New Concepts for Automatic Generation Control in Electric Power Systems Using Parametric Quadratic Programming

New Concepts for Automatic Generation Control in Electric Power Systems Using Parametric Quadratic Programming

Copyrlrh, fJ I.·AC Rnl Timt Di,;ul Con,roI Appl;"a'ioru c.. adalajarl. Mu i(o 198' NEW CONCEPTS FOR AUTOMATIC GENERATION CONTROL IN ELECTRIC POWER SY...

1MB Sizes 0 Downloads 25 Views

Copyrlrh, fJ I.·AC Rnl Timt Di,;ul Con,roI Appl;"a'ioru c.. adalajarl. Mu i(o 198'

NEW CONCEPTS FOR AUTOMATIC GENERATION CONTROL IN ELECTRIC POWER SYSTEMS USING PARAMETRIC QUADRATIC PROGRAMMING J.

L. Carpentier*, G . Cono** and P. L. N icderlander***

· Electricite de France , 2 rue Lo uis Murat , ]5008 , PorU , France "Elecfricife de France, 22 avenue de Wag ram , 7j 008 , Paris, Fra'lce " ' £ POS, J8 ru ede I'Yvette , 7j016 , Paris. Fran ce

Abstract. New concePts for Automa tic Generation Control in E'ectric Power Systems a r e presented , whe r e the two components of Automatic Generation Control , Load Frequency Contr ol and Economic Dispatch are perforned at the same rate . i . e . a few seconds , and where Economic Dispatch takes network security into account . This gives network secu r ity and iOOd transients, avoidin~ contradictory actions of Load Frequency Control and Economic Dispatch on the generating uni ts . The corner stone of the SOll:tion is the use of a new fast on- line Optimal Power Flow , using a new oarametric quadratic programming method . which is presented in details, ~eywords . Digital control : load dispatching : on- line operation : power system control : quadratic progranming : automatic generation control .

INTRODUCTION The Automatic Generation Control (AGC) o f an electric power system has two main objectives : first adjust power generation to load , second minimize the sum of the genera t ion costs. In the present state of the a rt! 1 1 [2] [3J (41. the first objective is met throuah the so-called "Load Frequency Contro l ' (LFC) , the second one using an "Economic Di spatch" func t lon (EO) , Usua lly, these two functions are cOr:lpletely separated in time , LFC taking the system dyn8r:lics closely into account , with A cycle rate between 2 and 10 seconds , whereas EO only changes the set points of LFC eVery fifth minute . E x c~pt a few exceptions. among which New Yo r k ~ower Pool may be mentionned (5' , no car e is t aken of power transmiSSion network security in the EconomiC Dispatch function . In these conditions , as mentionned in (1], the usua l Automatic Generation Control process has two main dra wbacks : LFC and EO control actions may of ten be contradictory , which results in undesi r able oscillations for the gene r ated powe rs. the power t ransmisSion networ k secu ri ty is not met . ElectriCite de France is now studying ne w concepts for AGC : avoiding these t wo dra wbacks was included in the objectives ,

UDCA-T

PRINCIPLES OF A NEW AaC

SYSTf'~.

Respective Functions of Load Frequency Control and Economic Dispatch The objective of the new Aac system project is to adjust generation to load in the most economical way, meetin~ generating unit constraints and transmission ~ecurity const r aints . If the latter constraints are well defined (transmission line and transformer currents within given limits under various conditions), it Is not the sa"'e ~or all the generatina unit constraints: o f course unit power as well as Dower ranps versus tine must remain within given limits: but also freql;er1: and large pO'lIer osci llati ons on a un!: must be avoided, which is not so clearly clefined ; if the latter constraint did not exist , the most economical AGC would only include EO :a few units , the "incremental units" , in economic balance with the syster:l , would adjust power generation to load in the ~ost economical way . But , as load variations versus time have an oscillatory corr_poner.t , large osc i llations might occu r on the powers of these i ncremen ta l un i ts . wh ich is unces i ra bE, To avoid such l arge oscillations on a few units , the load variations a r e partitionned into t wo components : the "Oscillato r " " component , for which LFC shares the oscill
'96

J . L. Carpen lie r, G. Co Cto and P . L. Nied e d a nde r

the average available powe r s of these un i ts a nd is costly . but this cost corr esponds t o spa r ins inc r emental unit "'e a r . Un it regul at-

i ng bandwidths are determined by schedul i ng one day in advance and i ncide ntall y ad justed eve ry hou r ,de f ini ng t he LFe possi b i li ties . Then , f or on - line cont r ol , the probl e m le ft is to apply Lf e to t he oscil la tory component of load var iations a nd EO to t he trend

e. ~ .

component .

Basic Pr inciples o f the New AGe System PrQject

The basic typical properties of the pr oposed system consist of performi n ~ ; Lfe and EO f unctions at th e s a ;ne r a t e , a few seconds , e . g . 10 seconds ( which , so f ar , seems sufficient for the F r ench system ) , ED with t r ansmission securi t y constraints (l.fC need not meet these constraints , due to the osc illatory character of phenorena and to the line thernal tire laj.'!sl .

In these conditions , the essenti a l functions related with AGC include: Every day , power scheduling with basic de termination of the unit regulating band.... idths . Every hour , final optlr.>al alloca t ion of re Ilulating band .... idths to the variOUS uni tS fo r the neKt hour lassociate d wi t h an opti l'1al po wer rlow) . Eve r y fifth minute , act i ve lonly) optimal po we r flow incl~dlnR elec t r i cal ne t work security analysis and spec i al ly p r ov i ding a " reduced r.>odel " of the product i on trans lIIission system , to be used eve r y 10th second . Eve r y 10th second , the on - l i ne AGC function itself . Including 1 . Decomposition of inout si~nals i n t o oscil latory and trend components (these input signals , which are the measurements of the variations of load to provide and of the currents to adjust in crit i cal lines , .... i l l be defined Mo r e accurately belo .... ) . 2 . Aoplication of secure Economic Dispatch to the trend co~ponents of input signals : this gives va r iations LIP to apply to the generating unit powe r s , 3 . Application of Load frequency Control to the osci llatory components of input signals this p.ives variations lIpf to ap p ly t o the generatina unit powers Cof course , LI P then LIP · • LIP • lIpf rreet the generat i ng unit constraints conct"rning the limits of POWB"'S and po .... e r ramps) , d . Sending of the orde r {)p " IJP · IJP f t o eac h Ilenerati ng unit , In p ut Signals

16

It was shown i n a p r ev i ous paper J tha t the i np ut signals ma y be tak en equ al t o

6Pe • I t . 6b , = -~ 0 ( lI F • --,--) d t

(1)

f o r th e me a su r emen t of t he var i a t i o n o f lo a d t o p r ovide

. -ot )

fo r

0

M!i.. dt ), 1.

(2(

t he me asurement of th e var i a t i on o f a

critical cu rrent to adjust , wi t h t he f ollowing notations t : time lIf " : frequency deviati.on

.

LI P; : e Kported power deviation M1 : cu rr ent deviation i.n 1 ine or t ra ns f o r mer L

L .I. . ), L : constants (possibl e tobe a dju s ted) giv i ng t he dynanlc prope r ties o f the cont r ol systen . Tt .... as denonstra,ed that using thes e inpu t signals allo ws to iet for ~he co~plete AGC the sal"e t r ansients as fo r the LfC i ntegra l control used n., .... in France , ''' hich 1S satisfacto r y , speciall', very stable . ~,

Load FrequenCl Control function

Th. Load freQ;.len cy con tr'J l lIPt:' J b.Pj

'J Secure

(31. '''i th

CBj/ I:B j'lIb; co~ponent

of

i. very Slnple

"r

relative

<0

the j th unit

regulating band .... idth of the j th unl t. Econo~ic

Dispatch function

The Secure Econonic Jispatch must use th e trend co~ponents of the input signa ls li b, ... lIbL ' " in order to provide the corrections LIP. This prcblen is not simple to solve within a very lit t le time and need ne .... techniques we shall call " fast On - Line Optina \ Po we r fl o w" ( " fOLOPF " ) and presen t be 10101 . Th e fe a s i b i1 ity of these new techniques is the co r ne r stone of the process and was the f i r s t s t ud y pe rformed after the process was Imag i ned . FOLOPf teChniques are nerived from t he tech niques of the usual Optimal Power Flo .... performed every fifth minute , and we sha ll fi r st briefly recall the latter .

5 r~INUTE RECURRENCE OPTIMAl. POWER fLOW Problem to be s'Jlved for this program , the data are the equip~ent on service for generation and trans~ission, the f!Jel costs , the l'Jads and the voltage magn I tudes Capprox imate if not we 11 known) a t every b>.Js . The results are the acti'/e powers P of the generatinll uni ts minimizing the total ruel cost and neeting gene r ating un i t constraints and trans~ission security constraints , for the Intact systeM and '_mde r contingency . Notice that the fuel costs ma y be take n a s they are o r biassed in order to ta ke a n a ccount of const r ai n ts (as po wer ra mp const r aints) like l y to appea r u p t o o ne or t .... o hours late r . Diffe r e nti a l Injections Met hod The problem is solved usi ng the "Dif f ere n t ial Injec t ions " method r 7] r 81 whi. c h bel ongs to t he f a mi.l y of " compact" Op t ima l Power fl o w Me t hods [9). These compa ct methods are c ha ra c t e r ized by the building of a "red uced mode l" of the problem , val uable i n a mo re o r l e ss Io'ide re gion , wh e re t he us eful co ns t r aints o f the p r obl e m ar e e Kp re s sed o nl y ve r sus the co n t r o l v ar i a bles P , i . e . t he

'97

New Concepts fo r Automa ti c Gene ration Control active powers of the generating uni ts "'hereas the compl ete proble~ is usually also expressed ve rsus the state variables e , voltage phase angles a t every bus of the network) . In the Diffe re ntial Injections I:lethod . the r educ ed model is non-linear, the losses being a quadratic function of p . and the fuel cost may be a quadratic function of P . In these conditions , the r educed model remains valuable on a wide region . Two main steps are performed ite ratively : reduced model build_ ing . r educed problem optimization . till the reduced problem is equivalent to the complete problem . The process nay be sunmarized as follo ws: 1. Start : find an initial solution pO(us",al_ Iy optimal but not feasible) . 2 . 'iith a physical load fIo ...· (Newton r.-ethod) comp u t e the electrical syste~ state e " f(P) . 3 . Analyse and se l ect the useful cons traints: this step includes contingency analysis if security under contingency is des ired ; usually t he useful const raints are not nume r ous , less than 12 fo r a pr oblem wi th 100000 potential constra ints. 4 . Build the coefficients o f the reduced problem t hrough sensitivity techniques (dOL/dP for the currents . dp/dP , d'p/dP' , p being the system 108ses) . At this point the reduced model is built . 5 . Test if the Just built reduced orobler!1 is iden t ical t o the previous one . If yes , the optimum is reached: i f no , go to 6 . 6 . Opti mi ze the r educed problem : Using the reduced model . one finds : 1 ) a feasibl e solution f o r P , 2) a feaSib l e and optimal solu ti on fo r P. Then go to 2 . Step I is initializati on , steps 2.3 .4 reduced model building , step S test, step 6 reduced problem opti~ization .

to 40 seconds with contingency in the security analysis . SECURE ECONOMIC DISPTACH TO BE SOLVED EVERY 10TH SECOND IN AUTOMAT IC GENERATIO N CONTROL Basic Ideas The basic idea to Ret a fast solut ion of the secure econom i c dispatch to be solved every 10th second in the AGe algorithm consists o f noticing that : 1) the reduced model built every fifth minute by the Different ial Injections method remains valuable to describe the system operation during the S following minutes (exceot the case when a sudden event occurs: then the corresponding addition to the reduced ~odel may be computed ve r y q·l!ckly . ·... hich keeps this proposition valuable) . 2) at the beginning of a 10 second cycle , we have the opt i mal solution P corresponding to the previous load: it is the solution of the previous 10 second cycle o r of the Differential !njectiorsal"ori t hm at the beginning of a S minute cycle : in these conditions . it is sufficient to trac k the optirr.a l solution ..,hen load changes, L e . to perform a parametric optimization . instead of solving an optimiz atio n problem with an initia l flat start . Pr oblem State me n t followlng the a bove basic ideas , t he problem to be solved ma.y be writ ten :

lA, J

IIPj

~A!.j

liP,

J

m

lIPj

~

"

" "tPI!'

1:.... fbL (L ~2 )

IIPj ~

J

Proble~

(S)

VJ

(a)

(6)

liP')

Reduced

(d )

X.I.'lP i Dj J liP J • Ob,

mlnimale

(7)

Optimization th lIPj va riat ion of the powe r Pj of t he jth unit (or Dart of un i t beinit defined by a quadratic cost) . IIb l • IIb L : trend component of t he input signals de f ined by (1) and (2) : they represent the variations of the conditions of the proble~ due to load changes since the las t ten second cycle . b L : current margin for current in line L at the end of the l ast 10 second cycle ; for those lines exactly on their limi ts, b L E 0 ap ap Alj • 1 - ~ • with ~ " lrst differen'oil

The rpduced problem optimiZat i on is performed using the Gene ra lized Reduced Gradien t (i.RG) technique [10!. In this algorithm , a t each step the variables are moved , when possible , in the opposi te d irection o f the gradient projected on the linearized constraints ; the length in this direction is given by the mi nimur!1 of the objective function o r by the fact a new constra int is rlet before ; then , at each step. non- linearities o f the const ra ints are take n into account thanks to Newton's me thod . A very il"'po r tant feature ~u !t; be noticed in this method : the gradient,i . e . the first derivative o f the objective func _ tion . o nly gives the steepest descent direction , but not the opt i mum point d i rection nor its posit ion . GRG is used here t o solve a quadratic program but it is a gene ral conve x programming method without special pr ope r ties fo r quadratic programs . Usually , 3 iterations ( reduced model building . reduced problem opt imization ) are sufficient. The nu~erical perfo rmances are as follows In IBM 30- 81 : f o r a system with SOO busses . 800 lines o r t ransformers, 65 the rmal olants , it takes about 20 seconds without cont ingency and up

tial losses (cons tant )

o

a 'p

.

.

Dij" apiOaPjo " 2nd dlfferental losses (constant ) ALj : sensitivity of current in line L versus IIPJ . inci denta l l y after trip o f a line or unit d . IIPj , IIP~J : bounds for IIP j . due to physica l bounds and ramp limits du ring 10 seconds. c J ' e J : coeffiCients of the ob jec tive func-

J . L. Car pen ti er , C . COt t o and P . L. Niede riander

598 tion I'"j

~O ,

ej

~ O) .

r ;. 'j t.f' j

( ll l represents rhe halarlCe h ... · ~,(>.~~ ;::"'r,"r,,'\ I.

IS

load and losses. represents ~ he - ra:'s-issi .

(6

constrai:l >s , r epresents the

S",

r l'.

r ''',

!n I inear pr -gra-- ir.l! . !IS ... • ,"', -e'_riz
•. -' .>:' ; '1:':'1 _.,.,

'";-'1;



•. •

act'ml val'le : In q;8.1ra"\·· I."_;'::·"-,~.:,': !. s I"'ethod does n':lt "Xlst : s , s L·· ,. problel'l , ... e had t' I",a~ine a ;;"-.' i rl:'gra:-C-i!h!. ",> · n,:.-: . side . '.
SEt'UHf: f." ;11. '; ! ' T'lI',-,U~:".

PR('~HA:·w::r

',:,\1. i

" . ,,;

;'';!'A·:,.:7~··

:.'1 r

I, i"

:';.;~;,.~

: ",.;,. T ll-' ','

oiEF Fi.. .";":

?

,'I"

r"';

,'.

....

First , two lt1"a,; al !,,', c'jl t:1 (of 'b' pr .hl ... :'· 1 ,. ,I '/'. "l'an inter,..r·diStl" : Hi,lr'l' i 'r' ;Jr-,· ~ l' '. 11 ,''',I' ronstrllinls : ',s" . ,'~-l'l:~'-"l ,!'<', ";' ' S ; in th ..se con{\i~i ,ns , ~.,. it,>_',",' 1; 1'- 'I ('~r'l r1etriC" 'l'Ja'jr't" j, pr-,>!r,,- ',:,'! \:1.' -,;' .5train~s and s,..all :i~"';SI .-"s . ,;j" " '" " 11tions neressar::' r,!;i 0>01'1::; ".' '\ r (8) , For the para-<:,tr.,· 1'''~ ;j'R.' ,., : , • .t:' ,itself , Wl' firs' r~,r;1 ,'. q,,, "'11'\,,'\ ''''hich wO'J11 be op~l~al j,!1 .. ,.- .'Itl, "0;'; " . ; ne'''' c'1nslrai:lt is -e' ; 'ht,:~ ,,' c ';." di r er ' i'm tj"!lnec b-; :,V '"I'S ',,:. 'I'> ,~ ,_ traint ra«es 1':: p'~ssir-·lo(' : .,~< .. ".. -"1"_' ',1ther s ~ ef' taklnl! in', 9 '<-,,tra!nt -",t . (,'c ,., T 'I"': 1:-' 6,." .... . . .. . ' ", ~~~.q',es 'lseri in :.inl"ar ;:'r' >o1ra--l." .,\:, .~, ,5":; GR"; cann'~t '0 ... jse~ f-i · :. .. r : 1'; :""-: , '.
!I,

Use of an

In,;er~eciate

"

",

!:-

;:; ' -: 'f.S'r!oll~ .

'l.f"!lera' ,'\I!

This pr.,ble-. is (I ,':flira " j< r~' '::-"- , ";" . " quadratic ')bje(' t i'.'" : "'" '.l n' ;;r.'i·" ;';8dratic (''l!1st:'ain' la '; ,••' ....'.':' , ~', . ,{" 'h' 'I \' if llb ' , tJ.r ~ is ' ; - '1. b' c.,,,,rs,," ft'8slb'>: 5' i . ~ .

5')1 • t \">n

_,

, 9'

t.b:

O~adra ' lc

?roO!ra-

~.'-_

linear Constraints It ::lay be shown :10' ~hat. :'.r '''_'' :~:j-alit-: oroperties, (6) is equivalent ~ o (11

r - r

t. P _

'. "'·__ :r,i!. -1'".1' ~ ;al va riable assoria 't"d ""ith .., • " , S ,';1" ' !,;", ~roble- lA , i.t is s ·· , 1"1", ' s",· .. {:: '.hen',:,correctc.P :' :~.:,' - l ( " " : ' as ~erf,:,r."er. i.!; the r,R~ "I'~':'i·r.'(, ~

"5'-

,

~f'laxa·I.,"

:!' r

'~s

':. Cons 'rainls

j.

.:'1 a:j-;an'ageJ'.. s t, ':-onsi'ler at 'r.VS" relati mships IfI'hirh are ,c':",, ' i'!c , W'J.,pin~ !Qr "r.,laxlng" the others ',"'" . !:,. '-U,,'Uli,m . O!" ro"rse this involves • h' 'asks ; ',;;]' -1-,\ :,I! • '.' . ' j j •• 'S ! t~t ~ar.i11I\S ,r the relax.·tj 'r.!;I'ralr. ' s ; If a -argin oec~~es ~t,r· . 'r.'· 'ns'rai:'.'s bec(>"',!!' f'ffecti·le . ~a'·r.::·,1I 'r'" ('!"'ec'\ve cons.raints '"hich s'. ,_~ '0'" I'f"iaxe; : 'his hap!)ens "'he:'. t.bl. l~ 2" ! ' tl::''j ':a> co)rresP'JtlC!ing d:.Jal variable :L h,,<" '· .... 5 P'Js i t i vc , or when the t.!' are not n~; f' ."'," a va!-iation 6bL which wou l d ~~"1


-", • : S' , I ' " , 5 "',. d,

",·;i tj "l1s , (111

!

'I

~.a~{

be restrictf'd

( ') 1

:. I

.I ( 12)

!. } 2

(; r ;,. t:.t

'~"'.:

,,

.{ t:.V'

(13)

is 'h" prc,b;e,.. rerair,lnFt t .... be solved 'I s ·pp ','here ~he set o:;~ constraints 'l~,"'S ~.?t change.

;;.sj·:" .::::

Fri~.C'if'e

.! 301,~tion

~!'a' ~c ~r~sra-

,f a Parare-ric O'~a ­

t:'~nstraineo

bj a Set of

:'ir.e;)r El,u'_ions T!':e gl::'... ral

fo r - o!" the croira- (:3) -ay be

tlri-:e~

).

~

;- "cx

x ,=,(;P ,Q b

el7)

x :; 8 ~XT JII

. C.P1'1 , 8

r (;'o l l '" AI DiJ It/j '" i , DU

'99

Nc,",' Co nc e p ts f o r Aut oma ti c Generati o n Co ntro l 'IJ e know for b .. 0 , x " 0 is the oPtlnal sible) sobtlon .

l'ca-

Reduced gradient. The variable...se!, ; x : is oartitionned into 2 $ubse~s I X, x ; ; :'~ll,:, ­ "" inJl, the sane oaftlt.!.on , A • fA, A'. The partition is s'.Ich that ;, 1S sq'~ar" ;;C'n sin"!';lar ; ~ is the basic co"'ponent of x. ;- ' he n':l!~-ba 5 i c co~ponen t . (161 gives rif . (~ , xT ~l d ! ~(c

...

(:4 Of

• X

- -, t -,;

gives A d ;

. • d;

•r.; •

,~



0 " Jx .. ('

7:.

(.9,

...cA' "-

1 ,,...

'T or "

~

-""~·ln.l! f"s ~

x . x' • 6x at the

0

and R ' x' , ( -,

122

,': 'r.

,

OJ

; OJ and a' >

-

.j

and

, 'j' j , 'j

'he reG·.cel .u·!es

15

and

a

.

gT

r.

· "P:T • r.

tile.

0

being a oarti tion 01

(2<1 ) 'j

such that.

(;oSI PrincipiI.' of the paranetric optil'liza·i'm. Th(' principle consists of tryilll~ ~o sol'le 'I': and (24) at a time . If x '" x' , .... lth g ~ ~' and llb ,. b - Ax' , .'~ '.·an ~

r.

Cl x '

.ob

.

-;

(25) ,

1If'(27) .

"r'''~si'-

III

I ilt '1- ~

Ril\~ , 7

8.

U,.dot(' P (p' . !\P) , '11 .... l-~la)(€'d constraint nar "Ith,' 11rS' ro:;,,,,, A, =>A, - llPTU·) . y

~!"

•...

n

·p' .• el,- '

,

I

YA- '

'ht""k 11 a '"rl\nll"'15sion line constraint -'Ill ' b,. dr :,p~d (6bl ., 0 and u L <: 0 or i··j ......sslbili'"l t, piv:H and 6b l tending to 10 "s'.·n 'he ,...,..,n!>.raintt . Ir so , redefi.ne 'A, -:-"""O."€, aRa1n "A- ' , '1 " y r;:- :, If p <: 1 and rlvr'inA: n~' n"r:'orMed . go to 9 ; lfp '" 1 ~ r p <:. al.'1 I'i"'':l~in-'l '"as JUSt perfo r red , f< ' ') 2 . !I • ir." . t, .. e~155ar:; because a cons t raint : :'('·:e,.teo P fr')~ being eq'Jal to 1 . This is j"."rl,r:-:"';"'It.h ~he sa~e selection cri t eria ,~ ir. :"'~a: 3i"'~lex "e~hod . Then go to 7 .

"r

'l-L 'llE

O?T:'~/,:'

P',)'IJER fLO'6'

P!'5'JLF

}

;':xlJ,:,:,j-en·, o:;,rganisation

iT

'lie chanp;e only ~ and x (6 x -

Exp':':'i-er.~s

which Iti'!es ~ he funda~en t al relations h ios to -ee ~ : A 6~ ' • Cl b

G 6; '

-:.,

,~.

,'n

fAST

0,

.: ~

J't '

1).,

A(x· · l l x · . b

A ll x '

';'i

i,,'h q,a~ ~ Cl ~ 1 and , : ! ' I ,'-1i:,5 ·('as:o:" . Cl ",a:: be 11m " , t· :"\ ."..- 67 , '. -,.'l."!'; ..... ne ~f i t5 bounds r· -, .!!( it ,;.. iax"d cr·ns~ralnt becol:les . ;-.. : the rlp'illi'i'le ·/al;.;(' 0 1 Cl nust •. ,~.,. ;,.' ••• I" .1~· a s,...,all correction to -."., 0 #. " RO t, 6 ; if Cl = 0 'I' " ;,,·i!!!!)l .. C. l 1;l'\ited Cl , g':l to 2 aft a',,1- 1: .•1 1 ,t'il! ','1"11 lIPj to:;, belong to t he • _"X ,. s('. c.F i.l Cl • t) for another reason , ' >,

At the point x ' , .Iet 1S oa r titjo'l x " Ix. : : ; such that ~j <: 'X J <: SJ or X'J ~ o.j and gj <: 0 ':lr X'J _ SJ anrl aJ > 0 , The op t i""H 11 t.'1 ",ay be r eached • r:ti ng to ge t & _ 0, or

'.. !lP_ ~

.:. .

.'

[0

:'p'

,l,

0'

j

y - . A.

~

ine llP s'lCh that llf-'j and gj , 0 > .;, :' .\f. '0.;" ;: ne . : Illb be il~g ,":'1 :'- ',: '1;1.' sec'md "'e-berl , ',". -j ", . ' i' .' iN ,. ...',·hp-1 ; ,.'11 E"~ g':"' to ~ ,

"

<0

rr-:'lble- constraints .

,.,

-:

Opt! Mill I ty cond i t ions . Thl"~; cor;:; ~I'''nrl '~ df ). 0 ror alll possibl .. ehanjZe ~ X, ... hich -'1b(' '.. rl tten :

'j

~ Cl -{ 1

Opti",al Po .... er Flow basic

~n - I..ir.e

,' ,

1?3

--,,

'~\e

'Ill

0

f~!.-_~PF aljZ')ri'h", fled'/ts directly fr~"l ' h'" ;:'''':i .s a:.al::s1s . I ' ~,a:1 be brie!!:. "'s,'rlb"'c as : ;.)11-:.0""5 ; '::'('1,:" :~')'" ll~' .. O. Use the part i tion t:.'F , r ,- the last ;":10",n solution ( which :""'!i; : ~9 ':"1 c.P " Q opti~al \ ......Co~P"J t e '"'' ,,,, ~ ~ cf, 'riP , rhe basis A , it s 11::":'';'' ~-' , "he ·'d·.;al vecto r " u z 1'A-' ,

12'

o~i.n.

""axli"-'''' such that

Cl

T~; ..

ro'" vector , is '11(' "rerLc,..1 p.ra:ii"~:'" f versus '; ; (? 1)\ sh,',",'s 1. is a 1 i :-.ear .- .:;!""'.i">n ,1 x . t!

6x' ,

a:!" :'i .!~~

i,

gradier.~

Cl

!l"F :

·,,1 th

A

C.x _

-Ilt

I)', .

(27)

,

This gives the variation Cl x . '" [Cl x • • Clx " 0 .... hich .... ould be opt1mal and ~eet the complete va riation llb of the second membe r if no ne'" const ra i n t appe ar ed . As a ma t t er of f act , the variation Axo is not a l .... a ys pos sibl e to perform , and the final variation for one itera tion will be defin e d by

... e r e carried o·~t :-o:;,r 29 case s or :-or ~ 1pica 1 1980 . 81 hours , ·... i t ~ ab':llt '}"" '::i;.lsses , 8QO lines , 65 t he r mal '~ni's , ... 1 th data cor-,in", fro~ s~'ste r:l me a S'J r e r>ents . Starting from an actua l case ( ab out 2-:>000 to 25000 therr:lal ;.j\v) l oads were inc reased or decreased by abou t 1200 MW fo r 15 c a ses and abo'~t 600 Woi' for 14 cases . On one ha nd , startina f~e OPtimal solution P of the actual case , the FOLOPf algorithm '..as "Ioolied ("fOLO?f solution") ; on the othe r ha nd , the optiC!lal so l ution .... as computed d irec tl y by the usual Optimal Po .... er Flo.... for the fi na l loa d conditions ("direct solution" ) . Th e o b j e c ti ve .... as to test ho .... siC!li l ar .... ere t he t ....o s o l u ti o m .~,'"

frf'r,o:~

s1s . e~ ,

600

J. 1... Carpentier , C. Cono and P. L. Niederlander

and how much CO~Dute~ time was required to get the FOLOPf solution .

Solution Accuracy Exper iments were carried out in two cases: with quadratic fuel cos t s and with linear fuel costs ....·ith quadratic fuel costs . the differences bet·... een the FOLOPF and direct solutions P were less than 3 MW . This proves 2 things: 1) fOLOP. algor! thn runs cor :,ectl~'

and accurately ; 2) the use of the reduced mode l al one remains valuable even for fairly large load changes . With linear fuel cos ts, a difficulty appeared : the fOLO?F solution

gave the same optimal cost as the direct solution , but the differences between the gene rated powers P could !'each :.Ip ~o 300 '.''11 . This is due to the existence of an infinit'l of equivalent optimal solutions with linear f'lel cos ts ; as the proposed AGC ~ay have to use solu t ions coming from both types of co::"putations , it wa s recolrl'lendeci to use only quadra tic fuel cosu .

Computation Speed for the 1200 MW load changes , the nur:ber of FOLOPF iterations where between 2 and :2 . the IBM 30- 81 computation time between 8 and 90 !fIi 11 iseconds ; for the 600 MW load chanp,es, the number of FOLOPF iterations were between 2 and 9 , the computati on times between 7 and 51 milliseconds . The nUl'lber of Iterations and computati on times ma inly depended on the nUMber of effective security constraints . These computations are very fast: always less than 90 (fIS f or 1200 MW load chan~e ; due to possib le additional computa ti ons ( e . g . in case of sudden system change) , t.he ma x imum e xpected computation time for FOLOPF may be estimated between 0 . 2 and 0 . 4 second : comDared with the 20 to 40 seconds necessary for a complete Optimal Power Flow , with FOLOPF the computation time is divided by a ratio lOO •

No w, there e x ist minicomputers us able in an Enerps Control Center with half the sDeed of the IBM 30- 81 : with such a minicomputer, the ma xi mum FOLOP F computation tine for a 1200 MW load change would be 0 . 4 x 2 • 0 . 8 second kno wi ng that in France 1200 Wi represents a ma x imum load change seldom reached within a 10 second internal , FOLOPF would need at most 8 % of the cycle time .

CONCLUSION New concepts f o r AGC were presented, 'IIhere on one hand transmission security Is net . and on the othe r hand LFC and EO are performed at the same rate, e . p' . 10 seconds . The feasibility o f these conce pts is due to the us e of a new tool , "Fast On- Line Optimal Power Flow" ("FOLOP F" ) , which divides by 100 the computer t i me of secure EO problems . This FOLOP F uses a new parametric quadratic Dro4 r amming algorithm ; it may be noticed this algo r ithm Is general and not limited to electric power s yste m appli cations ; wi th particular addi-

tions !'Iirroilar to those oresented in the pape r, it could even certainly in SOr1e cases be used AS an approxination for no re ~ene ral convex progra:nmin 4 · REFERENCES 1 . raavitsch H. ann Stoffel J ., "Automatic r.eneration Cor.trol , a seJrvey", International ,lo'.1rna1 of Electrical Power and Energy Systerr,s , Vol . 2 , No 1 . January 1980, po 21 - 28. 2 . Of' ~:ello LP" ~alls R. . T. ami B'Rells W.F., "Autoratic r,ene rati ~n Control . Part I . Process '~ odellin~ ", 'f.E!': rrans. PAS 92 . 1973 , pp 710_715 . 3 . De ··:el10 F.P .• ";ills R. J . and B'Re11s w.f., "A'.. to-atic r.eneration Control . Part :~ . uigital Control Techniques", IEEE Trans . PAS 92, 1973 , op 716-724 , 4 . Schells~ede n . and Wagner H., "Design aspects of a so!t war e package for Automatic ,';eneration Cont r o l 'II! th instantaneous economic dispatch and load forecasting functions" . IFAC Synoosium on A.utomatic Control in Power Generation , Distribution and Protection , Pretoria 1980 , OP 61 - 69 . 5 . Elacqua A. J . and Corey S . L . • "Security constrained disp at ch at the New York Power Pool" , IEEE paper No 82 WM Of\