condition (12) ~ a SystCIIl (11) has a. invaria.llt set. Ho\vever, onc canllot calculate thi~ ~et explicitly in contrar:v to the ca.-,-;e of linear S.ystCIllS (4). OIl the othel" hand, ~iucc one has got an upper bouud for the vallleH of ~(Xn), it. bccoInes pos~iblc to estirnate the radius of a lninirna.l invariant set, bOllnd(~d

the origin, and the Het

9Jt(L, J=)

---1-

::=

r(9J1)

9J1(L,~) +.=.6

+ EL6 + EL 2 6 + ... " (8)

K ll11tscvich and PSh(~Ilitehnyi (1997) proved that (8) is a IlIillirnal iIlvariaut set for the class of llOlltltationary :-;yst.(~!ns (7) for any ill(~q llali ty

1- q

iT

E Hat;i~:,;(ving

tIle

(1'

:

X E m1},

by the fono'\vinp; iterativc' procedure. ASSUlue .9J(0 ==.J"~

r( 9J1 o) == r( lr) == p-

+

Calculate 9R 1 ==:
~.

The

l'rLdill~

of 9Jt 1 is

obvjou~l.y

r-(9Jt 1) S; r( (~))

+ P :S () +

fLp.

f:

< ------

= 11lax{!lX I~

d(J)

~AJtc;r

+ E'

where d('J) i~ the dicunctcr of the set J' and (x is a C()n~tallt providing the flllfilrncuL of the ineqnality L n ::;; ()'qH. rrhe latter is correct. for Cl lnatrix satisf};iug the eOlldition (G) and a constant er > 1.

2.3 Inva.,·ia'll.t sets of d,!)na1n'£r;

8JJ.··dern.., 1j}ith

c~tilnating

the ra.dius Y"(ffit k ) by its up2:,i=O, ... lk 1".1, estiInating r(9J1,l,~+l) is stl'aightforwar
(J

'111€O'reu,. 8. The radius of the lninilual inva.riaut set flJt(
bonndAd

'f},oulin t-:nT pa'f't,.c;

The follo~"'ing class of SyHtcIIlS have been analizcd by KUlltscvich and Pshenitchnyi (199ua).

\vhere L rneet.s the couditioll (5) Ctud cfJ (-) is Cl. ------) R'Tn, there\vith

continuous vector fUllctiuIl R 1n

(10) i~ a cUIIlpaet

set ill

ll{'m.

Here e1(-)

IneallS

closillp.;

a. set.

It is ~trai~htfor"val'd that the syHteYflS (9,10) and (4,2) :it c
l\ nal:'lsis of 1l0Uallt.OllUlIlOtlS ~:V~tclIl::; v.,rith bOlllldccl lLonlillear parts leads to the sirnilar rc:snIt (S(:c K \l.llt.~evich and Pshenitchnyi (1 ggoa)").

2.5 M'ini'fnunJ, l'adin:; in1JaTiant set lfercinaft.er, t.he radius of a tninilual invariant set (or an cstirnatc for thit; .set) iH Hoed &'-) a c.ontrol pcrforlnaIlcc criterion.

A rninilnal invariant set, ~'"hich has rniuilnal ra.dius arnong- all other invariant sets, \vill be called a rnr:rl:irnu.fn 1'adiu.r,· iuvar-£anl ~,,·et. C~onsidcr

a control systenl (4,2) vlith 11

L==

rrIcCining

L

takc~

~l

1

II~ ~~

() 11

.(~l

tlle Htandard

(13)

Frobcniu~ forlll.

Think of L as the Hlun

L

=

Lo+ L

n?

2832

Copyright 1999 IFAC

ISBN: 0 08 043248 4

ANALYSIS AND SYNTHESIS OF OPTIMAL AND ROBUSTLY OPT...

of a uilpotent C-"vith the nilputeucy index ulat.rix

I~~)

14th World Congress of IFAC

TheorCTll 4 clairns the luinirnlllU raelius invariant set at L ::::::; L0 7 i.e.)

Hi.')

L m.

1

-=.:.::

Am,

I

bm,C ::::::

o.

Hence

L o == 110 () 10 0 1

'

and a luatrix

For vector controls: the synthesiH problcrIl canIlot be solved that elegantly and easily it is a particular ca~c of a l110dal control K:yllthesis p1'()hl~ln (se~ Klluts(-~vich aud I(untsevich (1994 1 1990) for instance). The solubon of the s.ynthesis problcIll in this ease ccrtaillly eaullot be given ill explit:it forIn and \viIl require solving rather (~OTnpJcx ntinirnax

l}/,r:orprn 4. DJt( L o , J") i~ a rninirn111H radius invariant. ~et uf ~ystcrn (4~2~ l:j).

probleJlls. rr"'heorerIl 4 claiInti that InininnuH of the radius is rCRched at the Ililput.ent par<11Ilct.cr IIlatrix L =

;).2

L o·

SJJrdhF:8i", of

cont-rol . . 'y,<.,'tcrn,"· u;ith

optirnal

11.ucertai·fl pa.'ra1netc'('8

PROOF. Considcr the sUIn of the fin·.;t two in fOTlnn]a (ti): ~1Jl(L;~)l

=?::

~~

+ L~~

=

(~+

LoJ)

Re1na.,.k. Synthe~iZtinp; a contrul 11\1 hich provides a lnini lInlIn radi us invariant Het foT' a closedloop sy~t.cnl~ onc: obviously ha.."') to check fir:::-;t

itcln~

l

+ L'm,~'

the robust stabilizability of the given

clR~~

uf

Ohviollsly, Lm.~ == {O}) if \I F E J Li=l ,. _. ,m 1,11 (L rrh(~ latter is trll(\ iff (11: ::=::: 0, 1: == 1, ... ,'ftL TIH~ ~Hun uf IlOIlCIIlpty sets is larg-cr thau each of thcrn, i.e. ~ 9J1(L o , J") 1 c 9)1(L, 3")1. Thcreforc~ r (ml(L o, 3')1) < r (9)7 ( L, J")t).

Unfort.ullately, the robust stahilizabilit:,'{ problcIll ha.Ll not been solved yet. "£hat i~ \vhy the robust stabiliz;al)ility of the clas:::-l uf systeIU:::-l ,ulder consideration is as~ulned here a priori.

SiJnilar eOllsiderations for the consecutive sets 9)1(L, tf)Ii~, the ~urn of k + 1 itcrns in fOl'nntla (6), lead t.o the salne result, Uy indnction, OIlC obtaill~

fiyst{~InS

The

H~e

of

ThcOT(~rIl

4 will be

~ho\vu

s.YHten)~.

IIcre~

the consideration i~ extended to the class of \vith Ullcertain pararHcters asslllning t.he

cor-:.flicir:nl...,' of L r!'l,ay take thei '(" 1Jalue,." in a .qiven cornpart

~on1Je.rr

...e.

Definition 2. AHs11lning S(:hUI'-Cohll cOllditioll (5) is fulfilled for every rnatrix L E )2;, a rnilliYllal invariant set. of the class of systerIlS ('1) with 1111certain para.lnctcr~ is the union of the luillilnal invariant 8cts (:alcllla.tcd for every particula.r Inatrix L froTIJ the set £:

ill the Ilcxt

scctiOll.

J. SYNT'I-IESlS OF CON'TllOL SYSTElvIS UNDE,R B()tJNDED DIST'lJll13ltNCES ;3-. -I Oplirnal cont'l'ol 8l/nthp-s'7.s 1vith.

."·et

m1(£,~) ==

U 9J1(L~ ~)LE£.

A-;no10n paTa111ef,el'

In particlllar, if

'f'J,att·i~T.

By virtne of

.£ =

Thcol"(~lIl 4~

the synthesis problclll i~ r(~dllCcd Lo (:(tlculaLing a feedback Inatrix C opt providing tlH~ needed property of n· Inatrix L. Such lua.trix C opt. alvlays 0xistti, if systerIl (1) iH cUIlL]'olhthl(~ accordiIl~ to Kahnan's criterion. Cah:lllating: CL fee( Ihc:u:k IllR,T,rix is t h(~ ea.:..;iest for scalar coutrols. In this ca::,K~~ it is a.ppropriate: t.o cOllsi(lcl' luatl'iccs A and B of the standard f01"111

A=II:}__ :__ ~II, ~I A~,

11=

JI

v;,;rithout loss of Rell(~rality. Obviollsly, IIlatrix L ==-A + .BC is of t.he ~taJldanl [orrn (l:j).

<_COllV

1

,,--1,<,. 11\

{L1:}~

~rith 1\l verticcs~ calcula.t.ing a Iuillilual invariant set bCCOII1CS le~s difficult. The foHo\\.Ting re:--;nlt wa.s first prp.se:uted hy KUllt-

lueaniug £ is a polytape

sevieh (199U). TheOTe'ffl, 5. If the condition

every L E

)2;~

9Jl(£,~)

([)) is fulfilled for

then :=;

(,()llv{~1)l(Li,:J): i

==

1 r ••• , N}

is i) nn invaria.nt cornpact convex Bet for tlYStern (4) \~l.ith uncertain paralnctcr lllntrix L E ,.e. and ii) a Hon-reducible upper hound convex (-~sti­ rnate for the Inininlal invariant :-let of the cla.~~ of Sy~tClllS.

2833

Copyright 1999 IFAC

ISBN: 0 08 043248 4

ANALYSIS AND SYNTHESIS OF OPTIMAL AND ROBUSTLY OPT...

Tll(~

14th World Congress ofIFAC

rnost siguifieaIlt curollary fronl rhe above i::l

TheOTe'lfl 6. Under the conditions of Theol'ClIl 4 and for the given set-valued parR.rnct,cI' estin1ates (16) thn feedhack veetor

ohvious:

l

[;o'FollaT?J 1, UIHler the cOJ1dition~ of TheorcIll

5,

o

the raclius of a COllV
(Jopt = -L1 m

is

l'hc ra.diu~ r(ffi1(£.,~)) call be calculated directly hy (~va.l11atiIlg the I1Urln~ uf the vertice::-;. Howcvel"~ the l1~e of sHch a JJlinituiza.t.ion criterion i~ lirnitcd t.o low dilne!lsion~ Tn,? sillce one has to pay t.rcnl(~n­ dOllS CUlIlputatioual efforts for

hi~hcr

. (:OH1,7

'l=l, ... )V

for the pl'obleTn (15).

PROOF. Further in the prouf, the index

-rn. \vill

be ornittcd. Introduce the notat.ion o

'P(~A,

dirnensiollH.

8b, (;) ===

IIA + 6.i! +

0

(l) + 8b )(;JI

(17)

for the objective function of t!L<' rniniInax: prohlelLl (lG). Here oA E 82l and 6') E c5b. A~S\llne

Alt,(~rnativcly, eOllHidcr t.hc; class of Hy~tcTnH v..!it,h scaJar (~ontroJ aud uncert.ain paralIlctcrs ..l4 m E 21 m . alHl lJ.m . E E7 1n , where Q1m.

a, lniIlilniz;(~r

0

/h rn

o

0

C opt = -..l4/b and substitute C hy

CO'Pt

in (17) to

obtain

{A~n}

(18)

is et polytope and b·rn. =

{b'nl. :

f!.m. ::;

IJ.,n ~ h1n } ,

Eqllival(,Htly~

is an interval, j-1~n.) 'i == 1 ,.IV, ar(~ the verticc;-.; of a p:;ivcn polyt.opc 2( 'rn and f!..m ~ 6111. arc g-i ven cOllstants. i







Let. llH point tu the f(·~a.Lnr{;s of Ho nilpotcnt lIHl.trix except thot;e ruakillf]; ll~e of eigcn va.lues. Consider the -(n.-tit degree of a lnatl'ix L of the standard

(1::1)

fOJ"tIl

11 L'[;, Ill' Lm=IIG(L~),,'

{i(O) =0.

(14)

It is straighLforw
(IG)

lllin UHi.X {IIA m . -t- lJ n1,(711}· A., .. EQi""

C

0- 1 0

introducing; the notation tiL == bA - 8bb A~ \vhere ~L take~ its valuc~ on the set 8£ ~ cS21 \vh(~n 0-

oob

1

0

.A..

Set 8£. iH a convex COl.l:lpact set, since tbe set~ 821 and hb aI'~ convex and cOJupact. FurtherIIlore, these two sets contain the origin , hence o£; contains l.1Jc urigin tuo, This iIllpOl'tant property ,viII be llsed l
cp(8A, 8b, C opt ) Le.,

G"opt i~

a rninirnizer for the pl'oblcHI (15).

Substitlltc C hy

b,,, Eb."j

It is ",vort.h to note that the optiInizatioll problcll1 (15) appcar~ to he quite the salIle as cOlu:~idcred ill I(lllltsevich a,nd I(uutsevich (]994), where another perfornlallC(~ (Titerioll wa~ ll~ed.

Expluring- the \vay the HYllthc~iH ploblclll has be(~ll solved a.bovc analyticfLlJy for a syst(~ln v;,rith knowll o

paralucter Inatrix) calculate the center . .4,nI. of the ~et 2l m . as the C(~llter of r,hc; tniuirnal upper hOl1 nd solid sphere con tainiug the polytope a.nd the C(~Jltcr b'l'l1. :;:= ()'[)(~m, + 6m J of the iIlterval set b (here, \ve aSSHIIlC () t/:. b). Ko\v) repres(~nt the given pa.rarnetcr eHtinHtte~ iu thQ centralized forlll 2lu~

o

0

::= ...4n~

+-

82l rn n.ud bm, ===

llm

+ ~bm;

(1(j)

Q

\vh(~r(~ b211'Tt "rlH~

==

follu\-ving;

pro1 )}Clll.

conv{ . A"~1 - A·rH theOl'CHJ.

:

1.

=

1~ ... ,.Z''l}.

gives tlH~ solntioll for the

~ cp(bA, 8/), (;),

(;upt

+

(71 and ol)tain 0- 1 0


=

/IbA - 8bb

. .4

0

I

(b

+ 8b)C1 1f.

Next, take into account the llotation.~ ftcceptcd il.l (19) and rewrite the ahove in the for-In:

No~r,

recall t.hat 1) is strietly llC)lll.tive 01' st.rictly positive on b (0 tf. b) and 8£ cont.ains the origin. lIence for any Cl, onc can find a v(~ctor /j L* E /j£, such that

TheurCUl G gX~!lcralizes the corre~pondiIlJ,!; result pHbliHh{'d by KUllt~cvich
Obviously) the n.Jhll.stly opt.ilnal control synthesis problclll should have ueen cOll~idcred at a pararIlctcr lnatrix hclong-ing- to et Sehllr-Cohn tict-.

2834

Copyright 1999 IFAC

ISBN: 0 08 043248 4

ANALYSIS AND SYNTHESIS OF OPTIMAL AND ROBUSTLY OPT...

14th World Congress ofIFAC

H(),,"vev(:~r,

it scenlS ahnost iUlpOSKiblc to overcorne the rnajor difficulty- of applyinr; thif) (:oHstraint, because there is no c:oll~trllctive (in viev...r of COIIlplltaLioIls) dcseript.ioIl of a :-\ct of ~tablc Inatriccs. IInl1ce, (In(~ ha.s to check the robust stahility of the sYIlthcH.ized cOlltrol systenJ aJter calculating: C:ojJta If the rohnst stahility conditioIl~ arc nut fulfillod, the only ""ay is to run a sct-rnernbership (~stiIIlai,illg procedure anCVl \vith the aiUl to l'CdllCe an uncertainty in parCl,rnetel's and then recalculate

V.1\'1. 1-( uu tsevich and .A_. V. K nu tsev ich. Robust, Htahility aua,l:rsis and control synthesis nndel" uncertainty. Syste'rn.." Sr.ience, 22(3):17-28, 1990. \'.~'!. I(Ullt~cvich CLud B.N. PHhcniLehllyi. Invariant. a.lld ~tn.J.ional'Y ~(~ts of nonlillcar discrete Sy8t.CU1H under bounded disturbances. P,,.ob~ !t-;rflY '(,ll)'((lvlcniya i info'{''fnatiki (Jou:rnal of A '(1to,.,nnJ1~on

(7apt .

4.

C~ONCLCSION

and Tn!01'1nat.io'fl,

~S1rjenr.f'j'l)~

(-1-2) :35

4G, 199Gb. ill Itnssian. V.1vL Kllut.::-levich and B,N. PShCIlitchllyi. :!'v1iniIual invariant sets of dynarnic ~ystc:nls with bounded dit-;tllrba.nce~. CybeTnct1:cs and ~')?J.'jte'ln,,, Analy8'i..'i\ ;12(1 ):58 (j4~ 1990a. TraIl~lated fruIn KibeTnetl~ka. i 8isternnyi Anali..z No. l~ pp.74-81: 1996, V.lvI. !(llntscvich and B.~. Pshcnit.chnyi. :Ylinilual invariant. sets of discrete systelIlS ~Tith Htablc linear p:1.-rts, P1"OblernJ) npTa.vlenjya i i'njo'l"rnat1~ki (.Jo1i.171,al of A1J.tornat£on and Info',"rnat£on /:;cience...-j, (1) ;81 -91, 19D7. ill Rnssiau. 1\11. ~/Iilane:-;e, J. NortoIl I H. Pi(~t-llahaIlicr, and E. "\i\laltel', editors. _Ro11..nd'£ng a.pP'f'oar;h,Po,'; (,0 b'rlstpJn ident(fir.at1:on. Plcnurn Press, Nfn,v- YOI·k Rnd London, 1996. E. vValt(~r and L. PrOllzato. Idt:ntift:cation of Pararoet'f'ir: AIodelb' frorn. E,7:perirnental Data. 1

By acccptin!!: the hard-boHud COllcept, one llccesHarily ha.s to refol'ln1l1at(~ anc~'" all Tnajor anal.)'THis and control synthesis pro ble1I1H and al~o onc: faces IH~'V unique problctns having- llothing; ill cornrllon \vit.h tbose rnet ",,.hell using- Ktochastie Ill0dclH. The purpose of this pap Cl' is finding; an (;asjly calculated r!lCaStlre of a systclYl reaction to bounded disturbance...". The pn~s(-~nted Inctbod~ of HllalYNis and synthesis ruake llSP of tJl(~ pCl'fOrIllancc eritcrioll, which is Iuinirui;c;illg the l'adiuH of CL Ininiulal i!lval'iant set of a. dynanlic (closed-loop control) :-;ystCIll, r-fhcse Inethods arc applied to linear and SOlnc typ0H of nOlllincar systcrns. [fhe cla.'-'b of l10nlinear rnodels llIHlcr (:oIlsideratioIl is l"CBtricted to tho~e \vith strictly bounded llonEnear parts. It i~ v(~ry iIIlporta,nt to extend this class.

T.

Springer-Verlag, Berlin Hcidclbcrg~ 1997. )ro~hiz;awa. 8tab1:Uf.1J lheo'r"y b:rJ Liapu.'nov,1.-; second ',nethod, NUlllbcr 9 in Publ. of the l\:lath. SOf~. uf Japan. The lvIathclnatical Society of Japan, 196u.

A lllore (lifficult (-llHl COIIlIllonly faced pl'oblclIl of the robust stabili~a,bility relnains yet. HIlsolvccL How·~cvcr, the existillg control syuthosis rnethods proved to he applicable (and ::-;oluctilIlCS very cffi~ CiCllt.) to a "vide class of feed life control ::-;ysteIIl~, where the hard-hound concept is the Duly acceptahle.

RcfcrCll(:CS

E,.A.. Barbashiu. Introdurtion to the theoTJj of ,4}tability. v\tolten-.-Noordhoff, Grouingeu 1970. U. \l. Dnlgakov. On the accllrnulatioll of dist.urbances in lillear oscillatory systculS 1~"ith (:011Ht.a.nt. paraJ·n(~tcrs. Dokladu Acad. },'r.i. URSS, 51::143 '345, ID4ti. (in RtlSSlan). A.. V. K tlutsevich. Sct-lllClnber~hip identification for l"OI)lU..;t (:ontroJ. III (/ESA :96 IJl1AC/5 11;1ulh:confwrenr.e (L-il.lt".-Franre; 1 996), v()lllnH~ 2, pa.u;CH 11G8 1172, Lillc, 199G. A....'l. Knutsevich and V.l\1. KlLllt~cvich. Lill(~a.r adrtptive control for' llonstat.ionary sy~tenJs Ulldel' bouuded noise. /,?/8fpJn and Cont'f"ol Le.ttPT.,,:, l

7

:11 ::)~) 40> 1997. V.M. T
2835

Copyright 1999 IFAC

ISBN: 0 08 043248 4