Reliability Assessment of Microgrid with Renewable Generation and Prioritized Loads

Size: px
Start display at page:

Download "Reliability Assessment of Microgrid with Renewable Generation and Prioritized Loads"

Transcription

1 Relablty Assessment of Mcrogrd wth Renewable Generaton and Prortzed Loads Osama Aslam Ansar, Nma Safar, and C. Y. Chung Smart Grd and Renewable Energy Technology (SMART) Lab Department of Electrcal and Computer Engneerng Unversty of Saskatchewan Saskatoon, Saskatchewan, Canada {oa.ansar, n.safar, Abstract Wth the ncrease n awareness about the clmate change, there has been a tremendous shft towards utlzng renewable energy sources (RES). In ths regard, smart grd technologes have been presented to facltate hgher penetraton of RES. Mcrogrds are the key components of the smart grds. Mcrogrds allow ntegraton of varous dstrbuted energy resources (DER) such as the dstrbuted generaton (DGs) and energy storage systems (ESSs) nto the dstrbuton system and hence remove or delay the need for dstrbuton expanson. One of the crucal requrements for utltes s to ensure that the system relablty s mantaned wth the ncluson of mcrogrd topology. Therefore, ths paper evaluates the relablty of a mcrogrd contanng prortzed loads and dstrbuted RES through a hybrd analytcal-smulaton method. The stochastcty of RES ntroduces complexty to the relablty evaluaton. The method takes nto account the varablty of RES through Monte- Carlo state samplng smulaton. The results ndcate the relablty enhancement of the overall system n the presence of the mcrogrd topology. In partcular, the hghest prorty load has the largest mprovement n the relablty ndces. Furthermore, senstvty analyss s performed to understand the effects of the falure of mcrogrd slandng n the case of a fault n the upstream network. Index Terms Relablty evaluaton, mcrogrds, Monte- Carlo, renewable, dstrbuted generaton I. INTRODUCTION For the last few decades, there has been an ncreasng concern about the clmate change and depleton of non-renewable energy sources. Increased publc awareness has called for reducton of carbon emssons whch are one of the man sources of clmate change. Ths requres gradual yet steady replacement of conventonal coal-based power plants wth envronmental-frendly renewable energy sources (RES). Smart grd technologes have been developed to facltate large-scale ntegraton of RES and to tackle dfferent challenges assocated wth t. Mcrogrds are one of the man buldng blocks of smart grds [1]. Mcrogrds do not only offer the ntegraton of RES n the dstrbuton system but also provde attractve features such as slanded mode of operaton and hgher flexblty. In order to fully understand the costbeneft analyss of dstrbuton system contanng mcrogrds, t s mportant to take nto account the relablty of the electrc supply at the customers ends. Mantanng system relablty s one of the prmary motves of any utlty. Sgnfcant costs and penaltes are ncurred as a result of system nterruptons. Hence not only the company s reputaton but also the fnancal reasons force utltes to ensure an acceptable level of system relablty. In ths regard, several standards have been developed to make sure that utltes guarantee the relable supply of electrcty to ther customers. In such scenaros, relablty studes become crucal for utltes. Relablty studes can also be consdered n dstrbuted energy resources (DER) szng, stng and operaton, and renforcement of the crucal elements n dstrbuton network [2], [3], [4]. The ncluson of RES ntroduces complexty n ther modelng for relablty studes. Furthermore, the output of dstrbuted generaton (DGs) based on RES are energy lmted and sporadc n nature. In ths case, the prorty order of the loads should be consdered to mantan electrc supply to the most senstve loads n the event of nsuffcent generaton. In lterature, several technques have been presented to evaluate the relablty at customer load ponts n the mcrogrd and of the mcrogrd as a whole. They can be categorzed nto analytcal methods, smulaton methods and hybrd methods. In [5], a method based on Monte-Carlo smulaton to evaluate relablty of an actve dstrbuton system wth multple mcrogrds s proposed. It has been shown that the ncluson of DGs and storage ncreases the overall relablty of the system. The method dscretzes the output of RES and evaluates probablty for each step. In [6] and [7], the relablty of a mcrogrd n slanded mode s evaluated. In [6], Monte-Carlo smulaton s used to model component falure and component repar and hstorcal data for RES. In [7], fault tree analyss s adopted whch can become mathematcally nvolved for large systems. Snce grd-connected mode of mcrogrd s not taken nto account n both of the papers, the values for the relablty metrcs are dfferent from the actual values. Markov modelng s used to evaluate the relablty of mcrogrd wth photovoltac (PV) generaton and energy storage systems (ESSs) n [8]. However the proposed method uses a smple two-state model for PV generaton whch s nsuffcent to ncorporate the hghly ntermttent nature of PV and other RES such as wnd energy. In [9], the ntermttent nature of PV /16/$ IEEE

2 based DGs s gnored by consderng that the ESSs are suffcently szed to make DGs dspatchable. The values of relablty ndces can be markedly dfferent f ntermttency s taken nto account. Some papers such as [10] evaluated the relablty ndces of a small solated power systems wth RES. However, they do not take nto account the abltes of a mcrogrd such as ts dfferent modes of operaton.. In [11], a hybrd model s presented to fnd out the relablty of a mcrogrd n the presence of renewable DGs and ESSs. The method does not consder the prorty order of the loads. Also n one of the cases, the method assumes that battery helps ntermttent DGs to supply the entre load or a large part of the load. Ths paper presents a hybrd method comprsng of both analytcal and smulaton technques to evaluate the relablty at customers load ponts n a mcrogrd contanng wnd and solar power generaton and prortzed loads. State samplng Monte-Carlo smulaton s combned wth the analytcal method. Smulaton methods take nto account the chronology of varaton of RES and the prorty order of the loads whereas analytcal methods consder the topology of the network and the falure and repar rates of the networks components. The paper s formatted as follows. Secton II brefly ntroduces the mcrogrd. Secton III presents the modelng of wnd and PV generaton sources for the relablty evaluaton. Secton IV delves nto the proposed method for calculatng the relablty ndces of the mcrogrd. The results are ndcated n Secton V. Secton VI provdes the concluson of the work. II. MICROGRIDS A mcrogrd s a group of nterconnected loads, DERs (such as DGs and ESSs), and management systems that can operate ether n grd connected mode or slanded mode (Fg. 1). From the utlty s pont of vew, a mcrogrd appears as a sngle controllable entty that can consume or supply power dependng upon the total generaton and the total load nsde the mcrogrd. The ablty of a mcrogrd to operate n slanded mode ncreases the relablty of the load ponts. The slandng mode can occur f there s a fault wthn the mcrogrd or n the upstream network to whch t s connected. In slanded mode of operaton, the loads depend upon the power generaton of DGs connected n the mcrogrd. In the case when DGs are unable to supply all of the load, energy management systems can voluntarly curtal the non-senstve loads utlzng advanced swtches and control strateges. Ths ensures that the senstve loads are not nterrupted [1], [4]. III. MODELING RENEWABLE ENERGY SOURCES One of the key features of the mcrogrd topology s ts ablty to ntegrate DERs such as renewable DGs, dspatchable DGs, ESSs and electrc vehcles (EVs) n a seamless manner. DERs not only provde an alternatve to the generaton expanson plannng but also assst n mprovng the relablty of the dstrbuton system to whch they are connected [4], [12], [13]. Nevertheless, RES poses dfferent challenges to the operaton and plannng of the power system. The most sgnfcant of these s ther ntermttent behavor. The output of RES depends on varous factors ncludng weather condtons and geographcal locaton. Moreover, RES have wdely dfferent characterstcs based on ther source of energy, for nstance, wnd and solar. Hence t s dffcult to use a general model for all RES or to use the same models as utlzed for conventonal generators. RES generators also have ther own falure rates and repar tme. It has been observed that the unavalablty of RES s prmarly because of the unavalablty of ther energy source rather than the falure of RES generators [14]. The method presented n ths work uses the above observaton. In relablty studes nvolvng smulaton technques, t s necessary to generate synthetc data for RES power generaton. The next parts of ths secton ntroduces the models to generate the data for ths purpose. A. Wnd Power Generaton Modelng Wnd energy s the most developed and the fastest growng form of renewable energy. Wnd power depends upon dfferent factors ncludng wnd speed, wnd drecton, and geographcal locatons etc. However, wnd power s largely dependent on the wnd speed. Therefore, n ths paper, the wnd power s modeled usng the wnd speed. Dfferent dstrbutons such as normal [15] and Webull [16] are used to model the wnd speed. In general, two parameter Webull dstrbuton provdes a better ft for modelng the wnd speed [17]. In ths paper, the model for wnd speed dstrbuton functon s obtaned from [16] whch uses Webull dstrbuton. The cumulatve probablty dstrbuton for the wnd speed F v s gven as: k ( vc / ) F () v 1 e (1) v, () w Fg. 1. Typcal structure of a mcrogrd w where, c s the scale parameter and k s the shape parameter. The values for two parameters are obtaned usng hstorcal data. They are adopted from [16] and gven n Table I. In order to smulate the wnd speed values for a requred duraton, the nverse transformaton method s mplemented [18]. Inverse transformaton method states that f a random varable (n ths case v) follows the U [0, 1] unform

3 TABLE I WEIBULL DISTRIBUTION DATA Parameter Regon 1 Regon 2 Mean speed (m/s) k c dstrbuton, then the random varable X = F -1 (U) has a contnuous cumulatve probablty dstrbuton functon F(X). Usng ths prncple, (1) can be expressed as: 1/ vc[ ln(1 X)] k (2) where X s a random number between 0 and 1. Snce X s a random number, (2) can be also be expressed as: 1/ vc[ ln( X)] k (3) Usng (3), daly values of wnd speed can be smulated by generatng random number X for each day. The wnd power s obtaned from the wnd speed by usng the WTG power curve shown n Fg. 2. The power curve expresses the wnd power as the functon of wnd speed and s provded by the wnd turbne manufacturer. The power curve of WTG s usually characterzed by the followng parameters: Rated speed (v rated ): The speed at whch the maxmum power can be extracted from the WTG. Rated power (P rated ): The maxmum power that can be produced by WTG at the rated wnd speed. Cut-n speed (v cut-n ): The mnmum wnd speed requred to produce the power from WTG. Cut-out speed (v cut-out ): The maxmum speed at whch WTG can operate. Usually after ths wnd speed, the WTG s shut down due to safety reasons. The wnd power curve s gven as: 0 0vvcut-n 3 av bprated vcut-n vvrated PWTG (4) Prated vrated vvcut-out 0 vcut-out v where, a and b are gven as: Prated a 3 3 vrated vcut-n (5) 3 vcut-n b 3 3 v v rated cut-n Usng the wnd power curve, the smulated wnd power s obtaned from the wnd speed. The smulated wnd speed and wnd power for one of the two WTGs s shown n Fg. 3. Table II provdes the data for the two wnd turbnes. B. Solar Power Generaton Modelng Wth the development of new solar panels, the penetraton of solar power s ncreasng and further growth s expected n TABLE II WIND TURBINES DATA Parameter Regon 1 Regon 2 Rated Power (kw) Rated speed (m/s) Cut-n speed (m/s) 3 3 Cut-out speed (m/s) the near future. Solar power of the solar panels manly depends upon the solar radaton and temperature. The temperature dependency of solar power s non-lnear and ntroduces complexty to the modelng. In ths work, solar radaton s utlzed to obtan the output power of solar panels. Hstorcal data of solar radaton s used to fnd out the probablty dstrbuton of solar radaton. The hstorcal data for 5 years s obtaned from [19]. It has been shown that the beta dstrbuton fts more accurately to solar radaton as compared to gamma and logarthmc dstrbuton [20]. Therefore n ths paper, beta dstrbuton s ftted on the hstorcal data of solar radaton. The probablty dstrbuton functon of beta dstrbuton s gven as: 1 1 x (1 x) f( x) (6) B(, ) where, B n terms of gamma functon (Γ) s defned as: ( ) ( ) B(, )= (7) ( ) After fttng the beta dstrbuton on hstorcal data, the two parameters for beta dstrbuton obtaned from statstcs and machne learnng toolbox of MATLAB are as follows: (8) After obtanng the correspondng probablty dstrbuton, nverse transformaton method s appled. Usng the nverse transformaton method, solar radaton s predcted for the requred nterval of tme. Smlar to wnd power curve, the solar power curve expresses solar power n terms of solar radaton. The expresson for output power of PV n terms of solar radaton s gven as [14]: 2 G Psn 0 G Rc GstdRc G PPV Psn Rc G Gstd (9) Gstd Fg. 2. Wnd power curve. P G G sn std

4 Fg. 3. Smulated wnd speed and wnd power for WT A where, P PV s solar output power n MW. G s smulated solar radaton n W/m 2. G std s solar radaton n the standard envronment. Usually ths value s set to 1000 W/m 2. R c s a certan radaton pont set usually to 150 W/m 2. P sn s the rated output power of solar panels. IV. RELIABILITY EVALUATION METHOD A. Relablty Indces For evaluatng the relablty of customer load ponts n a dstrbuton system, the system average nterrupton frequency ndex (SAIFI), the system average nterrupton duraton ndex (SAIDI), the customer average nterrupton duraton ndex (CAIDI), the energy not suppled (ENS) and average energy not suppled (AENS) ndces are used. These are defned as [21]: N SAIFI (10) NT UN SAIDI (11) N T SAIDI CAIDI SAIFI (12) ENS UL (13) LP AENS = ENS N T (14) where, s the falure rate, U s the outage tme, N s the number of customers, and L s the total load at load pont. N T s the total number of customers n the system. B. Evaluaton Method The steps for relablty evaluaton are as follows: Step 1. Obtan the cumulatve dstrbuton functons of wnd and solar radaton from ther hstorcal data. Step 2. Use cumulatve dstrbuton functons, and the nverse transformaton method to generate the smulated values of wnd speed and solar radaton for requred number of samples. Ths requres generaton of random number X for each sample. Here one sample can be consdered as one day. Step 3. Convert the wnd speed to wnd power usng wnd power curve (4) (5) and solar radaton to solar power usng (9). Step 4. At each day, sample the output of all WTGs and PV panels. The combned output of all RES s compared wth the load. In the case f RES cannot supply all of the loads, then the load wth hghest prorty s suppled frst followed by the next load n the prorty lst and so on. Step 5. Step 4 s performed for each day of the year. The number of occurrences at whch the load s suppled by RES for each of the load ponts s counted. At the end of year, the probablty of RES supplyng dfferent load ponts s calculated. Let P RES be the probablty that RES can supply the load at load pont. Ths probablty value along wth the system stochastc data (falure rate and repar tme) are used to obtan the average falure rate, unavalablty and repar tme at each load pont at the end of the year usng the followng equaton: j (1 PRES) up (15) U jrj (1 PRES) uprup where, and U are the falure rate and unavalablty at load pont, respectvely, up and r up are the falure rate and repar tme of upstream network, respectvely, and j and r j are the falure rate and repar tme of mcrogrd network components, respectvely that results n nterrupton at load pont. Step 6. Afterwards, the system ndces SAIFI, SAIDI, CAIDI, ENS and AENS are calculated usng (10) (14). Step 7. Steps 4-6 are repeated for each year untl the varaton n system ndces s less than the specfed tolerance or maxmum number of teratons s reached. In (15), each expresson conssts of two terms. The frst term arses from the stochastc data of the network components. The second term s derved from the followng condtonal probablty equaton. ( RES can supply) PRES (16) +( RES cannot supply) (1 PRES ) In the frst part of above expresson, n the case of falure n the upstream network, f RES can supply the load pont, the falure n the upstream network wll not result n any nterrupton at that load pont. Hence, the (16) reduces to: (1 P RES) up (17) The P RES calculated through smulaton takes nto account the ntermttency and varablty of RES. Durng a sample, f RES generaton s greater than the load, the excess generaton would be curtaled. V. CASE STUDY AND RESULTS A. Test System The proposed method s appled to a modfed verson of the dstrbuton system connected to bus 5 of the Roy Bllnton

5 Load Pont TABLE III LOAD DATA Load Level Type of Load (kw) No. of Customers LP Commercal 100 LP Offce loads 300 LP House loads 250 LP9 500 Governmental 50 Test System (RBTS) [21]. The advantages of usng the RBTS nclude the avalablty of stochastc data for network components, and manageable system sze. The test system s shown n Fg. 4. Table III provdes the assumed load data for the system. For ths system, n the case of a fault n a feeder secton, assume 4 for an nstance, the management system acts so as to reduce nterrupton duraton for the customers. In ths case, frstly F1 would open and then solatng swtches connected to secton 4 would open to solate the faulty secton from the rest of the network. Then F1 and normally open (N/O) swtch would close to restore the supply to rest of the customers. In ths stuaton, LP3 and LP4 would be nterrupted untl secton 4 s repared completely whereas, the other load ponts would be nterrupted for the duraton equal to the swtchng tme of F1, N/O and solatng swtches. As mentoned earler, a prorty lst s constructed to serve the most senstve load frst n the case of slanded mode of operaton resultng from falure n the upstream network. In ths case t s assumed that government load (LP9) has the hghest prorty followed by offce loads (LP3) and then house loads (LP4). On the other hand, commercal loads (LP2) are assumed to have the lowest prorty. Followng four cases are studed to consder the effects of DGs on relablty of mcrogrd. Case 1: Wthout DGs. Case 2: Wth 4 WTGs as DGs. Case 3: Wth 2 WTGs and 2 PV panels as DGs. Case 4: Case 3 wth varable load. The ratngs for the WTG and PV panels are gven n table IV. These ratngs are selected consderng that the capacty factors of DGs based on RES are qute low. Hence the combned rated power of all DGs s hgher than the total load n the mcrogrd. The falure rates and repar tmes of the test system are gven n [21]. The falure rate and the repar tme of upstream network are assumed to be 0.5 f/yr and 10 hours, respectvely. The swtchng tme and repar tme are assumed to be 3.5 hours and 30 hours, respectvely. These assumed values are frequently used n the lterature. The maxmum number of samples s set to 100,000. The varable load model s obtaned from [21]. TABLE IV RENEWABLE DGS DATA Type Locaton Rated Power (kw) WTG 1 LP WTG 2 LP PV 1 LP PV 2 LP B. Results The applcaton of the method on four cases ndcates that the presence of DGs n the mcrogrd ncreases the overall relablty of the system. Although the values of SAIFI do not change sgnfcantly, there s a notceable change n SAIDI and ENS values. An mprovement of 20% s observed n the values of ENS from case 1 to case 4. Furthermore, the mprovement s expected to be hgher for larger systems. The results for each of the load ponts are shown n Table V. Table VI shows the system ndces for all four cases from whch t should be nferred that the overall system relablty has mproved. The decrease n unavalablty and falure rate for the hghest prorty load (LP9) s sgnfcant as compared to the other load ponts. Fg. 5 shows pctorally the mprovement n the relablty at LP9. The results also ndcate that the relablty ndces for the most non-senstve load does not change sgnfcantly. C. Senstvty Analyss The successful slanded operaton of the mcrogrd n the case of upstream falure depends upon successful operaton of the solatng swtch. Ths solaton operaton usually has a hgh probablty of success. In prevous scenaros, ths probablty was taken to be unty.e. the slandng operaton s always successful. A senstvty analyss s performed to observe the effects of the probablty of swtchng on system ndces. Scenaro 3 s consdered agan and the probablty of successful solaton s vared from 100% to 0%. The results for all four cases are shown n Table VI. As expected, the results ndcate TABLE V RELIABILITY INDICES FOR ALL LOAD POINTS LP Index Case 1 Case 2 Case 3 Case 4 LP2 LP3 LP4 LP9 Fg. 4. Modfed Bus 5 of RBTS [21] λ (f/yr) r (hr) U (hr/yr) λ (f/yr) r (hr) U (hr/yr) λ (f/yr) r (hr) U (hr/yr) λ (f/yr) r (hr) U (hr/yr)

6 Case TABLE VI RELIABILITY INDICES FOR THE MICROGRID SAIFI (nt/yr SAIDI (hr/ yr CAIDI (hr/nt that the decrease n probablty of successfully slandng, reduces the relablty of the system. There s a decrease of 9.72% n the value of SAIDI f probablty of successful slandng decreases from 1 to 0. VI. CONCLUSION ENS (kwh/yr) AENS (kwh/yr Case Case Case Case In ths work, relablty evaluaton of a mcrogrd contanng wnd and solar energy sources was performed. The stochastc nature of wnd and solar energy ntroduces complexty n the relablty evaluaton methods. Ths stochastc nature s dealt through state-samplng smulaton whereas, analytcal methods are appled to evaluate the relablty metrcs. The prorty order of the loads s also taken nto account through smulaton. The studes ndcated that ntegraton of ntermttent wnd and solar energy sources n mcrogrd ncrease the relablty ndces of the system. In partcular, the most senstve load has the largest ncrease n ts relablty. It was also hghlghted that the relablty of the system decreases wth decrease n relablty of the slandng operaton. For future studes, dspatchable DGs and ESSs can be ncluded n the system to understand ther effects on the relablty ndces of the system. ESSs n partcular, can affect the relablty of the system as they can store excess renewable energy durng low-demand perods and can supply durng hgh-demand perods. For a more accurate evaluaton, hourly varatons of RES generaton can be utlzed. Moreover, relablty cost/worth analyss can be performed. Probablty of slandng success Fg. 5. Relablty ndces for load ponts 9 for dfferent cases. SAIFI (nt/yr TABLE VII SENSITIVITY ANALYSIS SAIDI (hr/ yr CAIDI (hr/nt ENS (kwh/yr) AENS (kwh/ yr ACKNOWLEDGMENT Ths work was supported by the Natural Scences and Engneerng Research Councl (NSERC) of Canada. REFERENCES [1] N. Hatzargyrou, H. Asano, R. Iravan and C. Marnay, "Mcrogrds: An overvew of ongong research, development, and demonstraton projects," IEEE Power and Energy Mag., pp , Aug [2] S. A. Areffar and Y. A. I. Mohamed, "DG mx, reactve sources and energy storage unts for optmzng mcrogrd relablty and supply securty," IEEE Trans. Smart Grd, vol. 5, no. 4, pp , Jul [3] S. Bahramrad, W. Reder and A. Khodae, "Relablty-constraned optmal szng of energy storage system n a mcrogrd," IEEE Trans. Smart Grd, vol. 3, no. 4, pp , Dec [4] N. Z. Xu and C. Y. Chung, "Relablty evaluaton of dstrbuton systems ncludng vehcle-to-home and vehcle-to-grd," IEEE Trans. Power Syst., vol. 31, no. 1, pp , Jan [5] Z. Be, P. Zhang, G. L et al., "Relablty evaluaton of actve dstrbuton systems ncludng mcrogrds," IEEE Trans. Power Syst., vol. 27, no. 4, pp , Nov [6] S. Kennedy and M. M. Marden, "Relablty of slanded mcrogrds wth stochastc generaton and prortzed load," n 2009 IEEE Bucharest PowerTech, pp. 1-7, Bucharest, Jun.-Jul [7] Z. L, Y. Yuan and F. L, "Evaluatng the relablty of slanded mcrogrd n an emergency mode," n th Int. Unverstes Power Eng. Conf. (UPEC), pp. 1-5, Cardff, Aug.-Sep [8] T. Tuffaha and M. AlMuhan, "Relablty assessment of a mcrogrd dstrbuton system wth pv and storage," n 2015 Int. Symp. Smart Elec. Dstr. Syst. Technol., pp , Venna, Sep [9] I. S. Bae and J. O. Km, "Relablty evaluaton of customers n a mcrogrd," IEEE Trans. Power Syst., vol. 23, no. 3, pp , Aug [10] R. Kark, and R. Bllnton, "Relablty/cost mplcatons of PV and wnd energy utlzaton n small solated power systems," IEEE Trans. Energy Convers., vol. 16, no. 4, pp , Dec [11] C. L. T. Borges, and M. Costa, "Relablty assessment of mcrogrds wth renewable generaton by an hybrd model," n 2015 IEEE Endhoven PowerTech, pp. 1-6, Endhoven, Jun.-Jul [12] P. M. Costa and M. A. Matos, "Relablty of dstrbuton networks wth mcrogrds," n 2005 IEEE Russa Power Tech, pp. 1-7, Jun [13] M. Fotuh-Fruzabad, and A. Rajab-Ghahnave, "An analytcal method to consder DG mpacts on dstrbuton system relablty", n 2005 IEEE/PES Transmsson & Dstrbuton Conf. & Expo.: Asa & Pacfc, pp. 1-6, Dalan, [14] J. Park, W. Lang, A. A. El-Keb et al., "A probablstc relablty evaluaton of a power system ncludng solar/photovoltac cell generator," n 2009 IEEE Power and Energy Socety General Meetng, pp. 1-6, Calgary, Jul [15] R. Kark, P. Hu and R. Bllnton, "A smplfed wnd power generaton model for relablty evaluaton," IEEE Trans. Energy Convers., vol. 21, no. 2, pp , Jun [16] A. P. Lete, C. L. T. Borges and D. M. Falcao, "Probablstc wnd farms generaton model for relablty studes appled to Brazlan stes," IEEE Trans. Power Syst., vol. 21, no. 4, pp , Nov [17] T. P. Chang, Estmaton of wnd energy potental usng dfferent probablty densty functons, Appled Energy, Elsever, vol. 88, no. 5, pp , May [18] R. Bllnton and R. N. Allan, Relablty Evaluaton of Power Systems, 2nd ed. New York, NY, USA: Plenum, [19] NASA Atmospherc Scence Data Center [Onlne]. [20] Z. Qn, W. L, and X. Xong, "Incorporatng multple correlatons among wnd speeds, photovoltac powers and bus loads n composte system relablty evaluaton," Appled Energy, Elsever, vol. 110, pp , Oct [21] R. Bllnton and S. Jonnavthula, "A test system for teachng overall power system relablty assessment," IEEE Trans. Power Syst., vol. 11, no. 4, pp , Nov