Reliability Assessment at Day-ahead Operating Stage in Power Systems with Wind Generation

Transcription

1 th Hawa Internatonal Conference on System Scences Relablty Assessment at Day-ahead Operatng Stage n Power Systems wth Wnd Generaton Le Xe, Member, IEEE, Ln Cheng, Member, IEEE, and Yngzhong Gu, Student Member, IEEE Abstract Ths paper proposes to evaluate relablty performances at operatng stage for power systems wth varable wnd generaton. Ths s a frst step towards recognzng the stronger couplng of power system decson makng across dfferent tmescales. In the day-ahead energy-reserve co-optmzaton unt commtment model, the valdty of usng operatng reserve as an approxmaton of long-term relablty requrement s tested wth realstc wnd generaton profle. Relablty ndces such as Loss of Load Probablty (LOLP) and Loss of Expected Energy (LOEE) are computed at each hour of the day at the concluson of the day-ahead unt commtment decson. Numercal experments conducted n the IEEE Relablty Test System suggest that whle operatng reserve requrement does not change wthn a day, the actual relablty performance of the system vares sgnfcantly dependng on () system loadng condtons, () wnd power varaton, and () operatng rules. Also, the operatng reserve requrement s shown to ether overapproxmate or under-approxmate the relablty requrement of power system. It ndcates that operatng reserve could be a very coarse determnstc approxmaton of system relablty requrement especally wth hgh wnd penetraton. A much more ntegrated approach for modelng, analyss, and decson makng s envsaged for provson of relable and cost-effectve electrcty servces. NOMENCLATURE The notatons used n ths paper are summarzed n Table I. I. INTRODUCTION Ths paper s motvated by the needs of the next-generaton power system, whch s expected to have a sgnfcant amount of varable energy resources, especally wnd power. Wth today s operatng and plannng paradgm under worst-case reserve desgns, the multple objectves (clean, affordable, and relable energy) of balancng supply and demand of power are often conflctng [1]. In partcular, t has been shown n varous ndustry study that wth the antcpated amount of renewable penetraton, the operatng reserves requred to mantan system relablty would be so hgh that largely offset the potental benefts of clean energy resources [2], [3]. The mpact of varable energy resources on power system relablty performance has accumulated a large body of lterature n recent years. Gven the conventonal herarchy underlyng today s power system engneerng practces, the lterature can be roughly dvded nto two categores. The frst L. Xe and Y. Gu are wth Department of Electrcal and Computer Engneerng, Texas A&M Unversty, College Staton, TX. Emal: Lxe@ece.tamu.edu. L. Cheng s wth the State Key Laboratory of Power Systems, Department of Electrcal Engneerng, Tsnghua Unversty, Bejng , Chna. Emal:chengln@mal.tsnghua.edu.cn TABLE I NOTATIONS C G Generaton cost of generator. C RS Reserve cost of generator. C U Startup cost of generator. C D Shutdown cost of generator. c k U Bnary ndcators of startng-up generator. c k D Bnary ndcators of shuttng-down generator. C Avalable capacty n tme perod. E Total load energy requred to serve the system demand. F k Vector of branch flow at the tme step k. LOLP Loss of load probablty. LOEE Loss of energy expectaton. L The forecast maxmum load n perod. P(C < L) Probablty that maxmum load exceeds avalable capacty. P L (t) Load n real tme t at locaton. PG k Output level of generator at tme step k. PG max PG mn P k RS Maxmum generaton output. Mnmum generaton output. Dspatched capacty for provdng reserve servces. PL k System wde load level at tme step k. ΔP L [k] Mnute-by-mnute load devaton around the forecast level. x k On/off status of generator at tme step k. category focuses on the mpact of varable generaton on longterm plannng practces, e.g. [4], [5]. A probablstc approach s used to determne the generaton capacty adequacy wth sample spaces large enough to cover uncertanty n the longterm (e.g. year-ahead or longer). Relablty ndces such as loss of load probablty (LOLP) or loss of energy expectaton (LOEE) are compared aganst the standards North Amercan Electrc Relablty Councl (NERC) or ts counterparts n other regons. As an determnstc approxmaton of relablty performances, operatng reserve requrements can be quantfed for operatonal practces [6], [7]. The second category of the lterature focuses on understandng the mpact of varable generaton on short-term operatng practces n power systems. At the operatonal stage, relablty requrement s typcally approxmated as a determnstc functon or rule, whch s typcally related wth peak demand at the operatng day and the largest unts n the system [6]. Wthn the year, as operatng condtons change from tme to tme, the rule or functon of operatng reserve typcally stay unchanged [7]. Regonal numercal studes as well as theoretcal analyss have been conducted to assess the ncreasng need of operatng reserve due to ncreased uncertantes n varable resources [2], [3], [8], [9]. Based on worst-case contngences, system operators typcally wll construct rule-based emprcal approach to quan /12 $ IEEE DOI /HICSS

2 tfyng operatng reserves n order to comply wth relablty standards specfed by NERC [10] [12]. Ths conventonal herarchy of plannng and operatng practce s based on the assumptons that (1) most generaton resources are dspatchable and can contrbute to generaton capacty; (2) system supply and demand uncertantes do not vary sgnfcantly n typcal operatng days; and (3) there s lack of onlne data and computatonal power for drect couplng of plannng and operatng practces [13], [14]. However, the ncreasng penetraton of varable uncertan wnd power may render these assumptons nvald [1]. Enabled by advanced sensng, communcaton, and computatonal technologes, t becomes not only possble, but also necessary to revst the boundary between plannng and operatng practces. Fg. 1 llustrates the couplng of plannng and operatonal plannng practces n power systems. Conventonal plannng (yearly and quarterly) and operatonal plannng (daly and hourly) dvsons n regonal transmsson organzatons (RTOs) share the responsblty of mantanng system relablty through the settng of operatng reserve requrement. However, the couplng s weak and un-drectonal. As an approxmaton to system relablty performances, the requrement of operatng reserve s determnstc and s updated very rarely [10] [12]. In ths paper, we propose a tghter condton-dependent feedback between operatng and plannng n order to mantan relable and affordable provson of electrcty servces n the future. In partcular, the contrbuton of ths paper s suggested as follows: The relatonshp between short-term operatonal reserve and long-term relablty requrement s revsted; A realstc wnd power generaton model s ncorporated n assessment of operatonal relablty at day-ahead operatonal stage; II. POWER SYSTEM RELIABILITY AT PLANNING STAGE The requred amount of system generatng capacty to ensure adequate supply can be dvded nto statc and operatng capacty requrements [7]. The fundamental dfference between statc and operatng capacty evaluaton s n the tme perod consdered. There are fundamental dfferences n the data used n each area of applcaton. The statc capacty s related wth the long-term evaluaton of the overall system relablty requrement. The statc reserve must be suffcent to provde for the overhaul of generatng equpment, outages that are not planned or scheduled, and load growth requrements n excess of the estmated values. Loss of load probablty or smlar expectaton methods are the most wdely used probablstc technques for evaluatng the adequacy of a gven generaton confguraton. In ths approach, the system capacty outage probablty table s combned wth the system load characterstc to gve an expected rsk of loss of load. In contrast wth statc capacty, the operatng capacty s related wth the short-term evaluaton of the actual capacty requrement for meetng a gven load level. Reserve must be scheduled n order to account for load forecast uncertantes and possble outages of generaton plant. The PJM method [6] has been consderably refned and enhanced for evaluatng operatonal reserve requrement, or unt commtment rsk. The bass of the PJM method s to evaluate the probablty of commtted generaton just satsfyng the expected demand durng the perod of tme that generaton cannot be replaced. Ths tme perod s known as the lead tme, whch may be several hours n the case of thermal unts or a few mnutes n the case of gas turbnes and hydroelectrc plant. The basc dfference n ths evaluaton s that the ORR (Outage Replacement Rate) of each unt s used nstead of the FOR (Forced Outage Rate) to construct the capacty outage probablty table. To evaluate the adequacy of a gven generaton confguraton, the followng relablty ndces are often used. The frst category s the Loss-Of-Load ndces. Such ndces are based on capacty or power. The LOLE (Loss-Of-Load Expectaton) s the expected perod durng whch the system load exceeds the avalable generatng capacty. The LOLP (Loss-Of-Load Probablty) gves the probablty that load exceeds the avalable generatng capacty n a gven tme perod. The LOLP can be calculated as follows Fg. 1. Relatonshp Between Plannng, Operaton, and Operatng reserve The rest of ths paper s organzed as follows. Bref overvew of power system relablty assessment at plannng stage s provded n Secton II. In Secton III we dscuss the quantfcaton of operatng reserve as an approxmaton of system relablty requrement. Secton IV formulates a realstc wnd generaton model whch can be used for operatonal relablty assessment at day-ahead unt commtment stage. Secton V formulates the day-ahead energy-reserve co-optmzaton unt commtment. Illustratve examples are shown n Secton VI. Concludng remarks and future work are dscussed n Secton VII. LOLP = LOLE p.u. = n P (C <L ) (1) =1 n whch refers to the ndex actng as tme step (e.g., an hour) and n refers to the full tme perod (e.g., 8760 hours or 24 hours). Avalable capacty n tme perod s represented as C. L refers to the forecast maxmum load n perod. P(C < L) refers to the probablty that maxmum load exceeds avalable capacty n perod. The second category of relablty ndces ncludes energy nformaton or the depth of power falure. The EUE (Expected Unserved Energy) s the expected amount of energy not suppled by the generatng system. The EUE s sometmes also called LOEE (Loss-Of-Energy Expectaton) or EENS

3 (Expected Energy Not Suppled). The EIU (Energy Index of Unrelablty) gves rato of the EUE and the total energy demand. LOEE = n E P (2) =1 EIU = LOEE(p.u.) = n =1 E P E n whch E refers to total load energy requred to serve the system demand. III. RELIABILITY REQUIREMENT AT OPERATING STAGE The fundamental objectve of power system operaton s to mantan the generaton and demand balance, subject to the transmsson network constrants and a set of plausble contngences. Conventonal power systems have been desgned and operated under well engneered spatal and temporal herarches, so that generaton s controlled through both feedforward optmzaton and feedback control to balance the fluctuatng loads [14], [15]. Load P L (t) n real tme t at locaton can be represented as: P L (t) = ˆP L [K]+ΔP L [k]+ɛ L (t) (4) where ˆP L [K] represents the forecast load at 10- or 15-mnute nterval K, ΔP L [k] corresponds to the mnute-by-mnute load devaton around the forecast level ˆPL [K], and ɛ L (t) represents the real-tme nose around the mnute-level load. Load forecastng technques have been developed n the past 50 years such that (a) ˆP L [K] ΔP L [k] ɛ L (t) ; and (b) ɛ L (t) s assumed to be Gaussan nose [16], [17]. Uncertanty ΔP L [k]+ɛ L (t) s consdered n determnng the operatng reserve requrement durng offlne plannng studes. Relablty threshold (e.g., LOLP) s reflected n the determnaton of operatng reserve requrement [6], [8]. Wth sgnfcant penetraton of varable resources, wnd/solar generaton are treated as negatve loads [9]. Ths ncreases sgnfcantly the overall net load uncertanty, leadng to a doubled or trpled reserve requrement for 20% wnd penetraton n regons lke Calforna and Texas [2]. Operatng reserve refers to a certan amount of capacty whch s reserved to avod power shortage due to loss of generatons, fluctuatons n demand and ntermttent resources [18]. Reserve are desgned to ensure the system relablty, but nstead of accurate assessment of relablty ndces, most of the ISOs consder the reserve requrement n a more practcally mplementable way. Despte the dfferent features and structures n detaled mplementaton, reserve quantfcaton and procurement n dfferent RTOs have shared core components [11], [19], [20]. These shared components nclude 1) uncertantes n loads and 2) falure of unts (contngency reserve). We take Electrc Relablty Councl of Texas (ERCOT), Independent System Operator (ISO)-New England and Chna Southern Power Grd as examples. In ERCOT, the requrement (3) of Non-spnnng reserve (NSRS) s determned by 1) the 95th percentle of the observed hourly net load 1 uncertanty from the prevous 30 days from the operaton day and from the same month of the prevous year, and 2) the capacty of the largest unt nstalled n the system. In ISO New England, the locatonal operatng reserve requrement s calculated accordng to 1) second generaton contngency nterface lmt (N-2 crteron), 2) the external reserve support (ERS), 3) forecast daly peak load, 4) mnmum capacty commtments requred for 1st contngency coverage (N-1 crteron). In Chna Southern Power Grd, the operatonal reserved requrement s calculated by 2% of system-wde peak demand plus (for load fluctuaton) the capacty of the largest unt n the system (for unt falure) [20]. In the numercal experments, the reserve determnaton methodology of Chna Southern Power Grd s adopted. The aforementoned approaches serve as mplementable approxmaton of power system relablty requrements at operatng stages such as n day-ahead unt commtment. IV. REALISTIC WIND GENERATION MODEL FOR RELIABILITY ASSESSMENT Wth the rapd ncrease of wnd power penetraton, ts stochastc power varaton challenges the grd operaton snce the wnd power generaton cannot be controlled lke the conventonal power generaton. An approprate wnd power varaton model can help the planners and operators to arrange adequate reserves. Worldwde efforts have been focused on ths topc and several models to descrbe the stochastc wnd power varaton have been proposed [5]. However, most of the exstng models smulate the wnd power varaton by assumng that wnd turbnes do not fal. Ths assumpton, however, s not supported by the feld experence. In [21], a model for assessng the wnd power varaton consderng the outage probablty of the wnd turbnes s proposed. Frstly, a new probablty-based model whch represents the relatonshp between the outage probabltes of wnd turbnes and wnd speed s proposed. Equaton (IV) s the proposed model for the outage probablty of wnd turbnes. p(v w )=av 3 w + bv 2 w + cv w + d (5) where v w represents the wnd speed and p(v w ) s the correspondng outage probablty of the wnd turbne. Then ths model s used to extend a frequency doman wnd power output model [22], whch was orgnally desgned wthout consderng the wnd turbne outages. The new model sgnfcantly mproves the accuracy of smulaton result of wnd power probablty dstrbuton, partcularly wthn the hgh wnd speed regon. Ths model s adopted to evaluate the plannng and operatng capacty requrements n the presence of wnd power penetraton. The mnmum and maxmum powers curves wth model correcton for turbne outages are plotted n Fg. 2 and are represented by sold and dash lnes. Fg. 2 also provdes an llustratve range n whch the wnd 1 Net Load s defned as the ERCOT load mnus the estmated total output from WGRs [11]

4 power output vares, so the model can be valdated by dentfyng whether the measurement data ponts are dstrbuted wthn the maxmum and mnmum power curves. If the wnd speed s gven, probablty dstrbutons of the wnd power output varatons can be obtaned from ths model. The multstate model of wnd farm s deduced and adopted to evaluate operatonal reserve requrements, or the unt commtment rsk. (approxmated by the output power correspondng to the peak probablty) obtaned from the model wthout outage model correcton s hgher than the measurement. Obvously, the outage of wnd turbnes decreases the onlne number of wnd turbnes, whch also changes the dstrbuton of wnd power output. Fg. 2. The Mnmum and Maxmum Powers Curves wth Model Correcton for Turbne Outages [21] Probablty dstrbutons of the wnd power output varatons are plotted n Fg. 3 to Fg. 4, correspondng to wnd speed regon of 5 m/s to 5.5 m/s, 10m/s to 10.5 m/s. In the low wnd speed regon, the probablty dstrbutons of the measurements, wth and wthout outage model correcton are smlar wth each other as shown n Fg. 3. The dstrbuton n ths fgure s a bt asymmetrcal. It s because the power output value cannot be negatve, thus the probablty dstrbuton n the low wnd speed regon concentrates n the postve-half of the plane. In Fg. 3, the three curves match well wth each other, confrmng that the low outage probablty has less effect on dstrbuton of the wnd power varaton at low wnd speed. Fg. 3. The Probablty Dstrbuton of Wnd Power wthn 5-5.5m/s [21] In the medum speed range, Fg. 4 shows that the model wth correcton for turbne outages stll provdes a good match to the feld data. However, the dstrbuton curve obtaned by the model wthout outage correcton starts to shft somewhat to the rght from the measured data. The mean power output Fg. 4. The Probablty Dstrbuton of Wnd Power wthn m/s [21] V. ENERGY-RESERVE CO-OPTIMIZATION IN DAY-AHEAD UNIT COMMITMENT In ths secton, we formulate the energy-reserve cooptmzaton n the day-ahead unt commtment operaton. Unt commtment s to determne when to startup or shutdown generatng unts and how to dspatch generators to meet system demand and reserve requrements whle satsfyng varous securty constrants over a specfc tme span (e.g., 24 hours), so that the overall operaton cost s mnmzed. The unt commtment problem can be formulated as follows, mn : f = T C G p k G + C RS p k RS + C U c k U + C D c k D G t=t 0 (6) Subject to p k G = PL,k k =1...T (7) G p k RS = DRS,k k =1...T (8) G F max F k F max,k =1...T (9) PG k + PRS k x k PG max, G, k =1...T (10) x k PG mn PG k x k PG max, G, k =1...T (11) 0 PRS k PG max, G, k =1...T (12) x k x k 1 c k U, G, k =1...T (13) x k 1 x k c k D, G, k =1...T (14) x k,c k U,c k D Bnary, G, k =1...T (15) where, C G s the generaton cost of generator ; C RS s the reserve cost of generator ; PG k s the output level of generator at tme step k, PG max and PG mn are ts upper and lower bounds; PRS k s the selected capacty n reserve servces, PL k

5 s the system wde load level at tme step k; F k s the vector of branch flow at the tme step k and F max s the vector of transmsson constrants of branches; C U s the startup cost of generator ; C D s the shutdown cost of generator ; x k s the on/off status of generator at tme step k, c k U and c k D are the bnary ndcators of startng-up and shuttng down generator. The objectve (6) s to mnmze the total operatng costs whch nclude generaton cost, reserve cost, start-up cost and shut-down cost. (7) are the power balancng equatons. (8) are the reserve balancng equatons. (9) are the branch flow constrants. (10) are the generaton and reserve sharng capacty constrants. (11) are the generaton capacty constrants. (12) are the reserve capacty constrants. (13) are the unts startng-up constrants. (14) are the unts shuttngdown constrants. (15) are the constrants for bnary decson varables such as the on/off status of each unt. In the next secton, system relablty performance s evaluated at each hour of day after the unt commtment operatng decsons. VI. NUMERICAL ILLUSTRATION Numercal experments are conducted to evaluate the relablty of the power system operaton wth wnd penetraton. Wnd generaton capacty s 20.9% of the overall generaton capacty. The test system used n ths study s a modfed IEEE relablty test system (24 bus) [23]. A securty constraned unt commtment (SCUC) s conducted over 24 hours wth hourly ntervals. The generators confguraton parameters, as shown n Table II, are based on [23]. The load pattern s also taken from [23]. The reserve costs of generators are confgured accordng to [24]. Wnd profles of 48 hours are collected from the Electrc Relablty Councl of Texas (ERCOT) [25]. Installed wnd capacty and producton potental are scaled accordng to the portons of overall system capacty. The reserved requrement s calculated by the sum of regular reserve whch conssts of 2% of demand plus the largest unt n the system, accordng to [20] and the reserve correlated wth wnd generaton uncertanty, descrbed n Secton IV. The dagram of the test system s presented n Fg. 5. The numercal experments are conducted n a PC wth Intel Core GHz CPU, 4GB Memory, Matlab 2011a and Wndows 7 Operatng System. TABLE II GENERATOR PARAMETERS Unt Group Type Pmax Pmn GEN.COST RS.COST (MW) (MW) ($/MWh) ($/MW) O12 Ol O20 Ol H50 Hydro C76 Coal O100 Ol C155 Coal O197 Ol C350 Coal N400 Nuclear W15 Wnd N/A Fg. 5. Dagram of IEEE Relablty Test System 24 Bus [23] A. Dscusson In ths subsecton, the smulaton results are analyzed. Accordng to dfferent rules handlng wnd generaton, the followng four cases are studed: Case A (optmstc case): The forecasted wnd producton potental s consdered 100% relable. Case B (pessmstc case): The forecasted wnd producton potental s completely unrelable. Case C: The falure rates of wnd turbnes are consdered (FOR = 0.10). Case D: The stochastc wnd power varaton model (descrbed n Secton IV) s consdered. The uncertanty of both wnd producton potental and turbne falure are taken nto account. Accordng to the SCUC results, the system load and onlne generaton capacty are provded n Fg. 6. From md-nght to early mornng, the spnnng reserve capacty stays at a low level. As the demand goes up durng the day, more unts are started and synchronzed to the power grd. In the late evenng, the demand drops. Therefore, several expensve unts were shut down. The relablty ndces of the aforementoned four cases are calculated for the operaton day. The loss of load probablty (LOLP) ndces of the four cases are presented n Fg. 7. Based on the same operatng reserve requrement, the actual system relablty vares throughout the day. For example, n Case D, durng the early mornng the LOLP s as low as 0.02 but durng the peak load perod (2 pm to 4 pm), the LOLP can reaches as hgh as Durng a typcal day the peak to valley rato of LOLP can be as hgh as 15 tmes. Accordng to the relablty assessment results, even n the optmstc case (Case A), the current reserve based approach gves a poor relablty level (LOLP 0.01) for the RTS 24 bus system

6 Fg. 6. System Load versus Onlne Capacty Fg. 8. Loss of Energy Expectaton VII. CONCLUSIONS AND FUTURE WORK Fg. 7. Loss of Load Probablty The loss of energy expectaton (LOEE) of the four cases are shown n Fg. 8. The LOEE results exhbt a smlar pattern as the LOLP results do. By comparng the four cases, the assumpton of wnd generaton uncertanty have sgnfcant mpact on the relablty assessment of the system. The dfference among four cases can be as hgh as up to 80%. If wnd generaton s consdered completely unrelable, a lot more reserve capacty has to be procured. Therefore, t s mportant to ncorporate a more accurate stochastc model of wnd generaton. In addton, the falure rate of wnd turbnes also nfluence the relablty of the system. In the smulaton results, especally from hour 16 to hour 21, there s a sgnfcant dfference (75%) between LOLP n Case A (not consder the FOR of wnd turbnes) and n Case C (consder the FOR of wnd turbnes). Case D provdes a relatvely accurate relablty assessment for the system. It s not as optmstc as Case A whch leads to potental rsks to the system. It s also not as pessmstc as Case B whch leads to extra unnecessary reserve cost. Accordng to the LOLP and LOEE patterns n case D, the current reserve determnaton methodology cannot effectvely ensure the relablty of the system and may result n unnecessary hgh cost of reserve capacty. Ths paper suggests a novel approach to evaluatng relablty performances at operatng stage n electrc energy systems. Gven the ncreasng penetraton of uncertan varable resources such as wnd and solar, t s mportant to recognze the stronger couplng of decson makng n power systems across dfferent tme-scales. As a frst step, ths paper proposes a ntegrated approach to evaluatng system relablty performances n day-ahead unt commtment stage. In the energyreserve co-optmzaton operatng model, the valdty of usng operatng reserve as approxmaton of relablty requrement s tested n the IEEE RTS-79 system wth realstc wnd generaton profle. Relablty ndces such as Loss of Load Probablty (LOLP) and Loss of Expected Energy (LOEE) are computed at each hour of the day at the concluson of the day-ahead unt commtment decson. Case study shows that whle operatng reserve requrement does not change wthn a day, the actual relablty performance of the system vares sgnfcantly dependng on () system loadng condtons, () wnd power varaton, and () operatng rules (e.g., whether wnd generaton s allowed to provde reserve). It suggests that operatng reserve could be a very coarse determnstc approxmaton of system relablty requrement especally wth hgh wnd penetraton. Smply doublng or even trplng operatng reserves as the soluton to the ncreased penetraton of varable resources wll be nether suffcent nor affordable. Instead a much more ntegrated approach for modelng, analyss, and decson makng wll be needed for power system plannng and operatng practces. Whle ths paper llustrates the need, advances n sensng, communcaton, and computng are the techncal enables of such an ntegrated approach to power system plannng and operaton wth quantfable relablty performances at an affordable cost. A number of mportant open research questons could be pursued. Frst, onlne assessment of operatonal relablty could be developed to reflect the nearterm relablty performances n power systems wth hgh wnd penetraton. Second, tghter operatng feedback-based assessment of relablty performance n plannng needs to be

7 developed. Thrd, a well-engneered approxmaton and quantfcaton of relablty performance n operatng stage needs to be developed. Operatng reserve requrement wth hgher tme resoluton updates could be a possble canddate. Last but not least, the valuaton of flexble demand n enhancng system relablty could be formally posed and studed [26], [27]. ACKNOWLEDGEMENTS The work of the authors from Texas A&M Unversty was supported n part by Natonal Scence Foundaton ECCS Grant # , and n part by Power Systems Engneerng Research Center. The work of the author from Tsnghua Unversty was supported by the Natonal Hgh-Technology Research and Development Program (863 Program) of Chna (2011AA05A103). REFERENCES [1] L. Xe, P. M. S. Carvalho, L. A. F. M. Ferrera, J. Lu, B. H. Krogh, N. Popl, and M. D. Ilc, Wnd ntegraton n power systems: Operatonal challenges and possble solutons, Proceedngs of the IEEE, vol. 99, no. 1, pp , [2] Y. Makarov, C. Loutan, J. Ma, and P. de Mello, Operatonal mpacts of wnd generaton on calforna power systems, Power Systems, IEEE Transactons on, vol. 24, no. 2, p. 1039, [3] J. Black, Plans for wnd ntegraton n ISO-NE: Progress and challenges, n Eghth Annual Carnege Mellon Conference on the Electrcty Industry, Pttsburgh, [4] S. Wang and M. Baran, Relablty assessment of power systems wth wnd power generaton, n Power and Energy Socety General Meetng, 2010 IEEE, July 2010, pp [5] R. Bllnton and Y. Gao, Multstate wnd energy converson system models for adequacy assessment of generatng systems ncorporatng wnd energy, Energy Converson, IEEE Transactons on, vol. 23, no. 1, pp , March [6] L. Anstne, R. Burke, J. Casey, R. Holgate, R. John, and H. Stewart, Applcaton of Probablty Methods to the Determnaton of Spnnng Reserve Requrements for the Pennsylvana-New Jersey-Maryland Interconnecton, Power Apparatus and Systems, IEEE Transactons on, vol. 82, no. 68, pp , Oct [7] R. Bllnton and R. Allan, Relablty evaluaton of power systems. New York and London: Plenum Press, [8] R. Doherty and M. O Malley, A new approach to quantfy reserve demand n systems wth sgnfcant nstalled wnd capacty, Power Systems, IEEE Transacton on, vol. 20, no. 2, pp , may [9] F. Bouffard and F. D. Galana, Stochastc securty for operatons plannng wth sgnfcant wnd power generaton, Power Systems, IEEE Transactons on, vol. 23, no. 2, pp , May [10] New York ISO. (2012) NYISO Ancllary Servces Manual. [Onlne]. Avalable: publc/webdocs/documents/ manuals/operatons/ancserv.pdf [11] ERCOT. (2011) ERCOT Methodologes for Determnng Ancllary Servce Requrements. [Onlne]. Avalable: content/mktnfo/ servces/kd/2010 Methodologes for Determnng AS Requrements.pdf [12] Calforna ISO. (2012) Spnnng Reserve and Non-Spnnng Reserve - Calforna ISO. [Onlne]. Avalable: docs/2003/09/08/ pdf [13] M. D. Ilć, From herarchcal to open access electrc power systems, Proceedngs of the IEEE, vol. 95, no. 5, pp , May [14] J. Wnkelman, J. Chow, J. Allemong, and P. Kokotovc, Mult-tmescale analyss of a power system, Automatca, vol. 16, no. 1, pp , [15] F. Wu, K. Mosleh, and A. Bose, Power system control centers: past, present, and future, Proceedngs of the IEEE, vol. 93, no. 11, pp , [16] F. D. Galana, An applcaton of system dentfcaton and st.ate predcton to electrc load modelng and forecastng, Elec. Power Syst. Lab., MIT, Cambrdge, MA, [17] G. Gross and F. D. Galana, Short-term load forecastng, Proceedngs of the IEEE, vol. 75, pp , [18] J. Wang, X. Wang, and Y. Wu, Operatng reserve model n the power market, Power Systems, IEEE Transactons on, vol. 20, no. 1, pp , [19] ISO New England. (2011) ISO New England Manual for Forward Reserve. [Onlne]. Avalable: proceds/sone mnls/ [20] Chna Southern Power Grd. (2012) Chna Southern Power Grd Reserve Operaton Gude. [Onlne]. Avalable: html [21] L. Cheng, J. Ln, Y. Sun, C. Sngh, W. Gao, and X. Qn, A model for assessng the power varaton of a wnd farm consderng the outages of wnd turbnes, Sustanable Energy, IEEE Transactons on, accepted, to appear. [22] P. Sorensen, N. Cutululs, A. Vgueras-Rodrguez, L. Jensen, J. Hjerrld, M. Donovan, and H. Madsen, Power fluctuatons from large wnd farms, Power Systems, IEEE Transactons on, vol. 22, no. 3, pp , Aug [23] C. Grgg, P. Wong, P. Albrecht, R. Allan, M. Bhavaraju, R. Bllnton, Q. Chen, C. Fong, S. Haddad, S. Kuruganty, W. L, R. Mukerj, D. Patton, N. Rau, D. Reppen, A. Schneder, M. Shahdehpour, and C. Sngh, The IEEE relablty test system a report prepared by the relablty test system task force of the applcaton of probablty methods subcommttee, Power Systems, IEEE Transactons on, vol. 14, no. 3, pp , [24] ERCOT Market Informaton. Feb.2008 ERCOT Ancllary Servces Offer Selected Daly Reports. [Onlne]. Avalable: [25] ERCOT Balancng Energy Servces Daly Reports Archves Balancng Energy Servces Daly Reports. [Onlne]. Avalable: mktnfo/ servces/ bal/2008/ ndex [26] M. D. Ilc, L. Xe, and J.-Y. Joo, Effcent coordnaton of wnd power and prce-responsve demand Part I: Theoretcal foundatons, Power Systems, IEEE Transactons on, vol. 26, no. 4, pp , [27], Effcent coordnaton of wnd power and prce-responsve demand Part II: Case studes, Power Systems, IEEE Transactons on, vol. 26, no. 4, pp ,