Electrical Power and Energy Systems 31 (2009) 562–569

Contents lists available at ScienceDirect

Electrical Power and Energy Systems

journal homepage: www.elsevier .com/locate / i jepes

Probabilistic evaluation of reserve requirements of generating systemswith renewable power sources: The Portuguese and Spanish cases

Manuel Matos a,*, João Peças Lopes a, Mauro Rosa a, Ricardo Ferreira a, Armando Leite da Silva b

Warlley Sales b, Leonidas Resende b, Luiz Manso b, Pedro Cabral c, Marco Ferreira d, Nuno Martins c,Carlos Artaiz e, Fernando Soto e, Rubén López e

a INESC Porto, Fac. Engenharia Univ. do Porto., Campus da FEUP, Rua Dr. Roberto Frias, 378, 4200-465 Porto, Portugalb Federal University of Itajubá, UNIFEI, Brazilc Rede Eléctrica Nacional, REN, Portugald REN Serviços, S.A, Divisão de Planeamento de Longo Prazo, Rua Sá da Bandeira, 567-4�, 4000-437 Porto, Portugale Red Eléctrica de España, REE, Spain

a r t i c l e i n f o a b s t r a c t

Article history:Received 30 September 2008Received in revised form 25 March 2009Accepted 29 March 2009

Keywords:Operating reserveMonte Carlo simulationRenewable energy sources

0142-0615/$ - see front matter � 2009 Elsevier Ltd. Adoi:10.1016/j.ijepes.2009.03.031

* Corresponding author. Tel.: +351 222094230; faxE-mail address: mmatos@inescporto.pt (M. Matos

This paper presents an application of probabilistic methodologies to evaluate the reserve requirements ofgenerating systems with a large penetration of renewable energy sources. The idea is to investigate thebehavior of reliability indices, including those from the well-being analysis, when the major portion ofthe renewable sources comes from wind power and other intermittent sources. A new simulation processto address operating reserve adequacy is introduced, and the correspondent reliability indices areobserved. Case studies on the Portuguese and Spanish generating systems are presented and discussed.

� 2009 Elsevier Ltd. All rights reserved.

1. Introduction

The increased use of electricity produced from renewable en-ergy sources constitutes an important part of the package of mea-sures needed to comply with the Kyoto Protocol to the UnitedNations Framework Convention on Climate Change. Renewable en-ergy source is defined as any energy resource naturally regener-ated over a short time scale that is derived directly from the sun(such as solar thermal and photovoltaic), indirectly from the sun(such as wind, hydropower and photosynthetic energy stored inbiomass), or from other natural movements and mechanisms ofthe environment (such as geothermal and ocean energy) [1].Renewable energies can play a major role in tackling the twin chal-lenge of energy security and global warming because they are notdepletable and produce less greenhouse-gas emissions than fossilfuels. The promotion of this type of electricity has a high priorityin many countries, in particular, in the European Union.

In the last few years, several discussions among European gov-ernments and associations have been carried, and in March 2007,Europe’s Heads of States agreed to a binding target of 20% renew-able energy by 2020 [2,3]. This decision gives a strong signal for

ll rights reserved.

: +351 222094050.).

Europe’s future energy policy as well as for the further expansionof the European renewable energy industry, which includes thedevelopment of new ways of operating and planning powersystems.

While contributions from renewable energy sources for elec-tricity production is small, with the exception of hydro, their mar-ket penetration is growing at a much faster rate than any otherconventional source. Although there are still many potential hydrosites in the world, severe restriction based mainly on environmen-tal aspects have limited their exploitation.

Wind has become a popular source of green electricity aroundthe world [4]. At the end of 2005, the worldwide capacity ofwind-powered generators was 59 GW; in Spain, about 10 GW ofinstalled capacity has produced more than 8% of its total energyneeds, and in Portugal, about 1 GW of installed capacity has pro-duced almost 4% of its needs. At the end of 2006, the global capac-ity of wind-powered generators reached almost 74 GW; Spain andPortugal have the second and ninth largest installed capacity in theworld.

The previous values put system operators and planners under ahuge pressure to come up with solutions bearing in mind thesenew technologies. The main reason is that the number of randomvariables and system complexities greatly increase, when renew-able energy sources are added to the system, due to the fluctuating

M. Matos et al. / Electrical Power and Energy Systems 31 (2009) 562–569 563

capacity levels of these sources. New computational models andtools have to be developed to deal with these new variables, partic-ularly those related with wind power. A huge number of technicalworks have recently been published in this area: see, for instance[5–12].

From the planning point of view, deterministic based ap-proaches have very attractive characteristics such as simple imple-mentation, easy understanding, assessment and judgment byplanners in relation to severe conditions like network outagesand system peak load. Unfortunately, the perception of many plan-ning engineers that past experience in addition to some knowncritical situations is enough to assess system risk conditions isnot valid. In addition, past experience with renewable sources likewind power is very limited. However, the principles of some deter-ministic standards (e.g. ‘‘N � 1” criterion) must be recognized asattractive.

Conversely, methodologies based on probability concepts canbe extremely useful in assessing the performance of power sys-tems [13–15]. They have been successfully applied to many areasincluding generation and transmission capacity planning, operat-ing reserve assessment, distribution systems, etc. The proper mea-sure of risk can only be achieved by recognizing the probabilisticnature of power system parameters.

A new framework, named system well-being analysis [16–18],has been built combining the deterministic perception with prob-ability concepts. This new framework reduces the gap betweendeterministic and probabilistic approaches by providing the abilityto measure the degree of success of any operating system state. In awell-being analysis, success states are further split into healthy andmarginal states, using the previously mentioned engineers’ percep-tion as the criterion. Well-being analysis has been applied in thelast decade to areas such as generating systems, operating reserveassessment, and composite generation and transmission systems.Chronological or sequential Monte Carlo simulation has been usedfor generating system well-being analysis, considering the loss ofthe largest available unit in the system as the deterministic crite-rion. These concepts can be extremely useful for dimensioningthe reserve capacities considering renewable resources [19–26].

This paper presents an application of chronological Monte Carlosimulation (MCS) to evaluate the reserve requirements of generat-ing systems, considering renewable energy sources. The idea is tostudy the behavior of reliability indices (conventional and well-being), when a major portion of the energy sources is renewable.Renewable comprises mainly hydro, wind, mini-hydro powersources, although other sources such as solar are present in muchlesser amounts.

Case studies with the Portuguese and Spanish generating sys-tems are presented and discussed. The work was developed inthe framework of a RTD project financed by the TSO of Portugaland Spain (REN and REE, respectively) within their activities re-lated to MIBEL (the Iberian electricity market), with the aim ofmaximizing the integration of renewable energy.

2. Proposed methodology

The assessment of capacity reserve requirements to ensure anadequate level of energy supply is an important aspect for bothexpansion and operation planning of generating systems. Thescheduling of the static capacity (i.e. adequacy) is a well knownactivity in generating expansion planning. However, if the amountof renewable energy sources being considered in the future is sig-nificant, the evaluation of operating reserve becomes also relevantin generating expansion studies.

The requirements for both reserves, static and operating, can beprobabilistically assessed based on two distinct representations:

state space and chronological modeling. State enumeration andnon-sequential MCS methods are examples of state space basedalgorithms, where Markov models are usually used for both equip-ment and load state transitions. Therefore, states are selected andevaluated without considering any time connection. Conversely,sequential simulation can perceive all chronological aspects and,hence, it is able to correctly represent equipment aging process,time varying loads, spatial and time correlation aspects, etc. Chro-nological MCS is very suitable due to its flexibility, as it allows therepresentation of non-exponential residence times, useful whendealing with chronological processes. Moreover, as dealing withrenewable energy sources and their natural uncertainties, due tohydrologic inflow sequences, wind speed variations etc., chrono-logical MCS seems to be the most effective way to adequately mod-el and solve these difficulties.

2.1. Chronological Monte Carlo simulation

The operation history of system states, for a simulation period T,is based on stochastic models of the components and on the loadmodel. The initial operating state is sampled from the probabilitydistributions of the generating equipment. After evaluating eachstate, performance indices are estimated using test-functions G(t):

EðGÞ ¼ 1T

Z T

0GðtÞdt: ð1Þ

Each performance index can be estimated using a suitable test-function. The failure probability, for instance, corresponds to the ex-pected value of an indicator function where G(t) = 1 if the systemassociated with time t is a failure state; otherwise G(t) = 0. Anotherway of estimating the expected value of G(t) is shown as follows:

eEðGÞ ¼ 1NY

XNY

k¼1

GðykÞ ð2Þ

where NY is the number of simulated years and yk is a sequence ofsystem states in year k. For instance, the energy not supplied will bethe summation of unsupplied energy associated with each interrup-tion of a simulated year. The uncertainty around the estimated indi-ces is given by the variance of the estimator:

VðeEðGÞÞ ¼ VðGÞNY

ð3Þ

where V(G) is the variance of the test-function. The convergence ofthe simulation process is tested using the coefficient of variation b[27].

b ¼ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiVðeE½G�Þq

=eE½G� ð4Þ

2.1.1. Modeling of thermal unitsA two-state Markov model is used for modeling the up/down

cycle of all thermal generating units. They are specified throughtheir failure (k) and repair (l) rates. Clearly, any non-Markovianmodel could be used if the necessary parameters are available.Fig. 1a shows the well known two-state Markov model [13]. Thegenerating capacity of the thermal units are fixed and pre-specified.

2.1.2. Modeling of hydroelectric unitsThe capacities of the hydro units will be defined for each month,

according to the corresponding hydrological series. These series aredefined for each hydraulic basin based on historical data and aim atcapturing the historical inflows, reservoir volumes and type ofoperation. Mathematical polynomials convert the monthly storagevolume of each reservoir into available power capacity for the

Fig. 1. (a) Two-state and (b) multi-states Markov models.

564 M. Matos et al. / Electrical Power and Energy Systems 31 (2009) 562–569

month. In the case of hydro power plants with pumpingcapacity, some additional evaluations are carried out. Historicalyearly series of volumes per power plant and per month have tobe provided.

2.1.3. Modeling of wind unitsUsually, in a wind power site, there are several generating units

and they will be grouped into an equivalent multi-state Markovmodel, as shown in Fig. 1b. Only two stochastic parameters arenecessary: unit failure and repair rates. Parameter N representsthe number of generating units of the wind farm. If C is the unitcapacity, the amount of power associated with the kth state is gi-ven by Ck = (N � k) � C, k = 0, . . . , N. The cumulative probability Pk

(from 0 to k) associated with this state can be easily calculated. Inorder to reduce the number of these states during the chronolog-ical MCS, a simple truncation process sets the desired order ofaccuracy. Therefore, instead of N + 1 states, a much smaller num-ber up to the capacity CL will limit this model; e.g. 1 � PL 6

tolerance.The productions of the wind generating units will be defined for

each hour, according to the hourly wind series for each geographicregion. The wind series try to capture the wind speed and powerconversion characteristics. Historical yearly series of per unitcapacity fluctuations per hour have to be provided.

2.1.4. Modeling of mini-hydroMini-hydro units are modeled similarly to the hydro generating

units from the hydrological point of view, but they are groupedinto multi-state units of Fig. 1b to simplify the modeling process-ing. Due to the lack of specific data in relation to the hydrologicalbasin where they are located, an equivalent reservoir is used tomodel the capacity variations with time. Obviously, if specific dataare available, they can be properly considered.

One has always to balance the benefits in terms of accuracy andthe cost in terms of computing effort and levels of model detail.

2.1.5. Modeling of co-generationCo-generating units are modeled similarly to the thermal units,

but like in the previous case, they are clustered as well by usingmulti-state units of Fig. 1b.

Moreover, an hourly utilization factor is specified, which mod-els the actual co-generation power used by the system. This factorvaries during the year following the tariff attractiveness and/or theindustry production cycle.

2.1.6. Maintenance aspectsA certain amount of power generation will be specified per

month in order to capture the maintenance activities along theyear. In order to deal with that, the proposed chronological MCSalgorithm, according to the generating power on maintenance,adequately increases the hourly load curve. This simplified modelis particularly useful for planning purposes, since it is difficult toaccurately specify the exact period for generating unit mainte-nance activities. For some nuclear power generators, due to itspeculiarities, it is possible to detail this period and the chronolog-ical simulation should account for that.

2.1.7. Load characteristicsA standard chronological load model containing 8760 levels,

corresponding to each hour, is used. The chronological MCS willsequentially follow these load steps during the simulation process.Two uncertainty levels, representing short and long-term loadforecasting deviations can be simulated through the MCS process.Gaussian or any other distribution can be used.

2.2. Conventional reliability indices

The conventional reliability indices are: LOLP = loss of loadprobability; LOLE = loss of load expectation; EPNS = expectedpower not supplied; EENS = expected energy not supplied;LOLF = loss of load frequency; LOLD = loss of load duration;LOLC = loss of load cost [13,22,26,27].

2.3. Well-being indices

The well-being indices are [16–18,26]: EH = expected healthyhours, which is the expected number of hours in a period (e.g. year)the system will stay in healthy states; EM = expected marginalhours, which is the expected number of hours in a period (e.g. year)the system will stay in marginal states; FH and FM = expected fre-quency associated with healthy and marginal states, respectively;DH and DM = expected duration of system residing in healthy andmarginal states, respectively. The deterministic criterion used todifferentiate between healthy and marginal states may be thespecified value for the secondary reserve, but the loss of the largestavailable unit in the system can also be used.

2.4. Operating reserve assessment

All previous risk indices are based on the following power bal-ance equation:

G� L < 0 ð5Þ

where G represents the system available generation and L is the to-tal system load. The random variable G depends on the equipmentavailabilities and on the capacity fluctuations due to, for instance,hydrology and wind variations, etc. The random variable L dependson the short and long-term uncertainties and also on the hourlyvariations.

In order to assess the performance of the operating reserve, newvariables have to be defined as shown in Fig. 2. In this work, theprimary (or regulation) and secondary (or spinning) reserves arepre-defined values. Obviously, the spinning reserve amount can al-ways be redefined, in case its associated performance is below apre-established acceptable value. The tertiary reserve (non-spin-ning) is composed by those generators that can be synchronizedwithin 1 h. This reserve is the most relevant in the present study.

The following power balance equation is set to assess the riskindices associated with the operating reserve:

ROPE ¼ RS þ RT < DLþ DPW þ DG ð6Þ

Fig. 2. Operating reserve assessment.

M. Matos et al. / Electrical Power and Energy Systems 31 (2009) 562–569 565

where DL represents the short term load deviation at hour ‘‘t”; DPW

represents the possible wind power capacity variation at hour ‘‘t”;and DG represents the generating capacity variation due to forcedoutages at hour ‘‘t”. From Fig. 2, one can observe an extra amountof capacity at the top of the tertiary reserve. This is due to the dis-crete effect of unit generating capacities.

Eq. (6) describes the risk of changes in the load, wind powercapacity and generating outages not being duly covered by theamount of spinning reserve, and also by those generators thatcan be synchronized within 1 h. Therefore, the same traditionaland well-being indices can be evaluated with this risk equation.

Although there are many reference values for LOLP or LOLE indi-ces (in capacity analysis studies), within the framework of genera-tion planning, there are no reference values for the well-beingindices nor for the operating reserve proposed framework.

2.5. Computational program characteristics

The implementation of the previous models is carried outthrough a Fortran (calculation mode) and VBasic (user interactivemode) algorithm. The convergence process is tracked through acoefficient of variation specified for the EENS index. Usually, onceensuring the convergence of EENS index, the others will have con-verged as well. The probability distributions of all, conventionaland well-being, reliability indices are also evaluated.

3. Additional studies

The simulation process described in the previous section is thebasis for adequacy evaluation of the generating system and of theoperating reserve. However, it is possible to take advantage of theinformation generated during the simulation to perform additionalstudies.

3.1. Wasted renewable energy

In wet periods that coincide with large amounts of wind powerand low load, it may happen that the system is not able to acceptall the available renewable energy. This is due to the need of main-taining in operation a specified amount of thermal power, in orderto guarantee the stability of the system.

Simulation was used to detect this kind of situations and calcu-late the total amount of wasted renewable energy in the simulatedyears. The ratio of this value to the total available renewable en-ergy is a measure of the risk of wasting renewable energy.

The effect of adding more pumping storage, thus increasing theload in valley hours, can also be evaluated in terms of reduction ofthe wasted renewable energy, helping the decision maker to assessthe impact of such a measure.

3.2. Interconnections

In neighboring countries, interconnections may be used to over-come situations of lack of generating capacity or lack of operatingreserve. In order to evaluate the correspondent benefits in terms ofrisk reduction, a four step procedure was implemented that usesthe basic simulation process as the main component.

Step 1: Perform a simulation on system A and build probabilitydensity functions (pdf) of the remaining available capac-ity (i.e. available capacity not used to serve the load), foreach month.

Step 2: Use the pdf and the net transfer limits defined for theinterconnections to generate an hourly series of avail-able import capacity for system B (neighbor to A).

Step 3: Add a new unit in system B with reliability parameterssimilar to the ones of the interconnection and with var-iable capacity equal to the hourly series defined in theprevious step.

Step 4: Perform a simulation on the modified system B.

The reliability indices obtained in Step 4, when compared to theones obtained by the normal simulation of system B, give a goodquantitative indication of the risk reduction associated to the exis-tence of the interconnection. However, more sophisticated ap-proaches may be necessary to capture the full complexity ofinterconnection impact on risk.

4. Results

The proposed algorithm has been tested under several condi-tions with different systems. A case using the Portuguese andSpanish Generating System (PGS and SGS, respectively) will be dis-cussed as follows. The case will show the results using the PGS andSGS configurations for the years 2005–2025.

In the base year 2005, the PGS had 1035 units with a total in-stalled capacity of 12.59 GW, distributed as follows: 4.38 GW (hy-dro); 5.43 GW (thermal); 0.98 GW (wind); 0.34 GW (mini-hydro);and 1.46 GW (co-generation). The annual peak load occurred in Jan-uary and it was approximately 8.53 GW. Also, the amount of renew-able power installed in the system is 45% of the total capacity.

Also in the year 2005, the SGS had 8150 units with a total in-stalled capacity of 70.2 GW, distributed as follows: 15.17 GW(hydro); 37.1 GW (thermal); 9.54 GW (wind); 0.79 GW (mini-hy-dro); and 7.6 GW (co-generation). The annual peak load occurredin January and it was approximately 43.14 GW. The amount ofrenewable power installed in the system is 36% of the total capacity.

In 2005, there were 35 hydro power plants in Portugal and 174in Spain. In this study, six hydrological basins in Portugal and alsosix in Spain were considered, and 16 years of monthly hydrologicalconditions were used (1990–2005). Fig. 3 shows the hydro produc-tion variations (dry, average and wet conditions) for Portugal andSpain. The driest year was 2005 for Portugal and 1992 for Spain,and the wettest year was 1996 for Portugal and 2003 for Spain.These monthly variations must be duly captured by the MCS-basedreliability assessment proposed algorithm.

Fig. 3. Hydro production, 1990–2005: (a) Portugal and (b) Spain.

566 M. Matos et al. / Electrical Power and Energy Systems 31 (2009) 562–569

In 2005, there were eight thermal power plants in Portugal and74 in Spain (not including co-generation). About 655 wind-gener-ating turbines (units) were in the PGS and 6365 in the SGS. Bearingin mind the wind series, Portugal was divided into seven regionsand Spain into 18 regions. In 2005, for the PGS, the day with thehighest peak capacity was November 7th, and the day with thelowest peak capacity was February 19th and for the SGS, the daywith the highest peak capacity was August 8th, and the day withthe lowest peak capacity was April 8th. Fig. 4 shows these two daysfor both systems, and also the average value of the annual windseries. In order to model the deviation DPW in (6), it was assumeda forecasting error by persistence. The error was then estimated bycomparing the wind power availability in subsequent hours.

The actual 2005 hourly load curves were used for both PGS andSGS. Uncertainty of the short-term forecasting (i.e. expression DLin Eq. (6)) was simulated using a Gaussian distribution. Sensitivityanalyses were preliminarily carried out to validate the distributionparameters used in Portugal and Spain.

Several other data involving mini-hydro and co-generationunits, maintenance, etc. have also to be processed. These data are

0.0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1.0

0 2 4 6 8 10 12 14 16 18 20 22

Prod

uctio

n (p

u)

Hour

Lowest Average Highest

(a) Fig. 4. Wind power typical fluctuations

Fig. 5. System evolution: (a)

not shown or discussed in this paper due to the lack of space,but they represent relevant information and also sources of varia-tions, which had to be carefully considered.

In order to simulate the remaining years of the study, the loadcurve shape was maintained, but the curve was adjusted to the fu-ture peak values forecasted by the two TSO. Wind power series, onthe other hand, were reconstructed taking into account the newforeseeable wind parks, the corresponding installed power andtheir geographic location. From year 2010 on, specific modelsand hourly series of solar power (PV and thermal) were specificallyconsidered, due to the relevant values foreseen in the future (up to6000 MW in 2025 in Spain). Configuration of the PGS and SGS alsochanged, with new units been added and some withdrawn fromthe systems. The main global data in the two systems can be seenin Fig. 5.

The proposed MCS algorithm was used to analyze severalpossible configurations of the PGS and SGS. Different scenariosinvolving not only hydro and wind unfavorable conditions, but alsoco-generation usage and maintenance strategies were beinganalyzed.

0.0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1.0

0 2 4 6 8 10 12 14 16 18 20 22

Prod

uctio

n (p

u)

Hour

Lowest Average Highest

(b) , 2005: (a) Portugal and (b) Spain.

Portugal and (b) Spain.

Table 1Traditional reliability indices in 2005 – (a) Portugal and (b) Spain.

Case LOLE (h) EENS (MWh/yr) LOLF (occ./yr) LOLD (h)

(a)Portugal generation adequacy

Base 0.005 0.835 0.005 0.946H+ 0.001 0.221 0.001 1.000H� 0.012 1.583 0.011 1.071HWM 0.021 3.458 0.019 1.099

Portugal operating reserveBase 0.031 3.293 0.041 0.768H+ 0.047 4.253 0.065 0.715H� 0.016 2.370 0.014 1.101HWM 0.028 4.863 0.027 1.029

(b)Spain generation adequacy

Base 0.088 69.0 0.06 1.41H+ 0.000 0.0 0.00 0.00H� 0.765 628 0.54 1.43HWM 2.531 2293.0 1.63 1.56

Spain operating reserveBase 0.227 131 0.32 0.71H+ 0.028 7.0 0.06 0.47H� 1.232 983.0 1.11 1.12HWM 3.978 3494.0 3.02 1.32

Table 2Well-being indices in 2005 – (a) Portugal and (b) Spain.

Case EH (h) FH (occ./yr) EM (h) FM (occ./yr)

(a)Portugal generation adequacy

Base 8760 0.085 0.083 0.082H+ 8760 0.022 0.022 0.021H� 8760 0.137 0.127 0.130HWM 8760 0.314 0.290 0.302

Portugal operating reserveBase 8758 2.088 1.769 2.070H+ 8757 3.722 2.949 3.681H� 8760 0.193 0.184 0.183HWM 8759 0.461 0.443 0.447

(b)Spain generation adequacy

Base 8759 0.720 0.700 0.699H+ 8760 0.003 0.003 0.003H� 8757 1.385 1.481 1.297HWM 8752 4.063 4.490 3.942

Spain operating reserveBase 8750 14.56 8.64 14.72H+ 8758 2.380 1.41 2.40H� 8750 12.80 8.06 12.80HWM 8739 24.80 16.59 25.68

M. Matos et al. / Electrical Power and Energy Systems 31 (2009) 562–569 567

For the Base case, all historical hydrological and wind serieswere simulated. In the H+ case, the wettest hydrological year wasconsidered, and in the H� case, the driest hydrological was simu-lated. The HWM case considers that the driest hydrological condi-tion occurs simultaneously with all observed wind series havingtheir capacities reduced by 50%. Also, the usual amount of poweron maintenance was increased by 20%. Certainly, this is a very se-vere scenario.

Table 1 shows the traditional reliability indices for Portugal andSpain in 2005. As it could be expected, the worst condition occursfor the ‘‘HWM” scenario. It is noted also that the operational risk isgreater than the generation risk, a fact that occurs in all thesimulations.

It is interesting to note that the Operational Reserve results inPortugal show an apparently contradictory result, the LOLE of the‘‘wet” scenario H+ (0.047) being greater than the LOLE of the otherscenarios. This is due to the commitment of more hydro units toserve the load, with a negative influence on the tertiary reserveavailable within 1 h when there is capacity surplus (as it was thecase in Portugal in 2005).

Table 2 shows the well-being reliability indices for Portugal andSpain, also in 2005. One can provide the following interpretationfor the SGS system, if everything goes wrong (i.e. HWM case) withthe 2005 configuration. In terms of generation adequacy, the SGSwill stay, in average per year, 8752 h in healthy states, 4.5 h in

Fig. 6. Risk evolution (generation adeq

marginal states, and 3.5 h in risk states. In what concerns the oper-ating reserve, the SGS will stay, in average, 8739 h in a healthystate, 16.6 h in a marginal state and 4.4 h in risk states.

The convergence criterion b was set to 5% in all tests for theEENS index, and a maximum of 3000 years were simulated. Allsimulations were carried out in a PC with 2.8 GHz. The CPU timewas, in average, 0.4 h for the PGS and 7 h for the SGS.

In order to assess the adequacy of the systems and operationalreserve in the period 2008–2025, the chronological MCS basedalgorithm was run for both the PGS and SGS. Figs. 6 and 7 showthe main results for Portugal and Spain. It is noted again higher riskvalues for the operating reserve assessment and can also see theinfluence of more adverse scenarios. It also possible to observethat, in 2015 the operational risk is greater in the wet scenarioH+ than in the Base and even Dry (H�) scenarios. A similar situationwas already reported and commented in the results of Portugal,2005.

An interesting feature of the implementation is the ability toprovide monthly values of the reliability indices. This feature wasimplemented as a result of the interaction with the two TSO, thestressed the need for this kind of information, even at the cost ofadditional simulation time. Fig. 8 shows the results for Portugaland Spain, 2025, for three of the scenarios. It is possible to observethat some months concentrate most of the risk, and also to see theimportance of the hydro component in the first three months of

uacy): (a) Portugal and (b) Spain.

02468

10121416182022

2005 2008 2010 2015 2020 2025

LO

LE

(h/y

r)

Years

Base HWM H- H+

(a)

0.0

0.5

1.0

1.5

2.0

2.5

3.0

3.5

4.0

4.5

2005 2008 2010 2015 2020 2025

LO

LE

(h/y

r)

Years

Base HWM H- H+

(b)

Fig. 7. Risk evolution (operating reserve): (a) Portugal and (b) Spain.

Fig. 8. Monthly variation of risk: (a) Portugal and (b) Spain.

0.000

0.200

0.400

0.600

0.800

1.000

1.200

1.400

1.600

1.800

Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec

LO

LE

(h/m

)

Month

H- M+ Base

(a)

0.000

0.100

0.200

0.300

0.400

0.500

0.600

0.700

Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec

LO

LE

(h/m

)

Month

H- M+ Base

(b)

Fig. 9. Effect of increased maintenance combined with decreased hydro inflow: (a) Portugal and (b) Spain.

568 M. Matos et al. / Electrical Power and Energy Systems 31 (2009) 562–569

the year, where the risk variation between scenarios is remarkable.So, in these months, the PGS is much more exposed to adverse hy-dro scenarios than in any other period of the year.

It is also possible to use the monthly variation of the indices toevaluate the influence of maintenance actions in the global reliabil-ity. Fig. 9 shows results for Portugal and Spain, 2025, both for thebase scenario and the increased maintenance (+20%) combinedwith decreased hydro inflow scenario (H� M+).

It is evident that in month 6 the risk increase (PGS case) is muchmore important than in months 1–3, even if the base risk value formonth 6 is smaller. This shows that the simulation tool can be usedto help deciding among different maintenance schedules, by char-acterizing the risk influence of each candidate schedule.

5. Final remarks

Renewable energy technologies will take a greater share of theelectricity generation mix in order to minimize the dependence onoil and the emission of CO2. While contributions from renewable

energy sources for electricity production are modest, with theexception of hydro, their market penetration is growing at a muchfaster rate than any conventional source. More renewable powersources cause, however, an increase in the number of random vari-ables and operation complexities in the system, due to the inter-mittent nature of these sources. Therefore, the determination ofthe required amount of system capacity and operating reservesto ensure an adequate supply becomes a very important aspectof generating capacity expansion analyses.

The dimensioning of operating reserve, spinning and non-spin-ning, plays an important role in systems with high penetration lev-els of renewable sources, mainly those from wind power, due to itsnatural volatility. Although there are many reference values forLOLE indices related with capacity analysis (e.g. 0.1 day/year [13],10 h/year [26]), there are no such standards for well-being indicesor for operating reserves.

Discussions on innovative criteria (e.g. the system has simulta-neously to survive the worst hydrological and wind conditions),operation strategies and assessment tools will be the new insights

M. Matos et al. / Electrical Power and Energy Systems 31 (2009) 562–569 569

of generating capacity expansion planning considering renewablepower sources for the years to come.

References

[1] EREC – European Renewable Energy Council. EREC’s Position on theFramework Directive for Renewable Energy Sources. EREC Website: <http://www.erec-renewables.org/fileadmin/erec_docs/Documents/Position_Papers/EREC_Position_Framework_Directive.pdf> (accessed 1/10/2007).

[2] Presidency Conclusions of the Brussels European Council (8/9 March 2007).Reference. DOC/07/1. European Website: <http://europa.eu/rapid/pressReleasesAction.do?reference=DOC/07/1> (accessed 09/03/2007).

[3] EU Commission Communication of 10 January 2007: Renewable Energy RoadMap – Renewable Energies in the 21st Century: Building a more SustainableFuture. European Website: <http://europa.eu/scadplus/leg/en/lvb/l27065.htm>.

[4] Rahman S. Green power: what is it and where can we find it? IEEE Power andEnergy Magazine 2003;1(1):30–7.

[5] Hatziargyriou N, Zervos A. Wind power development in Europe. Proceedings ofthe IEEE 2001;89(12):1765–82.

[6] Ramakumar R, Slootweg JG, Wozniak L. Guest editorial: introduction to thespecial issue on wind power. IEEE Trans on Energy Conversion 2007;22(1):1–3.

[7] Söder L, Hofmann L, Orths A, Holttinen H, Wan Y, Tuohy A. Experience fromwind integration in some high penetration areas. IEEE Trans on EnergyConversion 2007;22(1):4–12.

[8] Ummels BC, Gibescu M, Pelgrum E, Kling WL, Brand AJ. Impacts of wind poweron thermal generation unit commitment and dispatch. IEEE Trans on EnergyConversion 2007;22(1):44–51.

[9] Strbac G, Shakoor A, Black M, Pudjianto D, Bopp T. Impact of wind generationon the operation and development of the UK electricity systems. Electric PowerSystems Research 2007;77(9):1214–27.

[10] O’Malley M, Sanches-Gasca JJ. Guest editorial: special section on wind energy.IEEE Transactions on Power Systems 2007;22(3):899.

[11] Almeida RG, Pecas Lopes JA. Participation of doubly fed induction windgenerators in system frequency regulation. IEEE Transactions on PowerSystems 2007;22(3):944–50.

[12] Sørensen P, Cutululis NA, Vigueras-Rodriguez A, Jensen LE, Hjerrild J, DonovanMH, Madsen H. Power fluctuations from large wind farms. IEEE Transactionson Power Systems 2007;22(3):958–65.

[13] Billinton R, Allan RN. Reliability Evaluation of Power Systems. NewYork: Plenum Press; 1996.

[14] Silva MG, Rodrigues AB, Castro CLC, Coelho NA, Moutinho EA, Correa AC. Anapplication of predictive reliability analysis techniques in Brazil’s Northeastdistribution networks. International Journal Electric Power and EnergySystems 2007;29(2):155–62.

[15] Wangdee W, Billinton R. Reliability assessment of bulk electric systemscontaining large wind farms. International Journal Electric Power and EnergySystems 2007;29(10):759–66.

[16] Billinton R, Fotuhi-Firuzabad M. A basic framework for generating systemoperating health analysis. IEEE Transactions on Power Systems 1994;9(3):1610–7.

[17] Billinton R, Karki R. Application of Monte Carlo simulation to generatingsystem well-being analysis. IEEE Transactions on PWRS 1999;14(3):1172–7.

[18] Leite da Silva AM, Resende LC, Manso LAF, Billinton R. Well-being analysis forcomposite generation and transmission systems. IEEE Transactions on PowerSystems 2004;19(4):1763–70.

[19] Wang X, Dai H, Thomas RJ. Reliability modeling of large wind farms andelectric utility interface systems. IEEE Transactions on PAS 1984;103(3):569–75.

[20] Singh C, Gonzalez AL. Reliability modeling of generation systems includingunconventional energy sources. IEEE Transactions on PAS 1985;104(5):1049–56.

[21] Desrochers G, Blanchard M. A Monte Carlo simulation for the economicalassessment of the contribution of wind energy to power systems. IEEETransactions on Energy Conversion 1986;1(4):50–6.

[22] Leite da Silva AM, Melo ACG, Cunha SHF. Frequency and duration method forreliability evaluation of large-scale hydrothermal generating systems. IEEProceedings C 1991;138(1):94–102.

[23] Billinton R, Chowdhury AA. Incorporating of wind energy conversion systemsin conventional generating capacity adequacy assessment. IEE Proceedings C1992;139(1):45–56.

[24] Billinton R, Chen H, Ghajar R. A sequential simulation technique for adequacyevaluation of generating systems including wind energy. IEEE Transactions onEnergy Conversion 1996;11(4):728–34.

[25] Sayas FC, Allan RN. Generation availability assessment of wind farms. IEEProceedings C 1996;143(5):507–18.

[26] Leite da Silva AM, Manso LAF, Sales WS, Resende LC, Aguiar MJQ, Matos MA,et al. Application of Monte Carlo simulation to generating system well-beinganalysis considering renewable sources. European Transactions on ElectricalPower 2007;17(4):387–400.

[27] Leite da Silva AM, Manso LAF, Mello JCO, Billinton R. Pseudo-chronologicalsimulation for composite reliability analysis with time varying loads. IEEETransactions on Power Systems 2000;15(1):73–80.