13
Large scale integration of intermittent renewable energy sources in the Greek power sector Emmanouil Voumvoulakis a,b,n , Georgia Asimakopoulou a,b , Svetoslav Danchev a,c , George Maniatis a,c , Aggelos Tsakanikas a,d a IOBE (Foundation for Economic & Industrial Research), 11 Tsami Karatassou Street, Athens 11742, Greece b School of Electrical and Computer Engineering, National Technical University of Athens, 9 Heroon Polytechniou Street, 157 80 Zografou, Athens, Greece c Department of Economic Sciences, National and Kapodistrian University of Athens, 8 Pesmazoglou Street, 105 59 Athens, Greece d Laboratory of Industrial and Energy Economics, Department of Chemical Engineering, National Technical University of Athens, 9 Heroon Polytechniou Street, 157 80 Zografou, Athens, Greece HIGHLIGHTS c Greece needs 8.8 to 9.3 GW additional RES installations by 2020. c RES capacity credit varies between 12.2% and 15.3%, depending on interconnections. c Without institutional changes, the reserve requirements will be more than double. c New CCGT installed capacity will probably exceed the cost-efficient level. c Competitive pressures should be introduced in segments other than day-ahead market. article info Article history: Received 10 January 2011 Accepted 23 May 2012 Available online 11 August 2012 Keywords: Intermittent energy Generation expansion Reserve requirements abstract As a member of the European Union, Greece has committed to achieve ambitious targets for the penetration of renewable energy sources (RES) in gross electricity consumption by 2020. Large scale integration of RES requires a suitable mixture of compatible generation units, in order to deal with the intermittency of wind velocity and solar irradiation. The scope of this paper is to examine the impact of large scale integration of intermittent energy sources, required to meet the 2020 RES target, on the generation expansion plan, the fuel mix and the spinning reserve requirements of the Greek electricity system. We perform hourly simulation of the intermittent RES generation to estimate residual load curves on a monthly basis, which are then inputted in a WASP-IV model of the Greek power system. We find that the decarbonisation effort, with the rapid entry of RES and the abolishment of the grandfathering of CO 2 allowances, will radically transform the Greek electricity sector over the next 10 years, which has wide-reaching policy implications. & 2012 Elsevier Ltd. All rights reserved. 1. Introduction In January 2007, the European Commission published a Renewable Energy Roadmap which called for a mandatory target of 20% share of renewable energy sources (RES) in the EU’s energy mix by 2020. To achieve this objective, the EU adopted Directive 2009/28/EC in April 2009, which set individual targets for each member state. Under this Directive, Greece is committed to achieve a target of 18% RES penetration in its energy mix, which was revised upwards by the Greek state to 20% in Law 3851/2010. This translates to a 40% RES (including large hydro units) share in gross electricity consumption, under the law’s assumptions on RES integration in other activities. The integration of intermittent sources on such a large scale affects significantly the way the electricity system operates (IEA Wind Task Force 25, 2009; UK Energy Research Centre, 2006). The impact can be categorised into system balancing and reliability effects. Balancing refers to the relatively rapid short-term adjust- ment, necessary to manage demand and supply fluctuations over minutes or few hours. Reliability is related to the guaranteed availability of sufficient generation during peak demand. Inter- mittent generation increases the size of the system margin required to maintain a given level of reliability, as intermittent Contents lists available at SciVerse ScienceDirect journal homepage: www.elsevier.com/locate/enpol Energy Policy 0301-4215/$ - see front matter & 2012 Elsevier Ltd. All rights reserved. http://dx.doi.org/10.1016/j.enpol.2012.05.056 n Corresponding author. Tel.: þ30 210 92 11 223; fax: þ30 210 92 28 130. E-mail addresses: [email protected], [email protected] (E. Voumvoulakis). Energy Policy 50 (2012) 161–173

Large scale integration of intermittent renewable energy sources in the Greek power sector

Embed Size (px)

Citation preview

Energy Policy 50 (2012) 161–173

Contents lists available at SciVerse ScienceDirect

Energy Policy

0301-42

http://d

n Corr

E-m

emvoum

journal homepage: www.elsevier.com/locate/enpol

Large scale integration of intermittent renewable energy sources in the Greekpower sector

Emmanouil Voumvoulakis a,b,n, Georgia Asimakopoulou a,b, Svetoslav Danchev a,c, George Maniatis a,c,Aggelos Tsakanikas a,d

a IOBE (Foundation for Economic & Industrial Research), 11 Tsami Karatassou Street, Athens 11742, Greeceb School of Electrical and Computer Engineering, National Technical University of Athens, 9 Heroon Polytechniou Street, 157 80 Zografou, Athens, Greecec Department of Economic Sciences, National and Kapodistrian University of Athens, 8 Pesmazoglou Street, 105 59 Athens, Greeced Laboratory of Industrial and Energy Economics, Department of Chemical Engineering, National Technical University of Athens, 9 Heroon Polytechniou Street,

157 80 Zografou, Athens, Greece

H I G H L I G H T S

c Greece needs 8.8 to 9.3 GW additional RES installations by 2020.c RES capacity credit varies between 12.2% and 15.3%, depending on interconnections.c Without institutional changes, the reserve requirements will be more than double.c New CCGT installed capacity will probably exceed the cost-efficient level.c Competitive pressures should be introduced in segments other than day-ahead market.

a r t i c l e i n f o

Article history:

Received 10 January 2011

Accepted 23 May 2012Available online 11 August 2012

Keywords:

Intermittent energy

Generation expansion

Reserve requirements

15/$ - see front matter & 2012 Elsevier Ltd. A

x.doi.org/10.1016/j.enpol.2012.05.056

esponding author. Tel.: þ30 210 92 11 223; f

ail addresses: [email protected],

[email protected] (E. Voumvoulakis).

a b s t r a c t

As a member of the European Union, Greece has committed to achieve ambitious targets for the

penetration of renewable energy sources (RES) in gross electricity consumption by 2020. Large scale

integration of RES requires a suitable mixture of compatible generation units, in order to deal with the

intermittency of wind velocity and solar irradiation. The scope of this paper is to examine the impact of

large scale integration of intermittent energy sources, required to meet the 2020 RES target, on the

generation expansion plan, the fuel mix and the spinning reserve requirements of the Greek electricity

system. We perform hourly simulation of the intermittent RES generation to estimate residual load

curves on a monthly basis, which are then inputted in a WASP-IV model of the Greek power system. We

find that the decarbonisation effort, with the rapid entry of RES and the abolishment of the

grandfathering of CO2 allowances, will radically transform the Greek electricity sector over the next

10 years, which has wide-reaching policy implications.

& 2012 Elsevier Ltd. All rights reserved.

1. Introduction

In January 2007, the European Commission published aRenewable Energy Roadmap which called for a mandatory targetof 20% share of renewable energy sources (RES) in the EU’s energymix by 2020. To achieve this objective, the EU adopted Directive2009/28/EC in April 2009, which set individual targets for eachmember state. Under this Directive, Greece is committed toachieve a target of 18% RES penetration in its energy mix, which

ll rights reserved.

ax: þ30 210 92 28 130.

was revised upwards by the Greek state to 20% in Law 3851/2010.This translates to a 40% RES (including large hydro units) share ingross electricity consumption, under the law’s assumptions onRES integration in other activities.

The integration of intermittent sources on such a large scaleaffects significantly the way the electricity system operates (IEAWind Task Force 25, 2009; UK Energy Research Centre, 2006). Theimpact can be categorised into system balancing and reliabilityeffects. Balancing refers to the relatively rapid short-term adjust-ment, necessary to manage demand and supply fluctuations overminutes or few hours. Reliability is related to the guaranteedavailability of sufficient generation during peak demand. Inter-mittent generation increases the size of the system marginrequired to maintain a given level of reliability, as intermittent

E. Voumvoulakis et al. / Energy Policy 50 (2012) 161–173162

source plants are less likely to be fully utilised at times of peakdemand. As such, they contribute less to ancillary services (ETSO,2007).

In addition, a significant reduction of the mean average loading(capacity factor) of thermal units shall be expected, since a largeportion of them will not participate in the market during hours ofhigh energy production by intermittent sources. Meanwhile, thefrequency of shut-downs and start-ups of compatible units willincrease, leading to larger fuel consumption and maintenance costand lower lifetime and efficiency of the units (Poyry EnergyConsulting, 2009). Under these conditions the need of recent-vintage,fast-ramping, combined cycle gas turbines is obvious (Puga, 2010).

In this paper, we develop a method that determines theoptimum planning of generation expansion under the constraintof achieving the target of 40% share of RES in the Greek electricitysector by 2020. We utilise the software package WASP-IV in orderto establish optimal generation expansion plans, while a simula-tion process is executed in order to deal with the generation ofintermittent sources. We also estimate the fuel mix and thespinning reserve requirements of the Greek electricity systemand discuss the policy implications of our results.

The paper is structured as follows: Section 2 presents ourcontribution to the debate on the optimal structure of the Greekelectricity sector by outlining the major differences of our studyin comparison with previous research. Section 3 contains a briefoverview of the Greek power system, while Section 4 describesthe methodology of the paper. Section 5 presents the quantitativeresults of the study. The paper concludes with a discussion of thepolicy implications of our results.

2. Previous research and contribution of the study

Several studies have used WASP-IV to determine an optimumgeneration plan for the Greek Energy System (Agoris et al., 2004;Dagoumas et al., 2007; Kalampalikas and Pilavachi, 2010a, 2010b;Vassos and Vlachou, 1997). In the most recent of these,Kalampalikas and Pilavachi (2010a, 2010b) use economic, envir-onmental and energy criteria to evaluate the desirability ofmeeting the RES target under three alternative assumptions aboutthe composition of new thermal capacity. Assuming that electri-city demand grows at a fixed rate, the authors conclude thataccording to the environmental criteria the best choice is to meetthe RES target with higher share of natural gas, yet meeting thetarget as such is ‘‘almost impossible’’ according to the energycriteria and ‘‘uneconomical’’ according to the economic criteria, asit is associated with higher total cost.

Rampidis et al. (2010) use the BALANCE module of the ENPEPsoftware to examine the feasibility of announced investmentplans to develop new thermal capacity in the Greek electricitysector. The study concludes that various ambitious expansionscenarios are either unrealistic or inefficient, while an efficientand realistic scenario achieves only 6.3% reduction in CO2 emis-sions in 2020, in comparison with 2005. The extent of thisreduction is rather low, compared with the EU-wide goal of 21%for the ETS sectors. Meanwhile, the study includes options, suchas imported coal, which are not part of the current official energypolicy of the country, while it also assumes RES capacity that isnot sufficient to achieve the target of 40% penetration in demand.

In addition, Tolis et al. (2010) perform an analysis of the futurestructure of generation in the Greek electricity sector usingsequential quadratic programming in a long-term horizon (until2050). The authors model prices and variable costs as Markovprocesses. Electricity demand is assumed to follow a random walkprocess, while in an alternative scenario it is taken as a function ofelectricity prices. The intermittency of RES is not explicitly

modelled. Instead, the authors impose a higher reserve marginto take into account the probability that wind and solar might notbe available at peak demand and set a maximum level of RESpenetration (50% in annual demand) to limit the adverse effects ofintermittency. The target for minimum RES penetration in 2020 isset at 30%, below the 40% target in the RES law. Meanwhile, themaximum RES penetration of 50% is not sufficient to decarbonisethe electricity sector by 2050, especially without carbon captureand storage and nuclear technologies, which is not in line withthe ambition of the EU countries (Decision No 406/2009/EC) tolimit the emissions of green house gasses by 60–80% by 2050(in relation to 1990).

Our study differs in a number of ways. First, we do not attemptto evaluate the desirability of achieving the RES targets. Summingthe private costs of items traded in markets is not enough in orderto establish the economic desirability of a project from a socialperspective. Such an exercise should include monetary estimatesof all the external positive and negative effects, which is beyondthe scope of our study. It is assumed that the cost of not avoidingthe adverse effects of global warming, as well as the adversehealth effects from pollutants, such as SO2, NOx, dust, etc. are highenough to surpass the private cost premium from high RESpenetration. The external costs can be substantial—it has beenestimated that if we take into account the cost of environmentalexternalities, the production cost in the Greek electricity sectorwould be higher by 52% (Georgakellos, 2010).

For the same reason, our study does not include RES in theoptimisation algorithm. From the perspective of the investors, therevenues of the RES plants could have been modelled by usingthe current support mechanism, but the outcome of the optimisa-tion would not have necessarily indicated a cost-effective solu-tion. Alternatively, as in Tolis et al. (2010), the revenue of REScould have been modelled in terms of income from tradingemission allowances, coming from not producing electricity withfossil fuels. However, the RES generators in Greece do not haveaccess to such revenue, as they are not allocated allowances thatthey can sell. The RES revenue assumption might be reasonable ifviewed from a social planner’s perspective, but for consistentimplementation of this perspective, all external costs and benefitsof all options under review should also be taken into account,which is beyond the scope of our study.

In order to account for the intermittency of RES, we considerpower generation from intermittent RES as a negative load. Theequivalent load of the system in our paper is the load subtractingthe generation of intermittent power sources. In order to find theduration curve of this equivalent load, we perform a Monte Carlosimulation of the generation from intermittent energy sources.This approach also allows us to calculate the expected capacitycredit of the RES units at peak, instead of using the reserve marginas a criterion for scenario evaluation or as a constraint in theoptimisation problem.

Furthermore, we do not impose ex-ante the technology compo-sition of the new thermal plants in the system. Instead, thecomposition is provided by the optimisation algorithm of WASP-IV. This allows us to compare the current declared entrepreneurialinterest with the optimal path of generation expansion. Our studyrelaxes the zero loss of load probability requirement inKalampalikas and Pilavachi (2010a, 2010b), setting the constraintat 2.5%. This relaxation is substantiated, in our view, by the fact thatthe Greek system is connected with other electricity systems, whileGreece has been a net importer of electricity in the past few years.

In addition, we model the effects of the economic crisis, whichin Greece is expected to last longer, compared with othereconomies of the Euro area, due to the country’s fiscal problems.We model demand as a function of GDP growth and electricityprices, using macroeconomic projections over the 2010–2020

E. Voumvoulakis et al. / Energy Policy 50 (2012) 161–173 163

period and we assume interest rates that reflect the substantiallyincreased sovereign risk premium of the country.

We also incorporate the share of each RES technology in thetotal RES capacity from the Ministry of Energy’s ‘‘National Renew-able Energy Action Plan in the Scope of Directive 2009/28/EC’’,which implies that our results can be used in an impact assess-ment of the plan. In comparison with Kalampalikas and Pilavachi(2010a, 2010b) the Action Plan has substantially higher share ofwind and PV plants, at the expense of the share of small hydrounits. This implies that the electricity system projected in ourstudy is substantially more exposed to the adverse effects ofintermittency. Finally, we examine the impact of increasing theinterconnection and storage capabilities of the Greek electricitysystem on the optimal generation expansion plan and the reserverequirements, which to the best of our knowledge has not beenexamined in this context.

3. Description of the Greek Power System

The Hellenic Interconnected System (HIS) serves the needs of themainland and a few interconnected islands. Gross electricity demandduring 2009 was about 53.7 TW h. The compound annual growth rateof electricity demand reached about 4% in the last decade. Thedemand has been served mainly by thermal power, large hydroplants and imports, while the total net installed capacity of thesystem stood at 12.3 GW, as of end of 2009 (Table 1).

The main production centre is situated in North-West Greecein the vicinity of a lignite rich area. Significant hydro productiontakes place in the north and northwest of the country, while somelignite production is also available in South Greece (Peloponnesepeninsula). There are also wind farms (WFs) of total nominalcapacity 915 MW (December 2009), most of which are installed atthe island of Evia and in Thrace. These WFs contributed about 3%of the electricity needs during 2009. In addition, 413 MW of otherRES and small-scale combined heat and power (CHP) technologiesare in operation.

A number of islands, such as Crete, Rhodes and others are notconnected to the mainland system. These autonomous systemsrepresent approximately 8% of the electricity demand in Greece(Iiadou, 2009). The electricity generation on these islands relieson petroleum products and, to a much lesser extent, on RES.

The system is interconnected with Albania, Bulgaria, andFYROM via three 400-kV tie lines of total available transfercapacity of 600 MW and to Italy via an asynchronous 400-kVAC–DC–AC link with a transfer capacity of 500 MW. The system isalso connected with Turkey with a 400-kV line since the summerof 2008, but the commercial operation of this interconnection has

Table 1Installed capacity and the energy balance in the interconnected system in 2009.

Type Net capacity(GW)

Annual production(GW h)

Thermal 8.0 41,616

Lignite 4.8 30,542

Oil 0.7 1697

Natural Gas 2.5 9377

Large hydros 3.0 4955

RES & CHP 1.3 2937

Wind 0.9 1908

Small hydro 0.2 657

Biomass/biogas 0.04 182

CHP 0.1 144

PV 0.04 46

Total 12.3 49,508

not yet started, as the Turkish system synchronised with ENTSO-Eonly recently (September 2010).

Institutionally, the Greek wholesale market in the intercon-nected system is organised as a day-ahead mandatory pool,coupled with a capacity assurance mechanism. The generators inthe system can operate only if selected by the market operator(Hellenic Transmission System Operator HTSO), based on theirbids in the day-ahead market. The suppliers cannot purchaseenergy outside the pool. Bilateral contracts about financial settle-ments are allowed, but they cannot include physical delivery(Iiadou, 2009). The settlement of imbalances is performed ex postby the market operator at imbalances marginal price, calculatedusing the day-ahead bids and actual load during dispatch. Addi-tional payments are exchanged for forced production changes. Assuch, there is no provision for bidding in a real-time balancingmarket, while there is also no market for electricity derivatives.

The RES generators (with the exception of large hydro units)do not bid in the day-ahead market. The RES energy enters thepool with priority, provided that it does not put the security of thesystem at risk. All RES plants, apart from the large hydro units, arepaid administratively fixed feed-in-tariffs for the energy that theysupply to the system.

In addition, the regulatory framework envisages a capacityassurance mechanism with free negotiations between suppliersand generators over capacity certificates, which the suppliers arerequired to submit to HTSO to cover the peak load of theircustomers. This mechanism is still not operational. The durationof the transitory mechanism, where a fixed fee is paid per MW ofinstalled capacity, was recently extended for an additional year.

4. Method overview

We use time series of meteorological data, such as windvelocity, rainfall and solar irradiation, and the planned installedcapacity per RES technology and region, together with a numberof other inputs (Fig. 1), in order to estimate RES generation for theperiod 2010–2020.

The estimation for the intermittent RES generation is based ona Monte Carlo simulation (see Appendix A). We then checkwhether the RES generation is sufficient against the indicativepath of achieving the 2020 targets, set in Annex I of the Directive2009/28/EC and if necessary we adjust the production capacityaccordingly (Appendix B).

Based on the simulation of load demand and RES generation,we estimate an equivalent load duration curve (i.e., the residualdemand that must be served by thermal generation and largehydro units) on a monthly basis for the period under study.Taking into account the decommissioning plan of the incumbentproducer and other characteristics of the Greek electricity system,we use WASP IV to estimate the optimal generation expansionplan, needed to cover the equivalent load. We also estimate thespinning reserve requirements (see Appendix C).

4.1. Inputs

The inputs of our analysis are presented in Table 2.Each set of inputs constitutes a distinct scenario of the study. In

this paper we investigate 3 scenarios, which differ with respect to theinterconnection and storage capabilities of the system (Table 3).

In the 3 main scenarios, we assume constant real prices andcosts, implying that relative prices remain unchanged. The priceof CO2 allowances in the ETS after 2013 is fixed at 25 h/t CO2. Thepredictions on Greek GDP growth in the period 2010–2012broadly reflect the May 2010 projections of the Greek Ministryof Finance. In the remaining period, income growth stands at 2.8%

Table 2Description of the data inputs.

Input Description

Interconnected

system structure

Refers to the current configuration of the

interconnected generation system as well as to its

interconnection capacity to neighbouring power

systems.

Autonomous power

systems

Refers to the current configuration of the power

systems in the islands that are not connected to the

mainland system.

Transmission

expansion

Refers to the future interconnection capacity of the

mainland system to either its neighbouring systems or

with autonomous systems.

Storage capacity Refers to existing and future pump storage

installations.

Electricity demand

forecast

These forecasts include annual energy and peak load

forecast. The historical data of hourly load demand are

adapted according to the forecasts to provide hourly

time series, covering the 2010–2020 period.

RES Licence

Applications

This is a proxy for the expressed interest for

investment in renewable energy sources per

technology and region.

RES generation

intermittency

These data include historical hourly time series of the

production of intermittent sources per region.

Decommissioning

plan

This refers to an indicative plan to decommission a

number of plants that are not expected to be

economically viable or compliant with EU emission

legislation over the period under study. By 2020, we

have assumed that 2.9 GW of thermal capacity will be

decommissioned, including 1.7 GW lignite plants,

512 MW natural gas plants and all the plants using

fuel oil in the interconnected system.

Demand prediction The demand prediction is generated, using a demand

function with income elasticity of 1.25 and own-price

elasticity of �0.2. These elasticity levels broadly

reflect findings from previous research

(Christodoulakis and Kalyvitis, 1997;Christodoulakis

et al., 2000; Donatos and Mergos, 1991; Hondroyannis,

2004; Polemis, 2007; Rapanos and Polemis (2006))

and our own estimates. We used the GDP projections

of the macroeconomic unit of IOBE.

Fig. 1. Framework of the analysis.

E. Voumvoulakis et al. / Energy Policy 50 (2012) 161–173164

on average. We assume that once an island connects to themainland system, all the units burning fuel oil in the island areput in cold reserve.

4.2. Sensitivity analysis

The analysis presented in this paper is based on a substantialnumber of assumptions. Meanwhile, the output of the analysis

covers a substantial number of result variables. Presenting thechange in each result variable from sensitivity analysis of eachmajor input parameter would generate a great multitude ofnumbers, which would be very difficult to interpret in a concisemanner. Instead, we opted for building two more scenarios thatreflect the uncertainty on the economic conditions that the Greekelectricity sector faces. They are based on the intermediatescenario and are differentiated with respect to GDP growth (andthus electricity demand), prices of imported fuels (i.e., natural gasand oil) and prices of CO2 allowances.

In the stagnation scenario (M-S), we assume that the globaleconomic imbalances are not dealt with and the world growthremains sluggish. As a result of the repressed global demand, thenatural gas glut does not disappear until the end of the 2010s,while its real price declines by 3% annually (implying about 1%fall in nominal prices under 2% global inflation regime). The lowerlevels of energy generation in Europe and the continued avail-ability of cheaper natural gas on the spot market makes thecompliance with the emissions cap less costly, which pushesdown the price of the CO2 allowances to 10 h/t CO2 in the nextphase of ETS (2013–2020). We incorporated this into the modelby adjusting the variable component of non-fuel operation andmaintenance cost of each unit by a factor of 0.4, which corre-sponds to the ratio of carbon price in the stagnation and in thethree interconnection scenario (25 h/t CO2). As such, we assumethat the variable component of non-fuel and non-carbon opera-tion and maintenance cost is negligible. Meanwhile, the feebleglobal economic conditions keep risk aversion high, whichimpacts on the ability of the Greek government and privatecompanies to raise funds in the global capital markets. As aresult, the Greek economy diverges from that of the Euro area,growing at 0.8% annually on average in the period 2015–2020.

In the high-growth scenario (M-HG), the global coordinationon economic policies is considerably more successful and theglobal economic imbalances are addressed equitably. The reducedrisk aversion eases the credit crunch, facilitating investment andgrowth. Higher global economic growth implies that the naturalgas glut is quickly absorbed, which results in 3% annual growth innatural gas prices on top of consumer inflation. Lower relativeprice of coal and higher energy demand push the price of the CO2

allowances to 35 h/t CO2. As a result, the variable component ofnon-fuel operation and maintenance cost of each unit is multi-plied by a factor of 1.4, which corresponds to the ratio of carbonprice in the high-growth and in the three interconnection sce-naria (25 h/t CO2). Meanwhile, the improved conditions in theglobal capital markets ease the credit conditions in the Greek

50

55

60

65

70

75

2006

2007

2008

2009

2010

2011

2012

2013

2014

2015

2016

2017

2018

2019

2020

TWh

M-S M M-HG

Fig. 2. Annual load in the interconnected system.

Table 3Scenario description.

Scenario North Aegeaninterconnection

Crete & South Aegeaninterconnection

New pumped storagecapacity (MW)

Capacity of newinternationalinterconnections (MW)

Baseline (S) No No 0 0

Intermediate (M) Yes No 365 0

Ambitious (L) Yes Yes 865 1.000

E. Voumvoulakis et al. / Energy Policy 50 (2012) 161–173 165

economy, which moderates the reform fatigue, allowing deeperand more substantial implementation of the structural reforms.As a result, the Greek economy returns on a convergence path,growing at 4.2% on average in the period 2015–2020.

As a result of the different course of GDP growth and prices,gross electricity demand varies across the scenarios (Fig. 2).Under the stagnation scenario, electricity demand grows slowlyin the aftermath of the crisis, but does not reach the pre-crisislevels. In the other two scenarios, the pre-crisis gross demandlevels are reached by 2016.

By 2020, the gross electricity demand in the interconnectedsystem is assumed to vary in the range of 59–74 TW h. Thedifference in demand between the intermediate and the high-growth scenarios is less pronounced because the improvedmacroeconomic conditions in the latter scenario are accompaniedwith higher electricity prices, which dampen demand.

4.3. WASP-IV parameters

The estimation of the optimal generation expansion plan,using WASP-IV requires the specification of parameters such as:

Capital-investment costs of the expansion candidate technologies. � Operating costs (i.e., fuel, non-fuel and maintenance) of exist-

ing and candidate units.

� Cost of energy not supplied. � Discount rate. � Fuel and CO2 cost escalation. � System performance constraints.

We limited the choice of thermal technologies by excludingoptions that are not part of the current Greek energy strategy(e.g., hard coal and nuclear plants). The techno-economic char-acteristics of the candidate technologies are presented in Table 4.We consider four types of candidate plants, which differ in termsof fuel used, installed capacity, efficiency etc.

We set the discount rate at 10% to reflect the increased riskpremium on Greek debt, which most probably will remain high inthe future, compared to its levels in the past decade. It is hard to

believe that the spreads on Greek bonds over their Germancounterparts will return to the pre-crisis level. The belief that aEuro area country enjoys the same credit worthiness withGermany is highly unlikely to return under the current institu-tional arrangements in the Euro area.

Regarding the system performance constraints, we set theinstalled capacity in the critical period to lie above 20% of peakload demand. The maximum reserve margin was set at 40% of thepeak load. The annual loss of load probability (LOLP) was set to beless than 2.5%. The value of energy not supplied was assumed toequal 0.05 h per kW h not supplied.

4.4. Derivation of the optimal expansion plan

The WASP (Wien Automatic System Planning) software toolpermits the user to find an optimal expansion plan for a powergenerating system over a long period, within the constraintsdefined by the planner. It is maintained by the IAEA (InternationalAtomic Energy Agency), who has developed four versions of theprogram and distributed it to several hundred users.

In WASP the optimum expansion plan is defined in terms ofminimum discounted total costs. The entire simulation is carried outusing 12 load duration curves to represent each month, for up to amaximum duration of 30 years. Given the electricity demand for thefuture year, WASP explores all possible sequences of capacityadditions that could be added to the system within the requiredconstraints. These constraints can be based on achieving a certainlevel of system reliability, availability of certain fuels, build-up ofvarious technologies, or environmental emissions. The differentalternatives are then compared with one another using a costfunction, which is composed of capital investment costs, fuel costs,operation and maintenance costs, fuel inventory costs, salvage valueof investments, and cost of energy demand not served.

WASP was developed in the early 1970s for the needs of theInternational Atomic Energy Agency and, as such, it is not wellsuited to model intermittent sources of energy, which are oftentreated as hydro units with monthly variation of availability.However, we believe that the limitations of WASP regardingrenewables do not invalidate the key findings of our analysis.

The key limitation of WASP, i.e., its weakness to fully take intoaccount the cost of generation intermittency for the determina-tion of the optimal mix of renewable energy is not relevant in ouranalysis, as we take the RES mix as determined exogenously bypolicy. In addition, for the determination of the optimal genera-tion plan of the thermal units, we attempt to correct for theinability of WASP to model intermittent sources by consideringthe power generation from intermittent RES as a negative load.Still, there are several WASP limitations, such as the treatment ofthe interconnection capacities as exogenous, which remain in ouranalysis and are left for further research and modelling.

5. Results

The target of 40% RES share in the energy mix until 2020requires between 9.9 and 10.5 GW installed capacity in RES

Table 5Required RES installed capacity in the interconnected system (MW).

Technology 2009 2010e 2015 2020

Baseline scenarioWind 915 1314 3165 7536

Small hydro 183 193 227 305

Solar 46 200 843 2363

Biomass and other 41 57 117 259

Total 1184 1764 4351 10,464

Intermediate scenarioWind 915 915 2997 7509

Small hydro 183 186 222 296

Solar 46 60 789 2268

Biomass and other 41 44 111 255

Total 1184 1205 4119 10,328

Ambitious scenarioWind 915 924 2475 7073

Small hydro 183 186 208 261

Solar 46 66 617 2322

Biomass and other 41 44 94 307

Total 1184 1220 3394 9963

Table 6RES contribution to the reliability of the system.

Variable 2010e 2015 2020

Baseline scenarioTotal energy demand (TW h) 52.3 53.2 62.1

Equivalent energy demand (TW h) 46.9 44.1 41.8

Peak load (MW) 9717 10,395 11,527

Equivalent peak load (MW) 9380 9455 10,124

Capacity credit (MW) 337 940 1403

Intermediate scenarioTotal energy demand (TW h) 52.3 53.2 65.0

Equivalent energy demand (TW h) 48.0 44.5 44.0

Peak load (MW) 9717 10,395 12,147

Equivalent peak load (MW) 9470 9425 10,517

Capacity credit (MW) 247 970 1630

Ambitious scenarioTotal energy demand (TW h) 52.3 53.4 69.3

Equivalent energy demand (TW h) 48.0 45.9 47.6

Peak load (MW) 9717 10,395 13,152

Equivalent peak load (MW) 9476 9408 11,145

Capacity credit (MW) 241 986 2006

Table 4Technoeconomic characteristics of the candidate technologies a.

Parameter name Type 1 Type 2 Type 3 Type 4

Fuel Natural gas Lignite Natural gas Lignite

Min. operating level (MW) 200 300 122 150

Max. generating capacity (MW) 425 585 287 400

Heat rate at minimum operating level (kcal/kW h) 1515 2073 2403 2150

Average incremental heat rate between minimum and maximum operating levels (kcal/kW h) 1515 2073 2403 2150

Domestic fuel costs (c/106 kcal) 0 939 0 823

Foreign fuel costs (c/106 kcal) 3882 0 4550 0

Forced outage rate (%) 7 7 7 7

Number of days per year required for scheduled maintenance 28 42 21 35

Maintenance class size (MW) 400 585 287 400

Fixed component of non-fuel operation and maintenance cost (Euros/kW/month) 1.7 2.9 2.7 2.9

Variable component of non-fuel operation and maintenance cost (Euros/MW h)b 13.1 35.2 27.9 36.2

Heat value of the fuel used by plant, measuring the heat equivalent of 1 kg fuel used (kcal/kg) 8780 1366 9570 1750

Depreciable domestic capital cost ($/kW) of expansion candidate plant 0 1500 0 1500

Depreciable foreign capital cost ($/kW) 720 1000 720 1000

Plant life (in years and fractions of years) to be used for salvage value calculation 30 45 30 45

Interest during construction included in COSTL and COSTF (in %) 8.5 13.4 8.5 13.4

Construction time (in years) 2.5 4.0 2.5 4.0

a Data on the technical characteristics and the operating costs of the existing and candidate units was provided to the authors by the Hellenic Electricity Association

under a clause that we do not disclose the data for the existing units.b The cost of CO2 permits is incorporated in the variable component of non-fuel operation and maintenance cost, assuming that the permits’ price equals 25 h/t CO2.

E. Voumvoulakis et al. / Energy Policy 50 (2012) 161–173166

(Table 5), excluding the capacity of large hydro units. Given thecurrent level of RES capacity, this implies that Greece needs 8.8 to9.3 GW additional RES installations.

5.1. System reliability

The intermittency of RES does not allow them to participatefully in the task of meeting the peak of demand. Nevertheless, itwould be wrong to assume that they are completely unavailableduring the annual peak of the system. The contribution of theestimated installed capacity of RES to peak load in 2020 isestimated at 1.4–2.0 GW, depending on the scenario (Table 6).The capacity credit of RES, as percentage of peak demand,increases with their higher geographical dispersion through theinterconnection of the islands from 12.2–15.3%.

5.2. Balancing requirements

Load balance requirements are directly related to the genera-tion fluctuations of intermittent sources, which depend on the

time span of production planning. Under the current regime, theproduction plan for all thermal and large hydro generation unitsis drawn based on offers submitted the previous day.

With a 24 h gate closure span, the required spinning reservecapacity for the system is expected to reach 3.9–4.3 GW in 2020(Table 7). The spinning reserve requirements are higher under thescenarios with a more integrated system, because the intercon-nected system has higher load and RES production, which it hastaken up from the previously autonomous systems of the newlyconnected islands.

Nevertheless, under these scenarios the burden on thermalgeneration is lower, since the additional storage capacity(þ365 MW in scenario M and þ865 MW in scenario L) and, inthe case of scenario L, the increased import capacity (þ1000 MW)overcompensate for the higher spinning reserve requirements(þ287 MW in scenario M and þ389 MW in scenario L).

Reducing the planning period by better prediction algorithms,introduction of a balancing market, instituting new more flexiblepower plants and other measures could reduce these require-ments almost by half. Demand response programmes, which have

Table 8Installed capacity of thermal units in the interconnected system (MW).

Energy source 2009 2010e 2015 2020

Baseline scenarioLignite 4802 4802 4576 3548

Natural gas 2448 2873 2786 3636

Oil products 718 718 0 0

Total 7968 8393 7362 7184

Intermediate scenarioLignite 4802 4802 4576 3548

Natural gas 2448 2873 3211 3636

Oil products 718 718 0 0

Total 7968 8393 7787 7184

Ambitious scenarioLignite 4802 4802 4576 3548

Natural gas 2448 2873 3211 3211

Oil products 718 718 0 0

Total 7968 8393 7787 6759

Table 9Electricity generation fuel shares (%).

Energy source 2009 2010e 2015 2020

Baseline scenarioLignite 55.6 43.6 36.0 21.6

Natural gas 17.1 28.1 28.2 28.2

Oil products 12.1 12.0 8.9 8.8

Large hydro 9.0 7.4 7.7 6.5

Other RES 6.2 8.9 19.2 34.9

Intermediate scenarioLignite 55.6 44.0 33.7 22.9

Natural gas 17.1 29.2 31.2 29.6

Oil products 12.1 12.7 8.9 6.4

Large hydro 9.0 7.4 7.7 6.4

Other RES 6.2 6.6 18.5 34.8

Ambitious scenarioLignite 55.6 44.2 35.1 26.0

Natural gas 17.1 29.1 32.5 28.8

Oil products 12.1 12.6 8.9 0.0

Large hydro 9.0 7.4 7.7 7.0

Other RES 6.2 6.7 15.8 38.2

Table 7Spinning reserve requirements (MW) per planning-ahead period.

Planning-ahead period 2010e 2015 2020

Baseline scenario24 h 1899 2490 3940

1 h 1453 1602 2050

Intermediate scenario24 h 1862 2435 4227

1 h 1471 1589 2087

Ambitious scenario24 h 1828 2279 4329

1 h 1445 1562 2100

E. Voumvoulakis et al. / Energy Policy 50 (2012) 161–173 167

been shown in simulations to reduce considerably the reservecost and to improve the reliability indices (Shayesteh et al., 2010),can provide a viable, cost-efficient, large-scale solution underhigh penetration of intermittent renewable energy sources(Moura and de Almeida, 2010), especially if real-time dynamicpricing is deemed undesirable. Another reasonable option is tointroduce a new reserve category, in order to reflect the fact thatchanges in wind variability are much slower than traditionalcontingencies, while the replacement of reserves after a change inwind power output need not be as quick due to the low likelihoodof further major change in wind power output (Botterud et al.,2010).

5.3. Installed thermal capacity in the interconnected system

Given the residual load that has to be served and the plan todecommission old plants, we calculated the required new capa-city from thermal units in the interconnected system. Dependingon the scenario, the Greek interconnected system needs between1.7 and 2.1 GW of new capacity from thermal units by 2020(compared with 2009). Out of this, 1.3–1.7 GW represents capa-city of CCGT, using natural gas, and the remaining 400 MW refersto new lignite capacity.

Taking into account the optimal generation expansion plan,the total installed capacity of thermal power plants in theinterconnected system in 2020 varies between 6.8 and 7.2 GW,compared with 8.0 GW in the base year (Table 8). This impliesthat, given the projected load demand, the new thermal capacitywill substitute decommissioned plants, without increasing theoverall capacity of the thermal power system. As most of thedecommissioned capacity comes from lignite plants, the installedcapacity of lignite units will decrease by 26%. Meanwhile, theinstalled capacity of natural gas units will increase by 31–47%,reaching 3.2–3.6 GW. This result is consistent with the finding inKalampalikas and Pilavachi (2010a, 2010b) that the scenarioswith higher share of new natural gas capacity have lowertotal cost.

5.4. Electricity generation fuel mix

The change in the composition of capacity leads to a change inquantity of energy produced by each technology. The share oflignite units in national electricity production (interconnectedand autonomous systems) is expected to shrink from 56% in 2009to 22–26% in 2020 (Table 9). Meanwhile, the share of natural gasunits is expected to expand from 22% in 2009 to 28–30% in 2020.

5.5. Capacity factors of thermal power plants

The shift from lignite to natural gas, however, is not only due tochanges in installed capacity composition. The newly introduced

natural gas units will be more competitive than most of the olderlignite plants after the abolition of the grandfathering of CO2

allowances in 2013, under the variable cost assumptions in thisstudy. As a result, the average capacity factor of the lignite plantsdrops to 46–51% in 2020 (from 73% in 2009), while the averagecapacity factor of the natural gas units increases to 59–63%(Figs. 3 and 4). In terms of equivalent operating hours, this impliesthat the average lignite unit operates 4035–4494 h per year,depending on the scenario, while the natural gas units operate5130–5500 h per year on average. The fall in the capacity factor oflignite units is consistent with the result in Tolis et al. (2010),where the optimal usage factors of lignite plants are (in general)lower than those of CCGT units after 2013.

5.6. Spilled RES energy

The importance of interconnection and storage capabilities ismostly evident in the difference between the scenarios in terms offrequency and quantity of spilled energy. The production fromintermittent energy sources is curtailed when it surpasses theload demand of the system, net of the remaining must-run load,such as the technical minima of the thermal production units thatparticipate in the system in that particular hour. Under thebaseline scenario, the spilled energy in 2020 amounts to 7.6% of

0.00.20.40.60.81.01.21.41.61.8

2010

2011

2012

2013

2014

2015

2016

2017

2018

2019

2020

Spi

lled

Win

d E

nerg

y (T

Wh)

Βaseline Intermediate Ambitious

Fig. 4. Spilled wind energy.

Table 10Installed capacity of thermal units in the interconnected system—Sensitivity

analysis.

Energy source 2009 2010e 2015 2020

Scenario M–SLignite 4802 4802 4576 3548

Natural gas 2448 2448 2786 2786

Oil products 718 718 0 0

Total 7968 7968 7362 6334

Scenario MLignite 4802 4802 4576 3548

Natural gas 2448 2873 3211 3636

Oil products 718 718 0 0

Total 7968 8393 7787 7184

Scenario M–HGLignite 4802 4802 4576 3548

Natural gas 2448 2873 3211 4486

Oil products 718 718 0 0

Total 7968 8393 7787 8034

0%

10%

20%

30%

40%

50%

60%

70%

80%

2009

2020

-S

2020

-M

2020

-L

Lignite Natural gas Oil

Fig. 3. Capacity factors in the interconnected system.

E. Voumvoulakis et al. / Energy Policy 50 (2012) 161–173168

the total energy supplied by wind farms. Meanwhile, the spilledRES energy is limited to only 0.5% in the ambitious scenario due tosubstantially higher storage and interconnection capabilities.

5.7. Sensitivity analysis

The difference between the variable cost of lignite and naturalgas units is minimal after 2013 under the price assumptions ofour analysis. This implies that small changes in the relative pricesof fuels and allowances could overturn the result that CCGT unitsenjoy higher utilisation rate than lignite plants. Meanwhile, amore sluggish growth of the Greek economy would imply thatfewer new CCGT units are needed in the system, boosting therelative importance of lignite production. The impact of macro-economic variables, however, is constrained by the need todecommission lignite plants in order to comply with EU regula-tions on emissions from large combustion plants.

In order to quantify these counterbalancing forces, we ran themodel under two alternative sets of assumptions. Taking intoaccount the optimal generation expansion plan, the total installedcapacity of thermal power plants in the interconnect powersystem in 2020 varies between 6.3 and 8 GW (Table 10). Indeed,different assumptions on the demand growth rate have dramaticimpact on the required new thermal capacity from CCGT units,burning natural gas. More specifically, in the high growth scenariotwo additional natural gas units will be required, while theoptimal generation expansion plan for the stagnation scenariorequires two fewer natural gas units.

Largely due to the change in the composition of the generationcapacity, but also due to different carbon price, the share ofnatural gas units in electricity production is boosted by 3.4% inthe high growth scenario, while in the stagnation scenario the

lignite plants take the lead (Tables 11 and 12). Under a moreconservative approach with respect to the assumption on theevolution of natural gas prices (73% per annum), the differencesin the fuel mix would be more dramatic with natural gas takingeven higher boost in the high growth scenario and harder hit inthe stagnation scenario.

It is worth noting that even with fairly small carbon price(10 h/t CO2) the utilisation of lignite plants remains low. Thisimplies that in order for lignite to retain its current competitiveadvantage, natural gas prices should remain fairly buoyant,despite the continued sluggish economic growth underlying thelow carbon price assumption in the stagnation scenario.

Spinning reserve requirements are marginally higher in thehigh growth scenario by 56 MW (þ1.3%) in case of 24 h planning-ahead period and by 84 MW (4.0%). In contrast, the difference inthe spinning reserve requirements between the stagnation andthe reference scenario is much more pronounced, reaching�970 MW (�22.9%) in case of 24 h planning-ahead period and�400 MW (�19.2%). These differences reflect both the differentlevel of installed RES capacity and different demand volumes.

The spilled RES energy is significantly higher in the stagnationscenario (Fig. 5), reaching 5.6% against 3.8% in the referencescenario and 3.5% in the high growth scenario. This result islargely due to the fact that the decommissioning plan does notdiffer among the scenario, so that in the stagnation scenario,which has significantly lower energy demand, the share ofinflexible lignite units in the energy mix is higher.

6. Discussion

The decarbonisation effort is expected to transform radicallythe Greek electricity sector. Meanwhile, significant part of theexisting plant fleet will have to be decommissioned due to old ageand lack of compliance with the tightening rules for emissionsfrom large combustion plants. In order to replace this capacity,we estimated that the Greek electricity system will needbetween two and six new CCGT plants (850–2550 MW) and onenew lignite plant (400 MW) in operation by 2020, comparedwith 2009.

Already in 2010, however, two new CCGT plants were commis-sioned and one more plant is in advanced stages of construction,while six other CCGT units (2.7 GW) have applied for connection(Regulatory Energy Agency, 2010). This implies that if all projectsgo ahead as planned, the installed capacity of the Greek systemwill substantially exceed the cost-efficient level. This finding

Table 11Sensitivity analysis results.

Variable Values Deviation a

Scenario M–S Scenario M Scenario M–HG Scenario M–S Scenario M–HG

Installed RES capacity (MW) 8785 10,328 11,231 �14.9% 8.7%

Electricity generation (GW h) 56,983 67,221 70,731 �15.2% 5.2%

Lignite share (%) 27 23 19 4.0 �3.5

Natural gas share (%) 24 30 33 �5.3 3.4

Oil share (%) 6 6 6 0.0 0.0

Utilisation rate: Lignite plants (%) 49 50 44 �0.2 �5.3

Utilisation rate: Natural gas plants (%) 57 62 59 �5.7 �3.1

a Percentage deviation from intermediate scenario for variables measured in natural units and difference from intermediate scenario for indicators expressed in %.

Table 12Spinning reserve requirements (MW) per planning-ahead period.

Planning-ahead period 2010e 2015 2020

Scenario M–S24 h 1539 2003 3257

1 h 1462 1509 1687

Scenario M24 h 1862 2435 4227

1 h 1471 1589 2087

Scenario M–HG24 h 1533 2215 4283

1 h 1445 1580 2171

Fig. 5. Spilled wind energy—sensitivity analysis.

1 Non-disclosure agreement prevents us from revealing data per plant level

for existing units.

E. Voumvoulakis et al. / Energy Policy 50 (2012) 161–173 169

validates the conclusion reached in Rampidis et al. (2010) that ifall announced investment plans are realised, the Greek electricitysystem will have excessive capacity. This does not necessarilymean, however, that the new entrants will be at a disadvantagedue to this development.

Given the need to decommission old plants, the capacity oflignite plants falls under all scenarios, while the capacity ofnatural gas units expands in all but the stagnation scenario(Table 13). Under the fuel cost and carbon price assumptions ofthe model, this leads to a sharp change of the fuel mix inelectricity generation, with lignite’s share falling from 51% to19–27% and natural gas increasing from 19% to 24–33%.

The fuel mix difference between 2009 data and 2020 estimatesis boosted by the abolishment of the grandfathering of allowancesthat will force the Greek electricity generators to include the fullcost of carbon in their bids. Under a set of reasonable fuel andprice assumptions, this might put the new CCGT units higher onthe merit order than most of the existing lignite plants. Themarginal cost differences across the generation units in the Greeksystem from 2013 onwards will be smaller than in 2009, implying

that the merit order and thus the generation fuel mix will becomemore sensitive to changes in imported fuel and carbon prices,given that the lignite costs are less volatile 1 .

In addition, the intermittency of wind speed and solar irradia-tion will impose substantial increase of the reserve requirements.Under the assumption that the mandatory day-ahead pool ismaintained and its gate closure time is not reduced below 24 h,the reserve requirements will approximately double to 3.5–4.5 GWin 2020 from a 2009 estimate of 1.9 GW (Table 11). Without newopen cycle units or CCGT of sufficient flexibility, the spinningreserves should come from introducing more plants in the day-ahead plan, each working at a lower load. Given the fact that eachunit has a technical load minimum, below which it cannot operate,the increased reserve requirements will squeeze substantially theload margin over which each plant can bid in the day-aheadmarket. As such, the competitive segment in the day-ahead marketwill cover a substantially lower share of the served load. Thus, ourstudy indicates that with the current regulatory framework thegoals of liberalisation and decarbonisation may conflict. This runsagainst the finding in Szabo and Jager-Waldau (2008) that compe-tition in the electricity market and expansion of RES have mutuallyreinforcing effects.

It is still true, however, that under alternative institutionalarrangements decarbonisation and liberalisation can be compa-tible, particularly in the longer term when the RES technologieswill become more mature. Very important step in this directionwould be the establishment of a real-time balancing market, inplace of the current balancing mechanism, where the ex-postbalancing charges are implicitly calculated from the day-aheadoffers and bids. As we showed in Section 5.2 under planning-ahead period that does not exceed one hour (approximation ofreal-time balancing market with improved prediction capabilityand flexible thermal generation units), the spinning reservesrequirements can be kept at 2.1 GW, close to today’s spinningreserves requirements that are determined the previous day.

Our analysis falls short of performing cost-benefit analysis todetermine the desirability of additional storage and interconnectioncapabilities with Greek islands and neighbouring countries. Yet, ourresults indicate that there are substantial benefits from interconnec-tion and storage, related to lower spilled energy from RES andreduced reserve requirements that must be covered by thermalgeneration plants. An issue that remains for further research concernsthe impact of large-scale RES penetration on the drivers of finalelectricity charges that the consumers will have to pay (power systemcosts, ancillary service costs, capacity assurance costs, RES supportcost, etc.).

Table 13Summary.

Variable 2009 2020

Base-line Inter-mediate Ambitious Stagnation High growth

Installed RES capacity in HIS (MW)a 1184 10,464 10,328 9963 8785 11,231

Thermal capacity in HIS (MW) 7968 7184 7184 6759 6334 8034

Lignite 4802 3548 3548 3548 3548 3548

Natural gas 2448 3636 3636 3211 2786 4486

Oil 718 0 0 0 0 0

Electricity generation in Greece (TW h) 53.5 66.1 67.2 61.4 57.0 70.7

Lignite share (%) 51 22 23 26 27 19

Natural gas share (%) 19 28 30 29 24 33

Oil share (%) 10 9 6 0 6 6

Utilisation rate: Lignite plants (%) 73 46 50 51 49 44

Utilisation rate: Natural gas plants (%) 44 59 62 63 57 59

Spinning reserve with 24 h GCT (GW)b 1.9 3.9 4.2 4.3 3.5 4.5

Spilled wind energy rate (%) o0.1b 7.6 3.8 0.5 5.6 3.5

Source: HTSO, PPC and authors’ calculations.a Does not include large hydroelectric units.b 2009 authors’ estimate.

E. Voumvoulakis et al. / Energy Policy 50 (2012) 161–173170

Another issue that is left for further research concerns thefeasibility of servicing the increased spinning reserve requirementsby the fleet in the estimated generation plan. In addition, our studysuffers from the limitations (also exhibited by the other studies on theGreek generation expansion plan) coming from not incorporating theexternal costs of fossil fuel production in the optimisation algorithm.We chose not to incorporate these costs due to the lack of consensuson the size of the environmental external costs and the lack ofavailability of monetary estimates of the remaining external effects(e.g., security of supply, employment, etc.). This implies that theestimated generation expansion plan represents the most efficientoutcome in terms of monetary cost minimisation, but it should not beinterpreted as the socially optimal solution.

In conclusion, the large-scale penetration of intermittent genera-tion presents considerable technical, economic and institutionalchallenges, which substantiate the need for a redesign of the Greekelectricity market. Given the estimated doubling of the spinningreserve requirements, Greece needs to move from balancing mechan-ism, based on day-ahead bids, to real-time balancing market thatwould reward generation flexibility in a more efficient manner. Theintroduction of electricity derivatives would help the generators, thesuppliers and large consumers to hedge some of the risk that wouldcome from more flexibility in the market. In addition, the envisagedin the Greek legislation capacity market, whose operation has beenrepeatedly postponed, will have to replace the current transitionalcapacity assurance mechanism, in order to cover efficiently theincreased cost to the thermal power producers from the reducedload factors and the more frequent shut-downs and start-ups. Lastly,as the RES technologies mature and their share in the energy mixincreases, the Greek legislators should examine the possibility ofintroducing institutional arrangements that shift some of the pricerisk from the consumers to new RES generators. Such arrangementsinclude a two-part feed-in tariff (Lesser and Su, 2010), a feed-inpremium (Couture and Gagnon, 2010) and even a provision for RESbidding in forward and real-time markets that would provideincentives for improving wind power forecasts (Botterud et al.,2010) and the choice of sites (Hiroux and Saguan, 2010).

Acknowledgements

This paper is based on the study ‘‘Impact of Large Scale RESPenetration on Electricity Generation’’ (in Greek), conducted bythe Electricity Sector Observatory of IOBE. We would like toexpress our gratitude to the Hellenic Electricity Association for

the financial support, assistance with data, access to a WASP-IVmodel of the Greek electricity system and invaluable commentsprovided for the study. We would also like to thank participantsin the RENES 2010 conference (Athens, 11/05/2010), where anearlier draft of this paper was presented, and anonymous refereesfor their comments.

Appendix A. Monte Carlo simulation of intermittent sourcesand load

The output of many types of renewable electricity generation,such as wind and solar is intermittent in nature and varies withenvironmental conditions over which the operator has no control,such as wind velocity. A Monte Carlo simulation approach isadopted in this paper in order to take this into account. Theproduction of each intermittent source is considered to be arandom variable. The simulation of intermittent sources and loadis based on historical data of hourly values of wind velocity, solarirradiation and load demand. The simulation is performed for allthe years under study and for a large number of samples foreach year.

Fig. 6 illustrates the simulation procedure. Starting at the baseyear, load and RES are simulated for a large number of samples(100), where each sample stands for 8760-hour simulations. Theshare of the RES is calculated and if it is equal to the target sharethe procedure continues to the simulation of the next year.Otherwise the RES installation parameters are adjusted and thesimulation is repeated.

Load demand simulation

Load demand is stochastic and is simulated using as basis thehourly time series of load demand for year 2009, hereafterdenoted as P0(t). In order to simulate the hourly load demand ofyears 2010 to 2020, we take into account the annual load demandforecast. The load at hour t of year y, denoted as PL

simðy,tÞ, isconsidered to be a random variable that follows the normaldistribution as indicated in Eq. (1):

PLsimðy,tÞ �NðaðyÞ � P0ðtÞ,sLÞ: ð1Þ

where a(y) is the growth rate of load demand at year y, and sL isthe standard deviation of load demand in 2009.

Fig. 6. Monte Carlo simulation procedure.

P (M

W)

V (m/s)

vcutvNvin

Pmax

Fig. 7. Wind power curve.

E. Voumvoulakis et al. / Energy Policy 50 (2012) 161–173 171

Wind power

In order to simulate the power generation of wind parks, weused hourly time series of wind velocity for each region of theinterconnected system. The wind velocity at region k during hourt is assumed to follow the normal distribution as indicated byEq. (2):

Vsimðk,tÞ �NðVðk,tÞ,sk,vÞ

Vsimðk,tÞZ0: ð2Þ

where Vsim(k,t) is the simulated value of wind velocity, V(k,t) is thevalue of the hourly time series and sk,v is the standard deviation ofwind velocity in region k in 2009.

The power generation in a specific region of the system is afunction of wind velocity and depends on the installed capacity ofwind parks in the region. Eq. (3) describes the simulated value ofwind generation

PWsimðk,y,tÞ ¼ PW

instðk,yÞ � f ðVðk,tÞÞ: ð3Þ

where PWinstðk,yÞ is the installed capacity of wind parks in year y in

region k and f(V(k,t)) is the function that gives the generatedpower per installed MW for each level of wind velocity. Thisfunction is defined in Eq. (4) and illustrated in Fig. 7.

f ðvÞ ¼

0 0rvrvin

krampUðvN�vinÞ vinovrvN

Pmax vN ovrvcut

0 vcut ov

:

8>>><>>>:

ð4Þ

PV-plant

In order to simulate the PV plants generation, we used hourlytime series of PV generation per region of the system, denoted asPPV

0 ðk,tÞ with standard deviation sk,PV.The power generation in a specific region of the system is

assumed to follow a normal distribution with a mean value equalto the value of the PV generation time series multiplied by theamount of installed MW, denoted as PPV

instðk,yÞ as shown in Eq. (5).

PPVsimðk,y,tÞ �NðPPV

instðk,yÞ � PPV0 ðk,tÞ,sk,PV Þ: ð5Þ

Small hydroelectric plants

We followed the same procedure as above in order to simulatethe generation of small hydro (SH) plants. We use hourly timeseries of small hydro generation per region of the system, denotedas PSH

0 ðk,tÞ with standard deviation sk,SH.The power generation in a specific region of the system is

assumed to follow a normal distribution with a mean value equalto the value of the SH generation time series multiplied by theamount of installed MW as shown in Eq. (6).

PSHsimðk,y,tÞ �NðPSH

instðk,yÞ � PSH0 ðk,tÞ,sk,SHÞ ð6Þ

Non-intermittent RES

The power generation from non-intermittent renewableenergy sources, such as biomass and geothermal energy, isestimated using Eq. (7).

PBIgenðk,yÞ ¼ PBI

instðk,yÞ � CFBI: ð7Þ

where CFBI is an indicative capacity factor for these technologies(set at 0.7 in this study).

Equivalent load curve

After simulating the values of load and the intermittentsources, the residual load is calculated from Eq. (8). This load isassigned to the available conventional generation.

PresLsim ðy,tÞ ¼ PL

simðy,tÞ�X13

k ¼ 1

PWsimðk,y,tÞþPPV

simðk,y,tÞþPSHsimðk,y,tÞþ

PBIgenðk,yÞ

8760

!:

ð8Þ

The residual load must be above a minimum value, due to thefact that the generation from a conventional plant cannot fallbelow that plant’s technical minimum (9).

PresLsim ðy,tÞZPminðk,yÞ: ð9Þ

The technical minimum load for the system as a wholedepends on the availability of storage facilities and the system’sinterconnections to other power systems. In case the residual loadis lower than the specified threshold, intermittent generation iscurtailed.

The simulation is executed for all the years of the period understudy and is repeated for 100 samples. The duration curve of thesimulated residual load is the equivalent load duration curve.Fig. 8 illustrates the residual load duration curve for the baseline

0

2,000

4,000

6,000

8,000

10,000

12,000

0 2,000 4,000 6,000 8,000Hours

MW

Initial Load Duration Curve Equivalent Load Duration Curve

Fig. 8. Equivalent load duration curve in the baseline scenario, 2020.

E. Voumvoulakis et al. / Energy Policy 50 (2012) 161–173172

scenario in 2020. The results are used to provide the probabilitydensity function of the residual load which is then used as aninput in the WASP-IV model of the Greek system.

Appendix B. Determination of new installed capacity fromRES

The National Action Plan of the Ministry of Environment,Energy and Climate Change has determined, for each RES tech-nology, new installed capacity requirements for the fulfilment ofnational targets. The values of installed RES capacity per region

PWinstðk,yÞ, PPV

instðk,yÞ, PSHinstðk,yÞ and PBI

instðk,yÞ are not known in

advance. In order to determine these values, we followed a simplerecursive method. Our key assumption is that the new installedcapacity for each technology in each region is proportional to theinvestment interest there, as indicated by the number of applica-

tions for production licences ðPWlic ,PBI

lic ,PSHlic ,PPV

lic Þ, submitted to the

Regulatory Authority of Greece as per 31/12/2009 (Eqs. 10–13).

PWinstðk,yÞ ¼ aðkÞUPW

lic : ð10Þ

PBIinstðk,yÞ ¼ aðkÞ � PBI

lic: ð11Þ

PSHinstðk,yÞ ¼ aðkÞ � PSH

lic : ð12Þ

PPVinstðk,yÞ ¼ aðkÞ � PPV

lic : ð13Þ

where a(k) is a coefficient of proportionality, calculated with thesimulation process, which ensures that the interim and finaltargets for RES penetration are achieved.

The installed capacity of RES in autonomous islands isassumed to be limited to the 30% of peak load demand in eachautonomous system, due to transient stability constraints(Hatziargyriou et al., 1995). For the same reasons, RES generation

in the autonomous systems, denoted as PislandsRES ðyÞ, is estimated to

be equal to 17% of total energy production.We check whether the installed capacity of RES meets the 40%

target for 2020 and the interim targets for the period 2010 to2020 by using the constraint in Eq. (14).

PL�HðyÞþPislandsRES ðyÞþ

X13

k ¼ 1

PBIgenðy,kÞþ

X13

k ¼ 1

X8760

t ¼ 1

ðPWinstðk,yÞ PW

simðk,y,tÞ

þPSHinstðk,yÞ PSH

simðk,y,tÞþPPVinstðk,yÞ PPV

simðk,y,tÞÞ

¼ GðyÞ EðyÞ: ð14Þ

where PL�H(y) is the estimated production of large hydro units foryear y, G(y) is the target value of the share of RES in the energymix for year y and E(y) is the total electric energy consumption inyear y.

Appendix C. Reserve determination with the 3 sigma method

RES increase the unpredictable power fluctuations that have tobe managed by TSOs. As a result, additional spinning reserve isdeemed necessary to balance the system. The amount of reserveneeded to handle unpredicted short term variations, either due todemand and generation prediction errors or generation failures, isworked out through analytical techniques or simulation models,both based on statistical principles (IEA Wind Task Force 25,2009). The objective is to estimate the level of spinning reserve,which is adequate to deal with almost all the unpredictedfluctuations that can be envisaged.

In order to estimate the reserve requirements for the Greeksystem, we used the 3-sigma method. According to this method,reserves are sized to cover approx. 73-times the standarddeviation of the potential uncertain fluctuations that arise fromcombining the demand prediction error and the intermittentgeneration fluctuation, plus provision for the sudden loss ofthe largest single unit (n�1 criteria, or disturbance reserve).The 3-sigma criterion ensures that the probability of inability ofthe reserves to serve unpredicted demand or supply fluctuationsis limited to less than 1%, under the assumption of normaldistribution function for demand and intermittent supply(Milligan, 2003). The size of the reserves, denoted as R, is givenby Eq. (15).

R¼ 73ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffis2

dþs2RESþs2

LU

q: ð15Þ

where sd and sRES represent the standard deviations of fluctua-tions of demand and RES generation and sLU is the capacity of thelargest unit.

In this paper, the demand and RES generation fluctuations areestimated from the series that were derived by the simulationprocedure described in Appendix A.

References

Agoris, D., Tigas, K., Giannakidis, G., Siakkis, F., Vassos, S., Vassilakos, N., Kiliasa, V.,Damassiotis, M., 2004. An analysis of the Greek energy system in view of theKyoto commitments. Energy Policy 32, 2019–2033.

Botterud, A., Wang, J., Miranda, V., Bessa, R.J., 2010. Wind power forecasting in U.S.electricity markets. The Electricity Journal 23 (3), 71–82.

Christodoulakis, N., Kalyvitis, S., 1997. The demand for energy in Greece: assessingthe effects of the community support framework 1994–1999. Energy Econom-ics 19, 393–416.

Christodoulakis, N., Kalyvitis, S., Lalas, D., Pesmajoglou, S., 2000. Forecastingenergy consumption and energy related CO2 emissions in Greece: an evalua-tion of the consequences of the community support framework II and naturalgas penetration. Energy Economics 22, 395–422.

Couture, T., Gagnon, Y., 2010. An analysis of feed-in tariff remuneration models:implications for renewable energy investment. Energy Policy 38 (2), 955–965.

Dagoumas, A., Kalaitzakis, E., Papagiannis, G.K., Dokopoulos, P.S., 2007. A postKyoto analysis of the Greek electric sector. Energy Policy 35, 1551–1563.

Donatos, G., Mergos, G., 1991. Residential demand for electricity: the case ofGreece. Energy Economics 13, 41–47.

ETSO, 2007. European Wind Integration Study. Brussels, BE.Georgakellos, D.A., 2010. Impact of a possible environmental externalities inter-

nalisation on energy prices: the case of the greenhouse gases from the Greekelectricity sector. Energy Economics 32 (1), 202–209.

Hatziargyriou, N.D., Papathanasiou, S.A., Papadopoulos, M.P., 1995. Decision treesfor fast assessment of autonomous power systems with a large penetrationfrom renewables. IEEE Transactions on Energy Conversion 10 (2), 315–325.

Hiroux, C., Saguan, M., 2010. Large-scale wind power in European electricitymarkets: time for revisiting support schemes and market designs? EnergyPolicy 38 (7), 3135–3145.

Hondroyannis, G., 2004. Estimating residential demand for electricity in Greece.Energy Economics 26, 319–334.

IEA Wind Task Force 25, 2009. Design and operation of power systems with largeamounts of wind power. Eighth International Workshop on Large ScaleIntegration of Wind Power into Power Systems as well as on TransmissionNetworks of Offshore Wind Farms, 14–15 Oct. 2009, Bremen, DE.

Iiadou, E.N., 2009. Electricity sector reform in Greece. Utilities Policy 17 (1), 76–87.Kalampalikas, N.G., Pilavachi, P.A., 2010a. A model for the development of a power

production system in Greece—Part I: Where RES do not meet EU targets.Energy Policy 38, 6514–6528.

E. Voumvoulakis et al. / Energy Policy 50 (2012) 161–173 173

Kalampalikas, N.G., Pilavachi, P.A., 2010b. A model for the development of a powerproduction system in Greece—Part II: Where RES meet EU targets. EnergyPolicy 38, 6499–6513.

Lesser, J.A., Su, X., 2010. Design of an economically efficient feed-in tariff structurefor renewable energy development. Energy Policy 36 (3), 981–990.

Milligan, M., 2003. Wind power plants and system operation in the hourly timedomain. American Wind Energy Association Wind power Conference, 18–21May 2003, Austin, TX.

Moura, P.S., de Almeida, A.T., 2010. Multi-objective optimization of a mixedrenewable system with demand-side management. Renewable and Sustain-able Energy Reviews 14 (5), 1461–1468.

Polemis, M., 2007. Modelling industrial energy demand in Greece using cointegra-tion techniques. Energy Policy 35, 4039–4050.

Poyry Energy Consulting, 2009. Impact of Intermittency. Oxford, UK.Puga, J.N., 2010. The importance of combined cycle generating plants in integrat-

ing large levels of wind power generation. The Electricity Journal 23 (7),33–44.

Rampidis, I.M., Giannakopoulos, D., Bergeles, G.C., 2010. Insight into the Greekelectric sector and energy planning with mature technologies and fueldiversification. Energy Policy 38 (8), 4076–4088.

Rapanos, V., Polemis, M., 2006. The structure of residential energy demand inGreece. Energy Policy 34, 3137–3143.

Regulatory Energy Agency, 2010. National report to the European Commission.

Athens, GR.Shayesteh, E., Yousefi, A., Moghaddam, M.P., 2010. A probabilistic risk-based

approach for spinning reserve provision using day-ahead demand responseprogram. Energy 35, 1908–1915.

Szabo, S., Jager-Waldau, A., 2008. More competition: threat or chance for financingrenewable electricity? Energy Policy 36 (4), 1436–1447.

Tolis, A.I., Rentizelas, A.A., Tatsiopoulos, I.P., 2010. Optimisation of electricityenergy markets and assessment of CO2 trading on their structure: a stochasticanalysis of the Greek power sector. Renewable and Sustainable Energy

Reviews 14, 2529–2546.UK Energy Research Centre, 2006. The costs and impacts of intermittency: an

assessment of the evidence on the costs and impacts of intermittent genera-tion on the British electricity network. London, UK.

Vassos, S., Vlachou, A., 1997. Investigating strategies to reduce CO2 emissions fromthe electricity sector: the case of Greece. Energy Policy 25, 327–336.