IEEE Transactions on Power Systems, Vol. 9, No. 1, February 1994
UNIT COMMITMENT LITERATURE SYNOPSIS Gerald B. Sheble, Senior Member George N. Fahd, Member Dept of Electrical Engineering Decision Focus Incorporated Iowa State University Lo s Altos, CA 94022 Ames, Iowa 50011 ABSTRACT Sever al opt i mi zat i on t echni ques have been appl i ed t o t he sol ut i on of t he t her mal uni t commi t ment pr obl em. They r ange f r om heur i st i cs such as compl et e enumer at i on t o t he mor e sophi st i cat ed ones such as Augment ed LaGr angi an. The heur i st i cs have even r eappear ed as exper t syst ems. The pr obl emt o sol ve i s t he opt i mal schedul i ng of gener at i ng uni t s over a shor t - t er mhor i zon, t ypi cal l y 168 hour s. Thi s paper i s an over vi ew of t he l i t er at ur e i n t he uni t commi t ment f i el d over t he past t went y f i ve year s. Kevwor ds : Uni t commi t ment , enumer at i on, br anch- and- bound, dynami c pr ogr ammi ng, l i near pr ogr ammi ng, i nt eger pr ogr ammi ng, separ abl e pr ogr ammi ng, Lagr angi an r el axat i on, augment ed Lagr angi an, si mul at ed anneal i ng, exper t syst ems, and ar t i f i ci al neur al net wor ks. I NTRODUCTI ON Uni t Commi t ment ( UC) i s t he pr obl em of det er mi ni ng t he schedul e of gener at i ng uni t s wi t hi n a power syst emsubj ect t o devi ce and oper at i ng const r ai nt s. The deci si on pr ocess sel ect s uni t s t o be on or of f , t he t ype of f uel , t he power gener at i on f or each uni t , t he f uel mi xt ur e when appl i cabl e, and t he r eser ve mar gi ns. The gener i c UC pr obl emcan be f or mul at ed as: Mi ni mi ze Oper at i onal Cost Subj ect to: Mi ni mumup- t i me and down- t i me const r ai nt s Cr ewconst r ai nt s amp r at e l i mi t s Uni t capabi l i t y l i mi t s Der at i on of uni t s Uni t st at us Gener at i on const r ai nt s Reser ve const r ai nt s The obj ect i ve f unct i on i s oper at i onal cost s gr ouped i nt o t he f ol l owi ng cat egor i es: 1. Fuel cost s: They ar e r epr esent ed by an I nput / Out put (1/0) cur ve t hat i s model ed wi t h a pol ynomi al cur ve ( nor mal l y a quadr at i c or r educed cubi c) , a pi ecewi se const ant cur ve, or a pi ecewi se l i near cur ve. The f uel cost equat i on f or uni t I i s descr i bed by: 93 WM 155-2 PWRS A paper recommended and approved FCl ( Pl ) = FI * HI ( P l ) wher e Pl i s t he power l evel (MW) FCl ( Pl ) i s t he f uel cost ( $/ h) Fl I s t he f uel cost ( $/ MBTu) Hl ( Pl ) i s t he heat r at e cur ve ( MBTu/ h) 2 . The mai nt enance cost i s descr i bed by: MCl ( PI ) = BMI + I MI PI wher e MCl ( P1 ) i s t he mai nt enance cost BMl i s t he base mai nt enance cost I Ml i s t he i ncr ement al mai nt enance cost sTI = T S~F ~ + (1 - e( - Di / ASi ) ) ~~l ~l + MS ~ 3 . The st ar t up cost i s descr i bed by: wher e TSl i s t he t ur bi ne st ar t up ener gy ( MBTu) Dl i s t he number of ASl i s t he boi l er cool - down coef f i ci ent BSI I s t he boi l er st ar t up ener gy ( MBTu) MSl i s t he st ar t up mai nt enance cost The most nonl i near const r ai nt s ar e t he mi ni mum up- t i me and down- t i me r est r i ct i ons. I f a uni t must be on f or a cer t ai n number of hour s bef or e i t can shut of f , t hen a mi ni mumup- t i me i s set . By cont r ast , mi ni mumdown- t i me i s t he number of hour s a uni t must be of f - l i ne bef or e i t can be br ought on- l i ne agai n. Vi ol at i on of a down t i me const r ai nt may be al l evi at ed by banki ng or i dl i ng t he uni t . The pr eci se def i ni t i on of such i nt er medi at e t er ms has not been gener al l y r epor t ed. Cr ew const r ai nt s per t ai n t o t he number of uni t s t hat can st ar t at t he same t i me i n a par t i cul ar pl ant . The r amp r at e l i mi t s r est r i ct t he movement of a gener at i ng uni t bet ween adj acent hour s. The upper and l ower l i mi t s on t he gener at i ng uni t s f or ce t hemt o oper at e wi t hi n t hei r boundar i es. The der at i on of uni t s can change t he upper uni t l i mi t . The st at us of each gener at i ng uni t i s one of t he f ol l owi ng: 1. Uni t can cycl e on or of f ( Avai l abl e) . 2 . Uni t can be out aged due t o mai nt enance or 3 . uni t can be f or ced t o r un ( Must - Run) . 4. Uni t can have a f i xed out put ( Fi xed) . 5. Uni t t hat may be br ought on- l i ne qui ckl y wi t h no r egar d t o mi ni mumup- or down- t i me const r ai nt s ( Peaker s/ Gas Tur bi nes) . hour s down ot her unf or eseen pr obl ems ( Out aged) . by the IEEE Power System Engineering Committee of the IEEE Power Engineering Society for presentation at the IEEE/PES 1993 Winter Meeting, Columbus, OH, January 31 - February 5, 1993. Manuscript submitted January 28, 1992; made available for printing r eser ves. I n addi t i on t o t he above const r ai nt s, t he gener at or s must sat i sf y t he syst em l oad and l osses and pr ovi de t he r eady and spi nni ng November 4 , 1992. I Publ i cat i ons I n t he uni t commi t ment f i el d have abounded over t he l ast t went y f i ve year s. Thi s paper i nt ends t o summar i ze t he di f f er ent met hods used i n t he sol ut i on of t he UC pr obl em 0885-8950/94/$04,00 0 1993 IEEE _.,...., ~. . . . . . . . . . . . . , .. , . , , ...~-_,-- ,.. .I. .1_. . . . .I_ -. ,._ . - .. . 129 dur i ng t hi s per i od. For each met hod, t he avai l abl e paper s wi l l be r evi ewed. The r ef er ence j our nal s f r om whi ch t he publ i cat i ons ar e ext r act ed are: I EEE t r ansact i ons on Power Appar at us and Syst ems I EEE t r ansact i ons on Power Syst ems I EEE t r ansact i ons on Aut omat i c Cont r ol I EEE Pr oceedi ngs AI EE Tr ansact i ons I EE Pr oceedi ngs, Par t C El ect r i c Power Syst ems Resear ch J our nal Oper at i ons Resear ch I nt er f aces Amer i can Power Conf er ence Pr oceedi ngs I FAC Symposi umon l ar ge scal e syst ems PI CA Pr oceedi ngs GENERAL The backgr ound i nf or mat i on per t ai ni ng t o t he opt i mi zat i on t echni ques ar e di scussed i n sever al r ef er ences. Some of t he r ef er ences Cohen, A. I . and Sher kat , V. R. , "Opt i mi zat i on- Based Met hods f or Oper at i on Schedul i ng, " Pr oceedi ngs of t he I EEE, Vol . 75, pp. 1574- 1591, December 1987. Wood, A. J . and Wol l enber g, B. F. , Power Gener at i on. Oper at i on and Cont r ol , J ohn Wi l ey and Sons, New Yor k, N. Y. , 1984. Hi l l i er , F. S. , and Li eber man, G. J . , I nt r oduct i on ODer at i ons Resear ch, Hol den- Day, I nc. , Oakl and, CA, 1990. Bazar aa, M. S. , J ar vi s, J . J . , and Sher al i , H. D. , Li near Pr oar ammi n and Net wor k Fl ows, J ohn Wi l ey- and Sgons, New Yor k, Nemhauser , G. L. , and Wol sey, L. A. , I nt eaer Combi nat or i al ODt i mi zat i on, J ohn Wi l ey and Sons, New Yor k, N. Y. , 1988. Nemhauser , G. L. , I nt r oduct i on &Q Dvnami c Pr oar ammi ng, J ohn Wi l ey and Sons, New Yor k, N. Y. , 1966. Lasdon, L. S. , mzat f or Lar ue Scal e Svst ems, M%%k?f%v Yor k, N. Y. , 1970. Luenber ger , D. G. , I nt r oduct i on t o Li near and Nonl i near Pr o r ammi ng, Addi son- Wesl ey, Readi ng, MA, 1973. Cooper , L. and St ei nber g, D. , Met hods and ADDl i cat i ons of Li near Pr oar ammi ng, W. B. Saunder s, Phi l adel phi a, PA, 1974. Gi l l , P. E. and Mur r ay, W. , Pr act i cal ODt i mi zat i on, Academi c Pr ess, New Yor k, N. Y. , 1981. J ensen, P. A. and Bar nes, J . W. , Net wor k Fl ow Pr oar ammi nq, J ohn Wi l ey and Sons, New Yor k, N. Y. , 1980. N. Y. , 1990. Ref er ence 1 i s a gener al summar y of t he t echni ques used f or t hi s pr obl em. Ref er ence 2 i s t he now cl assi c t ext . Ref er ences 3 t hr ough 11 ar e var i ous Oper at i ons Resear ch t ext s, i n or der , whi ch t he aut hor s have f ound usef ul . EXHAUSTI VE ENUMERATI ON The UC pr obl em may be sol ved by enumer at i ng al l possi bl e combi nat i ons of t he gener at i ng uni t s. Once t hi s pr ocess i s compl et e, t he combi nat i on t hat yi el ds t he l east cost of oper at i on i s chosen as t he opt i mal sol ut i on. Thi s met hod f i nds t he opt i mal sol ut i on once al l t he syst emconst r ai nt s and condi t i ons ar e consi der ed. The f i r st t wo paper s ar e or i gi nal at t empt s t o r educe t he pr obl emt o mat hemat i cal t erms. Ker r , R. H. , Schei dt , J . L. . Font ana, A. J . , J r . , and Wi l ey , J . K. , "Uni t Commi t ment , I EEE Tr ansact i ons on PAS- 85, No. 5, pp. 417- 421, May 1966. Har a, K. , Ki mur a, M. , and Honda, N. , " A Met hod f or Pl anni ng Economi c Uni t Commi t ment and Mai nt enance of Ther mal Power Syst ems, 'I I EEE Tr ansact i ons on PAS- 85, No. 5, pp. 427- 436, May 1966. Happ, H. H. , J ohnson, R. C. , and Wr i ght , W. J . , "Lar ge Scal e Hydr o- Ther mal Uni t Commt i ment Met hod and Resul t s, I EEE Tr ansact i ons on PAS- 90, No. 3, pp. 1373- 1384, May/ J une 1971. PRI ORI TY LI ST Thi s met hod ar r anges t he gener at i ng uni t s i n a st ar t - up heur i st i c or der i ng by oper at i ng cost combi ned wi t h t r ansi t i on cost s. The pr e- det er mi ned or der i s t hen used t o commi t t he uni t s such t hat t he syst em l oad i s sat i sf i ed. Var i at i ons on t hi s t echni que dynami cal l y r ank t he uni t s sequent i al l y. The r anki ng pr ocess i s based on speci f i c gui del i nes. The Commi t ment Ut i l i zat i on Fact or ( CUF) and t he cl assi cal economi c i ndex Aver age Ful l - Load Cost ( AFLC) can al so be combi ned t o det er mi ne t he pr i or i t y commi t ment or der . The CUF met hod can be appl i ed t o ei t her a si ngl e- ar ea UC or a mul t i - ar ea uC. Shoul t s, R. R. , Chang, S. K. , Hel mi ck, S. , and Gr ady, W. M. , " A Pr act i cal Appr oach t o Uni t Commi t ment , Economi c Di spat ch, and Savi ngs Al l ocat i on f or Mul t i pl e- Ar ea Pool Oper at i on wi t h I mpor t / Expor t Const r ai nt s, It I EEE Tr ansact i ons on PAS- 99, No. 2 , pp. 625- 633, Mar ch/ Apr i l 1980. Lee, F. N. , "Shor t - Ter mUni t Commi t ment - A New Met hod, " I EEE Tr ansact i ons on Lee, F. N. , "The Appl i cat i on of Commi t ment Ut i l i zat i on Fact or ( CUF) t o Ther mal Uni t Commi t ment , " I EEE Tr ansact i ons on PWRS- 6, Lee, F. N. and Feng, Q., "Mul t i - Ar ea Uni t Commi t ment , " I EEE Tr ansact i ons on PWRS, Paper 91 WM 180- 0, New Yor k, 1991. heur i st i c of or der i ng can be t r ansl at ed r ul es and execut ed as an exper t svst emas PWRS- 3, NO. 2, pp. 421- 428, May 1988. NO. 2, pp. 691- 698, May 1991. not ed bel ow. Thus any of t hese Sechni hes can be t r eat ed as exper t syst emappr oaches si mpl y by usi ng an exper t syst emt ool . DYNAMI C PROGRAMMI NG Dynami c Pr ogr ammi ng ( DP) sear ches t he sol ut i on space t hat consi st s. of t he uni t s st at us f or an 130 opt i mal sol ut i on. The sear ch can proceed i n a f or war d or backwar d di r ect i on. The t i me per i ods of t he st udy hor i zon ar e known as t he st ages of t he DP pr obl em. Typi cal l y each st age r epr esent s one hour of oper at i on. The combi nat i ons of uni t s wi t hi n a t i me peri od ar e known as t he st at es of t he DP pr obl em. For war d DP f i nds t he most economi cal schedul e by st ar t i ng at t he i ni t i al st age accumul at i ng t ot al cost s, t hen backt r acki ng f r om t he combi nat i on of l east accumul at ed cost st ar t i ng at t he l ast st age and endi ng at t he i ni t i al st age. DP bui l ds and eval uat es t he compl et e deci si on t r ee t o opt i mi ze t he pr obl em at hand. Thus, DP suf f er s f r omt he "cur se of di mensi onal i t y" because t he pr obl em gr ows r api dl y wi t h t he number of gener at i ng uni t s t o be commi t t ed. To r educe t he sear ch space and hence t he di mensi on of t he DP pr obl em, sever al appr oaches have been adopt ed. Most appr oaches ar e based on t he above Pr i or i t y Li st t echni ques. One such met hod uses t he pr i or i t y l i st or der i ng wher e t he l east cost l y uni t s t o oper at e ar e commi t t ed f i r st and t he most cost l y uni t s ar e commi t t ed l ast . I n t hi s case, t he pr obl em i s reduced by consi der i ng combi nat i ons of uni t s sequent i al l y t ur ned on ( of f ) i n pr i or i t y l i st or der . Not e t hat unavai l abl e, must - r un, f i xed, and peaki ng uni t s ar e excl uded f r omt he pr esent pr i or i t y l i st . Anot her appr oach i s t o adopt a var i abl e pr i or i t y or der i ng scheme by or gani zi ng t he gener at i ng uni t s i nt o cl asses wi t hi n whi ch t he uni t s ar e pr i or i t i zed. A t hr eshol d and a wi ndow ar e def i ned i n each cl ass t o det er mi ne whi ch uni t s ar e aut omat i cal l y commi t t ed ( t hr eshol d) , whi ch uni t s ar e eval uat ed f or commi t ment ( wi ndow) , and whi ch uni t s ar e not consi der ed at al l . [ 19] Lower y, P. G. , "Gener at i ng Uni t Commi t ment Tr ansact i ons on PAS- 85, No. 5, pp. [ 20] Guy, J . D. , "Secur i t y Const r ai ned Uni t Commi t ment , " I EEE Tr ansact i ons on PAS- 90, No. 3, pp. 1385- 1389, May/ J une 1971. [ 21] Le, K. D. , Day, J . T. , Cooper , B. L. , and Gi bbons, E. W. , "A Gl obal Opt i mi zat i on Met hod f or Schedul i ng Ther mal Gener at i on, Hydr o Gener at i on, and Economy Pur chases, " I EEE Tr ansact i ons on PAS- 102, No. 7, pp. [ 22] Kusi c, G. L. and Put nam, H. A. , "Di spat ch and Uni t Commi t ment I ncl udi ng Commonl y Owned Uni t s, 'I I EEE Tr ansact i ons on PAS- 104, No. 9, pp. 2408- 2412, Sept ember 1985. by Dynami c Pr ogr ammi ng, I EEE 422- 426, May 1966. 1986- 1993, J ul y 1983. [ 23] Snyder , W. L. , Powel l , H. D. , J r . , and Raybur n, J . C. , "Dynami c Pr ogr ammi ng Appr oach t o Uni t Commi t ment , I' I EEE Tr ansact i ons on PWRS- 2, No. 2, pp. 339- 350, May 1987. [ 24] Hobbs, W. J . , Her mon, G. , War ner , S., and Shebl B, G. B. , "An Enhanced Dynami c Pr ogr ammi ng Appr oach f or Uni t Commi t ment , " I EEE Tr ansact i ons on PWRS- 3, NO. 3 , pp. 1201- 1205, August 1988. [ 25] Tong, S. K. and Shahi dehpour , S. M. , "Hydr ot her mal Uni t Commi t ment wi t h Pr obabi l i st i c Const r ai nt s Usi ng Segment at i on Met hod, " I EEE Tr ansact i ons on PWRS- 5, No. 1. pp. 276- 282, Febr uar y 1990. [ 26] Hsu, Y. Y. , Su, C. C. , Li ang, C. C. , Li n, C. J . , and Huang, C. T. , "Dynami c Secur i t y Const r ai ned Mul t i - Ar ea Uni t Commi t ment , " I EEE Tr ansact i ons on PWRS- 6, No. 3 , pp. Tr uncat ed DP i s anot her at t empt at r educi ng t he si ze of t he DP pr obl em. I n t hi s case, a smal l por t i on of t he sol ut i on space i s consi der ed wi t hi n pr i or i t y l i st or der i ng. The pot ent i al uneconomi cal commi t ment schedul es ar e t hen t r uncat ed. [ 27] Pang, C. K. and Chen, H. C. , "Opt i mal Shor t - Ter mTher mal Uni t Commi t ment , " I EEE Tr ansact i ons on PAS- 95, No. 4, pp. 1336- 1346, J ul y/ December 1976. [ 28] Pang, C. K. , Shebl B, G. B. , and Al buyeh, F. , "Eval uat i on of Dynami c Pr ogr ammi ng Based Met hods and Mul t i pl e Ar ea Repr esent at i on f or Ther mal Uni t Commi t ment s, I' I EEE Tr ansact i ons on PAS- 100, No. 3, pp. 1212- 1218, Mar ch 1981. Ref er ence 24 pr esent ed a basi c obser vat i on whi ch has not been addr essed by ot her DP appr oaches. Si mpl y st at ed, t he aut hor s r epor t ed t hat t he pr i nci pal of opt i mal i t y was f ound t o be vi ol at ed by t he syst em under st udy. Speci f i cal l y, i t was mor e opt i mal t o save sub- opt i mal sol ut i ons dur i ng t he f or war d pr ocess. Such an obser vat i on cl ear l y quest i ons t he use of DP f or t hi s pr obl em. The UC pr obl em may al so be decomposed i nt o smal l er subpr obl ems t hat ar e easi l y managed. Each subpr obl em i s sol ved wi t h DP. The subpr obl em coor di nat i on i s achi eved ei t her sequent i al l y wi t h Successi ve Appr oxi mat i on ( SA) or i n par al l el wi t h a Hi er ar chi cal Appr oach ( HA) . I n SA, t he sol ut i on of each subpr obl emi s subdi vi ded i nt o a smal l er gri d f or t he next subpr obl em and t he i t er at i ve pr ocedur e cont i nues unt i l no i mpr ovement i n t he sol ut i on i s det ect ed. I n HA, t he subpr obl ems ar e sol ved i ndependent l y of each ot her . The i nt er act i on bet ween t he subpr obl ems i s mani pul at ed by a coor di nat or t o conver ge t he sol ut i on of t he subpr obl ems t o t he over al l pr obl emsol ut i on. ~n al t er nat i ve t o t he decomposi t i on i s t o onl y appl y SA t o r est r i ct t he sol ut i on space of t he DP appr oach. One of t he var i ant s of t hi s appr oach uses Lagr angi an r educt i on of t he sear ch r ange by usi ng t he dual of t he UC pr obl em. Anot her var i ant uses a dual f unct i on of t he r el axed or i gi nal UC pr obl em. [ 29] Van den Bosch, P. P. J . and Honder d, G. , " A Sol ut i on of t he Uni t Commi t ment Pr obl em vi a Decomposi t i on and Dynami c Pr ogr ammi ng, If I EEE Tr ansact i ons on [ 30] Ni eva, R. , I nda, A. , and Gui l l en, I . , "Lagr angi an Reduct i on of Sear ch- Range f or Lar ge Scal e Uni t Commi t ment , I EEE Tr ansact i ons on PWRS- 2, No. 2, pp. 1049- 1055, August 1991. PAS- 104, NO. 7, pp. 1684- 1690, J ul y 1985. 465- 473, May 1987. 131 Systems, IFAC Symposium on Large Scale Systems, udine, Italy, June 1976. Dillon, T.S., Edwin, K.W., Kochs, H.-D., and Taud, R. J. , "Integer Programming Approach to the Problem of Optimal Unit Commitment with Probabilistic Reserve Determination," IEEE Transactions on PAS-97, No. 6, pp. 2154-2166, Nov/Dec 1978. Turgeon, A., "Optimal Scheduling of Thermal Generating Units, 'I IEEE Transactions on AC-23, No. 6, pp. 1000-1005, December 1978. Pereira, M.V.F. and Pinto, L.M.V.G., ltApplication of Decomposition Techniques to the Mid-Short-Term Scheduling of Hydrothermal Systems, I' Proceedings of the PICA, pp. 193-200, June 1983. Shaw, J.J. and Bertsekas, D.P., lvOptimal Scheduling of Large Hydrothermal Power Systems," IEEE Transactions on PAS-104, No. 2, pp. 286-294, February 1985. Habibollahzadeh, H. and Bubenko, J.A., "Application of Decomposition Techniques to Short-Term Operation Planning of Hydrothermal Power System, IEEE Transactions on PWRS-1, No. 1, pp. 41-47, February 1986. Recently, expert systems have been applied to the UC problem. UC is initially solved using available optimization techniques such as DP or variable window truncated DP and then the solution is refined by satisfying heuristic rules derived from knowledge of the system operation and conditions. Ouyang, 2 . and Shahidehpour, S.M., "Heuristic Multi-Area Unit Commitment with Economic Dispatch," IEE Proceedings, Part C, Vol. 138, No. 3, pp. 242-252, May 1991. Ouyang, 2. and Shahidehpour, S.M., i i ~ Intelligent Dynamic Programming for Unit Commitment Application, IEEE Transactions on PWRS-6, No. 3, pp. 1203-1209, August 1991. Finally, fuzzy DP has been used to solve the UC problem when the forecasted hourly loads are not exactly known. For an optimal solution, the DP model must express the hourly loads, the cost, and the system security in terms of fuzzy set notations. [33] Su, C.C. and Hsu, Y.Y., 8tFuzzy Dynamic Programming: An Application to Unit Commitment," IEEE Transactions on PWRS-6, All of the above techniques are use the same vision to solve the problem. They see the problem as a sequential decision process of when to start the next unit and which unit to start based on predicted (estimated) unit operation. This theme is expanded to include risk as shown below. INTEGERWMIXED--PROGRAMMING The solution of the UC problem based on the Benders approach partitions the problem into a nonlinear economic dispatch problem and a pure-integer nonlinear UC problem. The Mixed-Integer Programming (MIP) approach solves the UC problem by reducing the solution search space systematically through discarding the infeasible subsets. Dual programming is also suggested for the solution of the thermal UC problem. The general solution concept is based on solving a linear program and checking for an integer solution. If the solution is not integer, linear problems or subproblems are continuously solved. The problems are not similar because the number and type of integer variables are changed while holding the variables at a fixed integer value. Branching is the strategy adopted to determine which variables to hold constant. No. 3, pp. 1231-1237, August 1991. Garver, L. L. , "Power Generation Scheduling by Integer Programming - Development of Theory, AIEE Transactions No. 2, pp. 730-735, February 1963. Muckstadt, J.A. and Wilson, R.C., " An Application of Mixed-Integer Programming Duality to Scheduling Thermal Generating Systems," IEEE Transactions on PAS-87, Vol. 12, pp. 1968-1977, December 1968. Dillon, T.S. and Egan, G.T., "The Application of Combinatorial Methods to the Problems of Maintenance Scheduling and Unit Commitment in Large Power of these techniques may be viewed as a means of discarding the paths (branches) which are not expected to yield a better solution. BRANCH - - AND BOUND The Branch and Bound (B&B) approach essentially determines a lower bound to the optimal solution and then finds a near-optimal feasible commitment schedule. The branch-and-bound tree is searched for the "best" solution. The lower bound can be determined from a dual optimization problem that uses Lagrangian relaxation. Information obtained from the dual problem is instrumental in producing dynamic priority lists even though priority lists may not be necessary. These lists are useful in the determination of feasible solutions and help in the computation of an.upper bound on the solution. Only few nodes of the branch-and-bound tree are examined to obtain near-optimal solutions if an upper bound is found. Lauer, G.S. Sandell, N.R., Jr., Bertsekas, D.P., and Posbergh, T.A., "Solution of Large-scale Optimal unit Commitment Problems,f' IEEE Transactions on PAS-101, No. 1, pp. 79-86, January 1982. Cohen, A.I. and Yoshimura, M., "A Branch and Bound Algorithm for Unit Commitment," IEEE Transactions on PAS-102, No. 2, pp. 444-451, February 1983. The concept of a tree is most appropriate if a risk based approach'is to be used as discussed below. LINEAR PROGRAMMING Several Linear Programming (LP) approaches have been adopted to solve the large UC 132 pr obl em. Fi r st , t he pr obl em i s decomposed i nt o smal l er subpr obl ems vi a t he Dant zi g- Wol f e decomposi t i on pr i nci pl e. Each subpr obl emi s sol ved usi ng Li near Pr ogr ammi ng. Second, t he pr obl em i s sol ved wi t h t he r evi sed si mpl ex t echni que. A dual f or mul at i on wi t h a r educed basi s i s adopt ed al ong wi t h r el axat i on t echni ques. A mer i t - or der l oadi ng easi l y pr ovi des a st ar t i ng poi nt f or each schedul e. Last , t he di scr et e deci si on l i near pr ogr ammi ng appr oach i s appl i ed t o t he UC pr obl em wi t h heur i st i cs such as pr i or i t y or der i ng. Thi s t echni que i s an LP pr obl em wi t h di scr et e bounded var i abl es al l owed t o be ei t her t he l ower or upper bound of t he i nt er val . [ 44] St ot t , B. , Mar i nho, J . L. , and Al sac, O. , "Revi ewof Li near Pr ogr ammi ng Appl i ed t o Power Syst emReschedul i ng, " Pr oceedi ngs of t he PI CA, pp. 142- 155, May 1979. [ 45] Wai ght , J . G. , Bose, A . , and Shebl e, G. B. , "Gener at i on Di spat ch wi t h Reser ve Mar gi n Const r ai nt s Usi ng Li near Pr ogr ammi ng, n I EEE Tr ansact i ons on PAS- 100, No. 1, pp. 252- 258, J anuar y 1981. [ 46] Pi ekut owski , M. and Rose, I . A. , "A Li near Pr ogr ammi ng Met hod f or Uni t Commi t ment I ncor por at i ng Gener at i on Conf i gur at i on, Reser ve, and Fl ow Const r ai nt s, *I I EEE Tr ansact i ons on PAS7104, NO. 12, pp. 3510- 3516, December 1985. [ 47] Khodaver di an, E. , Br amel l er , A. , and Dunnet t , R . M. , "Semi - Ri gor ous Ther mal Uni t Commi t ment f or Lar ge Scal e El ect r i cal Power Syst ems, I' I EEE Pr oceedi ngs, Vol . 133, Par t C, No. 4, pp. 157- 164, may 1986. The ext ensi on of such t echni ques t o bi ddi ng pr ocedur es i s st r ai ght f or war d [ 3, 5, 7] . The essence of a bi ddi ng pr ocedur e i s anal ogous t o an open mar ket wher e each pl ant bi ds f or t he next cont r act t o pr ovi de power and ener gy. DYNAMI C AND LI NEAR PROGRAMMI NG The UC pr obl emi s sol ved usi ng r egul ar DP or DP wi t h successi ve appr oxi mat i on of t he sol ut i on space. LP sol ves t he economi c di spat ch wi t hi n UC f or t he cal cul at i on of t he pr oduct i on cost or t he opt i mal al l ocat i on of f uel . Dant zi g- Wol f e decomposi t i on, when used, par t i t i ons t he l i near pr ogr am i nt o smal l er , mor e manageabl e LP subpr obl ems. LP wi t h upper boundi ng i s al so an al t er nat i ve sol ut i on t echni que t o t he economi c di spat ch pr obl em. Wai ght , J . G. , Al buyeh, F. , and Bose, A . , "Schedul i ng of Gener at i on and Reser ve Mar gi n usi ng Dynami c and Li near Pr ogr ammi ng, I' I EEE Tr ansact i ons on PAS- 100, No. 5 , pp. 2226- 2230, May 1981. Van Meet er en, H. P . , "Schedul i ng of Gener at i on and Al l ocat i on of Fuel Usi ng Dynami c and Li near Pr ogr ammi ng, If I EEE Tr ansact i ons on PAS- 103, No. 7, pp. Shebl e, G. B. and Gr i gsby, L. L. , "Deci si on Anal ysi s Sol ut i on of t he Uni t Commi t ment Pr obl em, El ect r i c Power . Syst ems Resear ch, Vol . 10, No. 11, pp. 85- 93, November 1986. 1562- 1568, J ul y 1984. The essence of t hese t echni ques i s t o pr ovi de t he DP wi t h addi t i onal i nf or mat i on t o gui de t he sel ect i on of t he t r ee pat hs. Not e t hat r ef er ence 50 vi ewed t he deci si on pr ocess as how t o al l ocat e t he f i nanci al r esour ces f or uni t oper at i on. SEPARABLEPROGRAMMING Separ abl e Pr ogr ammi ng ( SP) assumes t hat t he obj ect i ve f unct i on i s concave and t he const r ai nt s ar e convex wi t h onl y one non- zer o var i abl e. Thi s speci al st r uct ur e can be expl oi t ed by LP. The A- separ abl e pr ogr ammi ng t echni que i s used wi t h gener al i zed upper boundi ng LP t o sol ve t he UC pr obl em. [ 51] Rahman, S. , "Power Syst em Oper at i on Schedul i ng usi ng Separ abl e Pr ogr ammi ng, " El ect r i c Power Syst ems Resear ch, Vol . 2, No. 4 , pp. 292- 303, December 1979. Thi s can al so be vi ewed as a t ype of LaGr angi an Rel axat i on appr oach. NETWORK ELQH PROGRAMMI NG Net wor k Fl ow (NF) Pr ogr ammi ng i s t he basi s f or schedul i ng most hydr o syst ems. Thus i t woul d be anot her anal ogy t o appl y t o t he uni t commi t ment pr obl em. The r esul t i s a nonl i near obj ect i ve f unct i on and a l i near set of const r ai nt s. Thi s pr obl emcan be sol ved wi t h a r educed gr adi ent al gor i t hm. I t can al so be sol ved wi t h a Fr ank- Wol f e t echni que. I n t hi s case, net wor k f l owr epl aces LP t o sol ve a par t of t he pr obl emi f t he non- net wor k const r ai nt s ar e not bi ndi ng. [ 52] Br annl und, H. , Sj el vgr en, D. , and Bubenko, J . A. , "Shor t - Ter m Gener at i on Schedul i ng wi t h Secur i t y Const r ai nt s, " I EEE Tr ansact i ons on PWRS- 3, No. 1, pp. 310- 316, Febr uar y 1988. [ 53] Habi bol l ahzadeh, H. , Fr ances, D. , and Sui , U. , "A New Gener at i on Schedul i ng Pr ogr am at Ont ar i o Hydr o, I EEE Tr ansact i ons on PWRS- 5, No. 1, pp. 65- 73, Febr uar y 1990. The anal ogy of Net wor k Fl ows al so appl i es t o t r ee based t echni ques as shown i n r ef er ence 11. LAGRANGI AN- The Lagr angi an Rel axat i on ( LR) opt i mi zat i on t echni que decomposes t he UC pr obl em i nt o a mast er pr obl emand mor e manageabl e subpr obl ems t hat ar e sol ved i t er at i vel y unt i l a The near - opt i mal sol ut i on i s obt ai ned. subpr obl ems ar e sol ved i ndependent l y. Each subpr obl em det er mi nes t he commi t ment of a si ngl e uni t . The pr obl ems ar e l i nked by LaGr ange mul t i pl i er s t hat ar e added t o t he mast er pr obl emt o yi el d a dual pr obl em. The dual pr obl em has l ower di mensi ons t han t he pr i mal pr obl emand i s easi er t o sol ve. For t he UC pr obl em, t he pr i mal f unct i on i s al ways gr eat er t han or equal t o t he f unct i on whi ch i s def i ned as weak dual ' i ty. The di f f er ence bet ween t he t wo f unct i ons yi el ds t he dual i t y gap f or whi ch t he pr i mal f unct i on i s an upper bound. The dual i t y gap pr ovi des a measur e of t he near - opt i mal i t y of t he sol ut i on. The LaGr ange mul t i pl i er s ar e comput ed at t he mast er pr obl em l evel . Once comput ed, t he 133 Pumped- St or age Hydr o, " I EEE Tr ansact i ons 1989. On PWRS- 4, NO. 3, pp. 1065-1073, August [65] Vi r mani , S., I mhof , K. , and Mukhenj ee, S., ' I mpl ement at i on of a Lagr angi an Rel axat i on Based Uni t Commi t ment Pr obl em, I EEE Tr ansact i ons on PWRS- 4, No. 4, pp. 1373-1379, November 1989. [66] Tong, S. K. and Shahi dehpour , S. M. , "An I nnovat i ve Appr oach t o Gener at i on Schedul i ng i n Lar ge- scal e Hydr o Ther mal Power Syst ems wi t h Fuel Const r ai ned Uni t s, " I EEE Tr ansact i ons on PWRS- 5, No. [67] Ruzi c, S. and Raj akovi c, N. , ''A New Appr oach f or Sol vi ng Ext ended Uni t commi t ment Pr obl em, " I EEE Tr ansact i ons on PWRS- 6, No . 1, pp. 269-275, Febr uar y 1991. 2, pp. 665-673, May 1990. Lagr ange mul t i pl i er s ar e passed t o t he subpr obl ems. The sol ut i on of t he subpr obl ems by f or war d DP i s f ed back t o t he mast er pr obl em and updat ed mul t i pl i er s ar e obt ai ned and used by t he subpr obl ems agai n. Thi s pr ocess i s r epeat ed unt i l t he sol ut i on conver ges. For t he shor t - t er mUC pr obl em, t he mul t i pl i er s ar e updat ed t hr ough a subgr adi ent met hod wi t h a scal i ng f act or and t uni ng const ant s t hat ar e det er mi ned heur i st i cal l y. For t he l ong- t er mUC pr obl em, t he mul t i pl i er s ar e updat ed wi t h t he var i abl e met r i c met hod t o pr event t he sol ut i on near t he dual maxi mum f romosci l l at i ng. [54] Fi sher , M. L. , "Opt i mal Sol ut i on of Schedul i ng Pr obl ems usi ng Lagr ange Mul t i pl i er s : Par t I , It Oper at i ons Resear ch, Vol . 21, pp. 1114-1127, 1973. [55] Muckst adt , J . A. and Koeni g, S. A. , I I A n Appl i cat i on of Lagr angi an Rel axat i on t o Schedul i ng i n Power Gener at i ng Syst ems, Oper at i ons Resear ch, Vol . 25, pp. 387-403, May/ J une 1977. [56] Ber t sekas, D.P., Lauer , G. S. , Sandel l , N. R. , J r . , and Posber gh, T. A. , "Opt i mal Shor t - Ter m Schedul i ng of Lar ge Scal e Power Syst ems, I' I EEE Tr ansact i ons on AC-28, No. 1, pp. 1-11, J anuar y 1983. [57] Mer l i n, A. and Sandr i n, P. , "A New Met hod f or Uni t Commi t ment at El ect r i ci t 6 de Fr ance, II I EEE Tr ansact i ons on PAS- 102, NO. 5, pp. 1218-1225, May 1983. [58] Fi sher , M. L. , " A n Appl i cat i on Or i ent ed Gui de t o Lagr angi an Rel axat i on, 'I I nt er f aces, vol . 15, No. 2, pp. 10-21, Mar ch/ Apr i l 1985. [59] Cohen, A. I . and Wan, S. H. , "A Met hod f or Sol vi ng t he Fuel Const r ai ned Uni t Commi t ment Pr obl em, " I EEE Tr ansact i ons on [60] Aoki , K. , Sat oh, T. , I t oh, M. , I chi mor i , T. , and Masegi , K. , "Uni t Commi t ment i n a Lar ge- scal e Power Syst emI ncl udi ng Fuel Const r ai ned Ther mal and Pumped- St or age Hydr o, " I EEE Tr ansact i ons on PWRS- 2, No . 4, pp. 1077-1084, November 1987. [61] Zhuang, F. and Gal i ana, F. D. , "Towar ds a mor e Ri gor ous and Pr act i cal Uni t Commi t ment by Lagr angi an Rel axat i on, " I EEE Tr ansact i ons on PWRS-3, No. 2 , pp. 763-773, May 1988. [62] Bar d, J . F. , "Shor t - Ter m Schedul i ng of Ther mal El ect r i c Gener at or s Usi ng Lagr angi an Rel axat i on, " Oper at i ons Resear ch, Vol . 36, No. 5 , pp. 756-766, Sept ember / Oct ober 1988. [63] Tong, S. K. and Shahi dehpour , S. M. , "Combi nat i on of Lagr angi an- Rel axat i on and Li near Pr ogr ammi ng Appr oaches f or Fuel Const r ai ned Uni t Commi t ment Pr obl ems, " I EE Pr oceedi ngs, Vol . 136, Par t C, No. 3, [64] Aoki , K. , I t oh, M. , Sat oh, T. , Nar a, K. , and Kanezashi , M. , "Opt i mal Long- Ter m uni t Commi t ment i n Lar ge Scal e Syst ems I ncl udi ng Fuel Const r ai ned Ther mal and PWRS- 2, No. 3, pp. 608-614, August 1987. pp. 162-174, May 1989. EXPERT SYSTEMS/AR= NEURAL NETWORKS Exper t syst ems combi ne t he i dent i f i cat i on of exi st i ng pr obl ems wi t h t he UC al gor i t hms and t he knowl edge of exper i enced power syst em oper at or s and UC pr ogr ammi ng exper t s t o cr eat e an exper t syst em r ul e base ( pr ocedur al dat a base) . The exper t syst em i mpr oves t he sol ut i on by adj ust i ng t he pr ogr am' s par amet er s t hr ough i nt er act i on wi t h t he syst em' s oper at or . Exper t Syst ems ar e mor e r ecent l y ref erred t o as Knowl edge Base Syst ems ( KBS) . Est i mat es of Ar t i f i ci al Neur al Net wor ks ( ANN) par amet er s ar e based on a dat abase hol di ng t ypi cal l oad cur ves and cor r espondi ng UC schedul es. The pat t er n of t he cur r ent l oad cur ve i s compar ed t o t he i nf or mat i on i n t he dat abase t o sel ect t he most economi cal uc schedul e. I n t he event t hat t he A" sol ut i on i s not f easi bl e f or t he ent i r e UC per i od, i t wi l l be used as an i ni t i al st ar t i ng poi nt f or a near - opt i mal sol ut i on. A " s r espond t o changes i n oper at i ng condi t i ons when pr esent ed wi t h suf f i ci ent f act s, even t hough t hey ar e t r ai ned of f - l i ne. [68] Mokht ar i , S. , Si ngh, J . , and Wol l enber g, B. , " A Uni t Commi t ment Exper t Syst em, " Pr oceedi ngs of t he PI CA, pp. 400-405, May 1987, [69] Shebl b, G. B. , "Sol ut i on of t he Uni t Commi t ment Pr obl emby t he Met hod of Uni t Per i ods ,' I I EEE Tr ansact i ons on PWRS-5, No. 1, pp. 257-260, Febr uar y 1990. [70] Wang, C. , Ouyang, Z., and Shahi dehpour , S. M. , and Deeb, N . , "uni t Commi t ment by Neur al Net wor ks, " Pr oceedi ngs of Amer i can Power Conf er ence, vol . 52, pp. 245-250, Apr i l 1990. [71] Ouyang, 2. and Shahi dehpour , S. M. , "Shor t Ter m Uni t commi t ment Exper t Syst em, " El ect r i c Power Syst ems Resear ch, Vol . 18, No. 1, pp. 1-13, December 1990. 721 Tong, S.K., Shahi dehpour , S. M. , and Ouyang, "A Heur i st i c Shor t - Ter m Uni t Commi t ment , " I EEE Tr ansact i ons on PWRS- 6, NO. 3, pp. 1210-1216, August 1991. 731 Ouyang, Z. and Shahi dehpour , S. M. , " A Mul t i - St age I nt el l i gent Syst em f or uni t Commitment," IEEE Transactions on PWRS, Paper SM 322-8, San Diego, 1991. Sasaki, H., Watanabe, M., kubokawa, J., Yorino, N., and Yokoyarna, R., "A Solution Method of Unit Commitment by Artificial Neural Networks, '' IEEE Transactions on PWRS, Paper 91 SM 437-4, San Diego, 1991. Ouyang, Z. and Shahidehpour, S. M. , "A Hybrid Artificial Neural Network-Dynamic Programming Approach to Unit Commitment," IEEE Transactions on PWRS, Paper 91 SM 438-2, San Diego, 1991. It is notable that considerable research is underway to relate the trees of A"s to those of KBSs. It should be noted that strict conversion, from one type of KBS to ANN, is not possible. However, the addition of simple decoupling neurons (nodes ) makes the transition trivial for several practical cases. RI SKANALYSI S Scheduling and rescheduling of generating units are based on certain observable events such as load changes and unit forced outages. A risk analysis is warranted to determine the necessary additional generating capacity to meet the system load and reserve requirements. One such probabilistic analysis proposes that the UC must satisfy two risk levels: one at the isolated system level and another at the interconnected level. Another approach adopts a stochastic model that reflects the sequence of events associated with scenario-based sequential rescheduling decisions. The random sequence model is expressed in terms of available capacity to reduce the dimensions of the problem. [76] Chowdhury, N. and Billinton, R., "Unit Commitment in Interconnected Generating Systems Using a Probabilistic Technique," IEEE Transactions on PWRS-5, No. 4, pp. 1231-1237, November 1990. [77] Lee, F.N. and Chen, Q., "Unit Commitment Risk with Sequential Rescheduling," IEEE Transactions on PWRS-6, NO. 3, pp. Risk Analysis is most recently extended to include operational and planning events beyond unit Commitment as discussed below. a MULATm ANNEAL1 NG The UC problem has been compared to the annealing of a metal. When the metal is cooled slowly (annealed), its energy tends to assume a globally minimal value. The states of a metal correspond to the various feasible solutions of the problem to minimize and the energy of a state is analogous to the cost of a feasible solution. Simulated Annealing (sm) generates near-optimal and fast solutions. Feasible solutions are generated randomly and are accepted as the next generation to continue the solution process if the cost of the current solution is less than the previous one. Otherwise, the current solution is accepted with a certain probability. 1017-1023, August 1991. [78] Zhuang, F. and Galiana, F.D., "Unit Commitment by Simulated Annealing," IEEE Transactions on PWRS-5, NO. 1, pp. 311-318, February 1990. AUGMENTE D MGRANG IAN This is an optimization technique that handles static and dynamic constraints. The two types of constraints are decomposed into subproblems of reasonable size, homogeneous nature, and well-known structure. The objective function to optimize is unconstrained and continuously differentiable. Ill-conditioning is avoided. Artificial constraints and variables are added to the original problem to decompose it. Next, the artificial constraints are handled via a dual approach and an augmented Lagrangian (AL) relaxation technique which adds a quadratic penalty function to the constraints. Finally, The Auxiliary Problem Principle is applied to decompose the problem by linearizing the nonseparable terms of the cost function and by adding separable quadratic terms (if properly chosen) to the cost function. [ 791 Batut, J. and Renaud, A . , "Daily Generation Scheduling Optimization with Transmission Constraints: A New Class of Algorithms," IEEE Transactions on PWRS, Paper 91 SM 429-1, San Diego, 1991. DECISIONANALYSIS Decision Analysis is the art of analyzing the choice of decision options by subjectively assessing the outcomes of each decision and the probability of each outcome. The reference which the authors have used IS: [80] Raiffa, H. A., DecisiQn An alvsis - : y n c e r t a w , Addison-Wesley Publishing Company, Reading, Massachusetts, 1968. Such an approach builds a decision tree to show the outcomes possible from each decision. Such a tree resembles the paths generated by a full DP approach. If the outcomes are assumed to be certain,then all of the above may be viewed as tree based techniques. Many of the above optimization techniques may be viewed as different approaches to pruning the tree. Some approaches prune by excluding paths above a pre-determined cost value (truncated DP, priority list modification, KBS). Other methods prune by excluding branches (paths) by estimating the minimum cost and/or feasibility (MIP, B&B, LP, SP, LR, AL). Finally, the latest approach is to select the best path by evaluating the energy (quality of solution) and randomly searching for alternative decisions to reduce the energy ( S A n ) . The authors have started efforts on solving the uc problem with Genetic Algorithms and will present such work when definitive results are available. Since the tree approach is being used for bulk power assessment, the key to future efforts will be to explicityly link the above techniques with decision trees. It has been realized for some time, that the reason for solving the unit commitment problem is not just to determine a unit schedule. The mtroductorv Lect ures Choices under 135 Gerald B. Sheble' ( U 71, SM 85) i s an Associ at e Pr of essor of El ect r i cal Engi neer i ng, I owa St at e Uni ver si t y. Dr . Shebl 6 r ecei ved hi s B. S . and M. S. degr ees i n El ect r i cal Engi neer i ng f r om Pur due Uni ver si t y and hi s Ph. D. i n El ect r i cal Engi neer i ng f r omVi r gi ni a Tech. Hi s i ndust r i al exper i ence ext ends over f i f t een year s. Hi s academi c exper i ence i ncl udes r esear ch i n t he appl i cat i on of power syst ems t echni ques f or spacecr af t . Hi s pr esent r esear ch cent er s i n t he opt i mal oper at i on and schedul i ng of power syst ems. George Fahd (S 85) r ecei ved hi s B. S. E. E. i n December of 1985 and hi s M. S. E. E. i n August of 1987 f r om t he Uni ver si t y of Al abama at Bi r mi ngham. He r ecei ved t he Ph. D. maj or i ng i n Power Syst ems f r omAubur n Uni ver si t y i n J une, 1991. He hel d a post doct or at e posi t i on at I owa St at e Uni ver si t y bef or e j oi ni ng Deci si on Focus I ncor por at ed I n December of 1991. Hi s i nt er est s i ncl ude uni t commi t ment model i ng, economi c di spat ch, opt i mal power f l ow, opt i mi zat i on t echni ques, power syst em oper at i on and cont r ol , and economy t r ansact i ons. pr i mar y r eason f or sol vi ng t he uni t commi t ment pr obl em i s t o pr ovi de a cost basi s f or t r ansact i on pr i ci ng. As such, r esear ch f or t he f ut ur e shoul d concent r at e on r el at i ng t he uni t schedul e t o t he avai l abl e t r ansact i ons wi t h t he i nt ent of sel ect i ng t he l east cost , yet r el i abl e, opt i on. SUMMARY Thi s paper gi ves a l i st of t he r ef er ences avai l abl e f or t he sol ut i on of t he t her mal UC pr obl em. A var i et y of t echni ques have been appl i ed t o t hi s compl ex, nonl i near , mi xed- i nt eger pr ogr ammi ng pr obl em. A cl ear consensus i s pr esent l y t endi ng t owar d t he Lagr angi an Rel axat i on appr oach over ot her met hodol ogi es. The Augment ed Lagr angi an i s a r el at i vel y newcomer t hat i s not t hor oughl y t est ed yet . Anot her ar ea of r esear ch i nvol ves t he appl i cat i on of genet i c al gor i t hms t o t he sol ut i on of t he UC pr obl em. Thi s i s a r andom sear ch met hod whi ch i s based on nat ur e' s sur vi val of t he f i t t est t heor y. Cl ear l y, t he met hods t o be used f or t hi s pr obl em wi l l cont i nue t o evol ve.