Generator Cycling due to High Penetrations of Wind Power

AbstractPower systems have changed considerably in recent years. The introduction of deregulationhas brought about competitive electricity markets, forcing generators to operatein a more flexible manner. Coupled with this, the rapid integration of wind powerworld-wide has introduced increased levels of variability and uncertainty into powersystem operation. This has led to generators being started up and shut down, rampedand operated at part-load levels more frequently, in order to meet an increasinglyvariable net load (load minus wind generation) and respond to unexpected net loadchanges. As base-load units are designed to achieve maximum fuel efficiency they tendto have limited operational flexibility and consequently this type of cycling operationresults in serious degradation of plant equipment through various mechanisms suchas thermal fatigue, erosion, corrosion, etc. leading to more frequent forced outagesand reduced plant lifetime. Increased costs for base-load generators will also resultfrom cycling operation, the most apparent being increased operations and maintenance(O&M) and capital costs resulting from deterioration of the components. However,fuel costs, environmental penalties and income losses will also arise. Quantifying thesecosts is challenging given the vast array of components affected and the time delay thatis typical between cycling operation occurring and the damage manifesting itself. Theuncertainty surrounding cycling costs can lead to these costs being under-estimated bygenerators, which in turn can lead to increased cycling.I

This thesis examines how the operation of base-load units, coal and combined-cyclegas turbines (CCGTs) in particular, are impacted with increasing penetrations of windgeneration on a system. The technical characteristics of these units, such as their startuptimes and contribution to system reserve requirements, are shown to influence thetype and level of cycling that will be experienced. Despite collective agreement thatmore flexible generation is needed to support the variability and uncertainty of windgeneration, it is shown here that paradoxically it is the most inflexible generation (i.ecoal plants) that are the most rewarded as wind generation increases.Having identified that CCGT units are severely impacted by increasing wind penetrationsand in many cases are forced into mid-merit operation, a novel operatingstrategy is investigated for these units. Many CCGTs include bypass stacks allowingthem to vent exhaust gas directly into the atmosphere and bypass the steam sectionof the plant entirely. Running in this open-cycle manner, CCGTs will have reducedefficiency but can start-up quickly. This thesis examines if a system with increasingwind penetration can benefit from increased flexibility when CCGT units are allowedto operate in a multi-mode regime. It is shown that such operation can improve systemreliability by increasing the sources of replacement reserve and that production frompeaking capacity is displaced, reducing the need for such units to be built.Other options which are commonly cited as improving the flexibility of power systemsinclude pumped storage, compressed air energy storage, interconnection and demandside management. Each of these can assist in balancing net load variability and so areconsidered beneficial to the integration of wind power, however typically their impacton the operation of base-load units has not been examined. This thesis investigates howvarious forms of flexibility can alleviate or aggravate cycling of base-load generation.It is found that many of these options will in fact be in competition with base-loadgeneration to provide energy and/or reserve to the system and so can actually increaseplant cycling.On the premise that penetrations of variable renewables will continue to increasefor the foreseeable future, and that cycling operation will be a growing concern forgenerators, this thesis presents a novel formulation for cycling related costs to be representedin a unit commitment algorithm. Incremental cycling costs related to start-upsII

or ramping can be represented using the new formulation and depending on the level ofknowledge that is available, the resulting cost function can be linear, piece-wise linearor step shaped. This new approach to modelling cycling costs has applications for bothlong-term planning studies and real-world scheduling models. A case study on a 20unit system was carried out and the inclusion of the new cycling cost formulation wasshown to reduce cycling operation, distribute the burden of cycling more evenly acrossthe units and reduce overall system costs relative to the case where cycling costs werenot modelled.III

ContentsAbstractPublications Arising from ThesisAcknowledgementsAcronyms and SymbolsNomenclatureIVIIVIIIXXII1 Introduction 11.1 Evolving Power Systems and the Rise of Wind Power . . . . . . . . . . . 11.2 Impact of Wind Power on System Operation . . . . . . . . . . . . . . . 41.3 Wind Power on the Irish Power System . . . . . . . . . . . . . . . . . . 71.4 Thesis Objectives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101.5 Summary of Thesis Contributions . . . . . . . . . . . . . . . . . . . . . . 111.6 Thesis Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 122 Cycling of Thermal Plant 142.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 142.2 Damage to Power Plants Due to Cycling . . . . . . . . . . . . . . . . . . 152.3 Cycling Costs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 182.4 Next Generation Thermal Plant . . . . . . . . . . . . . . . . . . . . . . . 213 Unit Commitment with High Wind Power Penetrations 223.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 223.2 The Wilmar Planning Tool . . . . . . . . . . . . . . . . . . . . . . . . . 233.2.1 The Scenario Tree Tool . . . . . . . . . . . . . . . . . . . . . . . 233.2.2 The Scheduling Model . . . . . . . . . . . . . . . . . . . . . . . . 243.3 Other Unit Commitment Models . . . . . . . . . . . . . . . . . . . . . . 283.4 The Irish 2020 Test System . . . . . . . . . . . . . . . . . . . . . . . . . 29IV

4 Cycling of Base-load Plant on the Irish Power System 334.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 334.2 Scenarios Examined . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 344.3 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 374.3.1 Increasing Wind Penetration and the Operation of Base-Load Units 374.3.2 Sensitivity Analysis . . . . . . . . . . . . . . . . . . . . . . . . . 434.3.3 Effect of Modelling Assumptions . . . . . . . . . . . . . . . . . . 494.4 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 505 Multi-mode Operation of Combined-Cycle Gas Turbines 535.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 535.2 Methodology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 565.3 Test System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 605.4 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 615.4.1 Utilization of the Multi-mode Function . . . . . . . . . . . . . . 625.4.2 Benefits Arising from Multi-mode Operation . . . . . . . . . . . 665.4.3 Sensitivity Studies . . . . . . . . . . . . . . . . . . . . . . . . . . 715.5 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 776 Power System Flexibility and the Impact on Plant Cycling 796.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 796.2 Methodology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 816.3 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 846.3.1 Impact on the Operation of Base-load Units . . . . . . . . . . . . 846.3.2 Impact on Wind Curtailment and CO 2 Emissions . . . . . . . . . 916.4 Summary of Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 926.5 Other Flexibility Options . . . . . . . . . . . . . . . . . . . . . . . . . . 936.5.1 Battery Electric Vehicles . . . . . . . . . . . . . . . . . . . . . . . 936.5.2 Maintenance Scheduling . . . . . . . . . . . . . . . . . . . . . . . 946.5.3 Control of Wind Power Output . . . . . . . . . . . . . . . . . . . 966.5.4 Market Options . . . . . . . . . . . . . . . . . . . . . . . . . . . . 967 Unit Commitment with Dynamic Cycling Costs 987.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 987.2 Formulation of Dynamic Cycling Costs . . . . . . . . . . . . . . . . . . . 997.2.1 Cycling Costs Related to Start-ups . . . . . . . . . . . . . . . . . 1007.2.2 Cycling Costs Related to Ramping . . . . . . . . . . . . . . . . . 1047.3 Model and Test System . . . . . . . . . . . . . . . . . . . . . . . . . . . 1087.4 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1137.4.1 Start-up Related Cycling Costs Results . . . . . . . . . . . . . . 1137.4.2 Ramping Related Cycling Costs Results . . . . . . . . . . . . . . 1177.4.3 Start-up and Ramping Cycling Costs Results . . . . . . . . . . . 1187.5 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1208 Conclusions 1228.1 Future Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 124V

References 127Appendix A. Probability distribution of net load ramps 136Appendix B. Cycling data for CCGT and coal units 138Appendix C. Base-load cycling with/without storage/interconnection 141Appendix D. Fuel Cost Curves 143Appendix E. Publications 145VI

Publications Arising from ThesisJournal Publications:1. Troy, N., Flynn, D., Milligan M. and O’Malley, M. “Unit Commitment withDynamic Cycling Costs”, IEEE Transactions on Power Systems, in review.2. Troy, N., Flynn, D. and O’Malley, M. “Multi-mode Operation of Combined-CycleGas Turbines with Increasing Wind Penetration”, Accepted to IEEE Transactionson Power Systems3. Troy, N., Denny, E. and O’Malley, M. “Base-load cycling on a system with significantwind penetration”, IEEE Transactions on Power Systems, vol. 25, issue2, pp. 1088 - 1097, 2010.Conference Publications:1. Troy, N. and O’Malley, M. “Multi-mode Operation of Combined-Cycle Gas Turbineswith Increasing Wind Penetration”, in Proceedings of the IEEE Power &Energy Society General Meeting, Minnesota, USA, July 2010.2. Troy, N. and Twohig, S. “Wind as a Price-Maker and Ancillary Services Providerin Competitive Electricity Markets”, in Proceedings of the IEEE Power & EnergySociety General Meeting, Minnesota, USA, July 2010.3. Troy, N., Denny, E. and O’Malley, M. “Evaluating which forms of flexibility mosteffectively reduce base-load cycling at large wind penetrations”, in Proceedingsof the 8th International Workshop on Large-Scale Integration of Wind Power,Bremen, Germany, October 2009.4. Tuohy, A., Troy, N., Gubina, A. and O’Malley, M. “Managing wind uncertaintyand variability in the Irish power system”, in Proceedings of the IEEE Power &Energy Society General Meeting, Calgary, USA, July 2009.5. Troy, N., Denny, E. and O’Malley, M. “The relationship between base-load generation,start-up costs and generator cycling”, in Proceedings of the 14th AnnualNorth American Conference of the International Association of Energy Economics,Louisiana, USA, December 2008.VII

AcknowledgementsI would like to thank everybody whose help and support contributed to this thesis, butin particular the following:My supervisor Professor Mark O’Malley for his guidance and encouragement over thepast four years. Through his hard work and efforts I have benefited from many wonderfulopportunities for which I am extremely grateful and consider myself lucky to havefound such a dynamic mentor.My co-supervisor Dr Damian Flynn for the enthusiasm and time he invested in mywork. His attention to the finest detail is exemplary and I am very thankful for theeffort he put into my thesis.Dr. Eleanor Denny for her support, insights and advice in the earlier stage of myPhD.Dr. Aidan Tuohy, to whom I am indebted for getting me up to speed with the Wilmarmodel and answering my many annoying questions even when busy writing up his ownthesis.Dr. Michael Milligan for hosting me at NREL and indeed all the other members of theGrid Integration Group. My time at NREL was both insightful and enjoyable and Iam very grateful for the opportunity.Dr. Jonathan O’Sullivan and Sonya Twohig for hosting me at EirGrid and for manyinteresting discussions during that time.Ms Magdalena Szczepanska for all her help over the past four years and for keepingthings running smoothly.All the students at the ERC who have been good fun and great friends. I look forwardto many more adventures together!VIII

My parents, grandparents, extended family and friends, for their support and for providingrelief from my academic pursuits.But most especially I’d like to thank Shane for always being kind, supportive andpatient. I could not have done it without you.IX

Acronyms and SymbolsADGT Aero-derivative gas turbineAIGS All Island Grid StudyARMA Auto-regressive moving averageBNE Best new entrantCAISO California Independent System OperatorCCGT Combined-cycle gas turbineCEMS Continuous Emissions Monitoring SchemeCHP Combined heat and powerCO Carbon monoxideCO 2 Carbon dioxideDOE Department of Energy (US)DSM Demand side managementEDUD Expected duration of unmet demandELCC Effective load carrying capabilityEPRI Electric Power Research InstituteERCOT Electric Reliability Council of TexasEU European UnionEV Electric vehicleEWEA European Wind Energy AssociationX

GADS Generating Availability Data SystemGAMS Generic Algebraic Modeling SystemGE General ElectricHRSG Heat recovery steam generatorIRRE Insufficient ramping resource expectationLOLE Loss of load expectationNEPOOL New England Power PoolNERC North American Electric Reliability CouncilOCGT Open-cycle gas turbineO&M Operations and maintenancePHEV Plug-in hybrid electric vehicleREFIT Renewable energy feed-in tariffROC Renewable Obligation CertificateROCOF Rate of change of frequencyRPS Renewable Portfolio StandardsSEM Single Electricity MarketSONI System Operator Northern IrelandSPP Southwest Power PoolSTT Scenario Tree ToolTR1 Tertiary operating reserve (Ireland)UK United KingdomUS United StatesV2G Vehicle-to-GridXI

NomenclatureChapter 3 & 5Indicesccgt CCGT unitsccgt open CCGT units in open-cycle modeg Unitsi Interval of the start-up processs Scenariost TimeParametersP mingP maxgMinimum power output for unit ’g’ (MW)Maximum power output for unit ’g’ (MW)P U (g, i) Power output for unit ’g’ at interval ’i’ of the start-up process (MW)Startfuel gUD gStart-up fuel required by unit ’g’ (GJ)Duration of start-up process for unit ’g’ (h)Binary VariablesV Onlines,t,gV Starts,t,gV Shuts,t,g0/1 variable equal to 1 if unit ’g’ is online in scenario ’s’, at time ’t’0/1 variable equal to 1 if unit ’g’ is started in scenario ’s’, at time ’t’0/1 variable equal to 1 if unit ’g’ is started in scenario ’s’, at time ’t’XII

p(s,t,g) power output for unit ’g’, in scenario ’s’, at time ’t’ (MW)Positive VariablesF uel Starts,t,gP OffgStart-up fuel used if unit ’g’ is started in scenario ’s’, at time ’t’ (GJ)Offline contribution to replacement reserve from unit ’g’ in scenario ’s’, at time’t’ (MW)Chapter 7Indices/Setst, T Time step, set of time stepsg, G Units, set of unitsi, I Interval of cycling cost function, set of intervals of cycling cost functionj, J Level of ramp, set of all ramp levelsl, L Segment of the piecewise linearisation of the variable cost function, set of allsegments of the piecewise linearisation of the variable cost functionConstantscost S gCycling cost increment for each additional startTh S g (i) ith threshold corresponding to cumulative start-upscost S g (i) Cycling cost increment for each additional start-up, while N S (t,i) < Th S (i+1)R gproduction change (MW) over time period ‘t’ deemed damaging for unit ‘g’R g (j) jth production change (MW) over time period ‘t’ deemed damaging for unit ‘g’cost R gCycling cost increment for each additional ramp > RTh R g (i) ith threshold corresponding to cumulative rampscost R g (i) Cycling cost increment for each additional ramp, while N R (t,i) < Th R (i+1)cost X gCycling cost increment for each additional bi-directional rampTh X g (i) ith threshold corresponding to cumulative bi-directional rampsXIII

cost X g (i) Cycling cost increment for each additional bi-directional ramp, while N X (t,i) R g between time t and t-1,XIV

g (t, j) equal to 1 when a unit undergoes ramp > R g (j) between time t and t-1,step R g (t, i) equal to 1 when N S (t,1) ≥ Th R (i) at time t,up g (t) equal to 1 when production at time t > production at t-1,down g (t) equal to 1 when production at time t < production at t-1,x g (t) equal to 1 when ramping switches direction between consecutive periods,step X (t, i) equal to 1 when N X (t,1) ≥ Th X (i) at time t.Positive VariablesN S g (t) Cumulative start-ups,N S g (t,i) Cumulative start-ups beyond threshold Th S (i),C S g (t) Total cycling cost attributed to start-ups,N R g (t) Cumulative ramps > R g ,N R g (t,i) Cumulative ramps > R g beyond threshold Th R (i),C R g (t) Total cycling cost attributed to ramping,N X g (t) Cumulative incidents of bi-directional ramping,N X g (t,i) Cumulative incidents of bi-directional ramping beyond threshold Th X (i),C X g (t) Total cycling cost attributed to bi-directional ramping,c p g(t) Production cost for unit ‘g’ at time ‘t’,c s g(t) Start-up fuel cost for unit ‘g’ at time ‘t’,p g (t) Output (MW) for unit ‘g’ at time ‘t’,D(t) System demand (MW) at time ‘t’,δ l (g,t) Variable used in the linearization of the variable cost function of unit ‘g’ at time‘t’, represents the l th segment (MW).XV

CHAPTER 1Introduction1.1 Evolving Power Systems and the Rise of Wind PowerRECENT years have seen the power generation sector undergo significant changes.Traditionally electricity systems were operated by vertically integrated monopolieswhose main aim was to meet the demand as opposed to minimising cost (Narula et al.,2002). However, by the 1980s deregulation and unbundling of utilities was seen as ameans of improving economic performance. In 1982 Chile kick-started electricity deregulationby passing a law which allowed large consumers of electricity to choose theirretailer and negotiate their prices freely. In 1990 the United Kingdom (UK) governmentprivatised the electricity supply industry, which led to other Commonwealth countries,notably New Zealand and Australia, also pursuing deregulation. The European Union(EU) directive 96/92 introduced in 1996 required Member States to create competitiveelectricity markets, whilst by the late 1990s many states in the United States (US) werealso moving towards deregulation (Al-Sunaidy and Green, 2006).1

Chapter 1. Introduction 2Figure 1.1: Increased cycling due to introduction of electricity market in Ontario(APPrO, 2006)In the resulting competitive and volatile marketplaces that were created, generatorswhich had previously operated as base-load plant were often forced into flexible operation(Kitto Jr et al., 1996; Narula et al., 2002). Figure 1.1 which illustrates increasedplant start-ups following the introduction of a competitive electricity market in Ontarioprovides an example of how greater flexibility is required in competitive markets. In acompetitive marketplace, energy traders or suppliers, seeking to maximise profitability,will offer generation into power, exchange and ancillary service markets requiring unitsto have short start-up times and good cycling capabilities. The ability to operate flexiblycan bring considerable economic advantage for generators, as they have increasedopportunities to earn revenue from the market, such as through hourly and seasonalmarket arbitrage or peak shaving for example (Balling and Hofmann, 2007). However,the financial pressure to reduce capital costs in a competitive market can often lead topower generating companies purchasing plants with cheaper and consequently poorerperforming components, which are more susceptible to cycling related wear and tear.As such, older coal plants have been found to be more rugged and cost effective to cyclecompared to newer combined-cycle units (Lefton and Besuner, 2006).Meanwhile, the acceptance that anthropogenic greenhouse gas emissions are resultingin climate change has led to the introduction of energy policies seeking to reduce the

Chapter 1. Introduction 3environmental impact of electricity generation. Supporting renewable energy sourcesand energy efficiency measures has been identified as vital to achieving emission reductions.Coupled with this, rising fossil fuel prices and instability in countries where fossilfuels are sourced has led to widespread backing of renewables as a means of improvingsecurity of supply and reducing exposure to fossil fuel price volatility. In 2008 the EUimposed demanding climate and energy targets known as the ‘20-20-20’ targets whichare to be met by 2020. These aim to reduce EU greenhouse gas emissions by 20% below1990 levels, supply 20% of energy consumption from renewable energy sources andreduce primary energy consumption by 20% through energy efficiency measures (EU,2008). Although the US has no comprehensive long-term energy policy, initiatives suchas Renewable Portfolio Standards (RPSs) and Renewable Energy Certificates (RECs)have been taken at a state level to increase the use of renewable energy (Black & Veatch,2011). Thus 28 out of 50 US states have set compulsory targets seeking renewable energypenetrations up to 40%, with a further 5 states having voluntary targets (DOE,2009).Wind power, now a proven and mature technology which offers near-zero emissionsand operating costs, represents a feasible means of meeting emissions and renewableenergy targets and consequently has experienced rapid growth over the past decade.The cumulative installed wind power capacity in the U.S stood at 41.4 GW in the firstquarter of 2011 (AWEA, 2011b), just behind China which is set to reach 58 GW bythe end of 2011 (Castano, 2011). In Europe the total installed wind capacity exceeds84 GW, with countries such as Germany, Spain and Denmark representing the largestshares. In terms of energy penetration however, Denmark, Portugal, Spain and Irelandlead the way, as seen for 2009 in Figure 1.2 (IEA, 2010).In spite of the unprecedented economic downturn, the annual growth rate for windpower has remained high with the installed wind power capacity in the EU increasingby 12.4% in 2010, compared with 15% in the US (AWEA, 2011a; EWEA, 2011c). Thisrapid pace of wind power installation is set to continue through the coming years, aswith the majority of hydro resources already exploited, wind power (and solar power insome countries) represents the most scalable and competitive means of achieving 2020

Chapter 1. Introduction 4Figure 1.2: Top 10 highest wind penetrations, as % of electricity consumption, in EUcountries (IEA, 2010)targets. Up to now wind power has been supported by some form of subsidy such afeed-in tariff or renewable certification scheme. However, with growing sales and largerand more efficient turbines the cost of wind power is, in some countries (Brazil, Sweden,Mexico and US) similar to the cost per MWh of coal generation and consequently itmay be possible to phase out subsidies over the coming years without hampering windpower development (Bloomberg, 2011). EWEA (European Wind Energy Association)predicts between 230 and 265 GW of installed wind power in Europe for the year 2020(40 GW of which is assumed to be offshore wind) which would supply between between14.4% and 16.7% of the total electricity demand (EWEA, 2011b).1.2 Impact of Wind Power on System OperationAs higher penetrations of wind power are achieved, power system operation becomesincreasingly complex due to the variable and unpredictable nature of wind power. Traditionallysystem demand has been largely predictable as demand profiles follow daily,weekly and seasonal patterns, allowing generation to be efficiently committed. Windpower however introduces another element of uncertainty and thus systems with significantlevels of wind need to utilise wind forecasts when committing and dispatching

Chapter 1. Introduction 5generation. Approaches to wind forecasting can be categorised as physical or statistical,with modern forecasting systems employing a combination of the two. Physicalapproaches, namely weather prediction models, which are typically used for horizonsof 6 to 72 hours, utilise data such as land and sea surface temperatures to physicallymodel atmospheric dynamics. Statistical approaches transform meteorological predictionsinto wind generation (often using artificial-intelligence based models) and arefound to give better accuracy for horizons up to 6 hours (Monteiro et al., 2009). Thedesire of system operators for information regarding the reliability of forecast has led toensemble or probabilistic forecasts becoming popular. Ensemble forecasting producesmultiple forecasts, by varying the input parameters or by using multiple weather predictionmodels, to generate a probability density function of the most likely outcome(Möhrlen et al., 2007). Wind power forecast error however, increases with the forecasthorizon and even when these state-of-the-art methods of forecasting are employed, theday-ahead wind forecast error (root mean square error) for a region can be 8-12% of thetotal wind capacity as reported in Siebert (2008), which can result in thermal units beingover- and under-committed (Ummels et al., 2007). Thus power systems with largewind power capacities will need to re-evaluate commitment decisions on a continualbasis as more up-to-date wind forecasts become available. The unpredictable natureof wind power also requires conventional plant to carry additional reserves in order tomaintain system reliability, should an unexpected drop in wind power occur. Many approacheshave been proposed to determine how much wind power increases the reserverequirement on a given system and it has often been found that the increased reserverequirements represents only a small percentage of the wind power output (Dany, 2001;Doherty and O’Malley, 2005; Holttinen et al., 2008; Holttinen, 2005)The variable nature of wind power will increase variability in net load (load minuswind generation), which must be met by conventional generation on the system, resultingin a greater demand for operational flexibility from these units (Ummels et al., 2007;Holttinen, 2005). Expected or unexpected reductions in net load, which can arise dueto declining wind power output, will force conventional plant to ramp up their output,or if sufficient ramping capability is not available, fast-starting units will need to come

Chapter 1. Introduction 6online. Periods of low demand coinciding with high wind power output can lead toconventional plant being shut down, a problem which has been exacerbated of late dueto a reduction in demand as a result of widespread economic recession (Axford, 2009).The culmination of adding more variability and unpredictability to a power system isthat thermal units will undergo increased start-ups, ramping and periods of operationat low load levels, collectively termed “cycling” (Braun, 2004; Göransson and Johnsson,2009; Holttinen and Pedersen, 2003; Meibom et al., 2009). Furthermore, in some systemswind is allowed to self-dispatch, so the forecast output from wind farms is notincluded in the day-ahead schedule. This can lead to increased transmission constraintswhich will further intensify plant cycling (GE, 2005).Many systems are currently experiencing increased plant cycling as a result of windpower and wind integration studies are predicting this problem to worsen. The SouthwestPower Pool (SPP) wind integration study noted that in order to accommodatehigher wind penetration levels more operational flexibility (i.e. more start-ups andcycling of units) would be required and this would increase as the forecast error increases(Charles River Associates, 2010). The Californian Independent System Operator’s(CAISOs) renewables integration study had similar findings, but with combinedcyclegas turbine (CCGT) units specifically identified as undergoing increased cycling.Relative to a 2012 reference case, CCGT plant start-ups increased by 35% with 20% renewableson the system. Both the ‘NYISO 2010 Wind Generation Study’ and the ‘NewEngland Wind Integration Study’ also found that the operation of CCGTs was significantlyimpacted by an increased penetration of wind power (NYISO, 2010; GE, 2010).The Nova Scotia wind integration study predicted that start-ups for large thermal unitswould be significantly increased as wind penetrations increased and acknowledged thatthe cost impact of this was not fully understood (Hatch, 2008), while Xcel Energy arecurrently experiencing cycling of their coal fleet due to high wind penetrations in Colorado.In Göransson and Johnsson (2009), which studied the power system of WesternDenmark, the capacity factor for units with low start-up and turn-down performanceand high minimum load levels (i.e. base-load units) were found to be the most significantlyimpacted by wind power, while Oswald et al. (2008) found that more ramping

Chapter 1. Introduction 7will be required from fossil fuel plants on the British system to maintain the powerbalance.Wind power will also impact a system’s dynamic performance. As wind power willtend to displace conventional generation, it will also displace the inertial response providedby these units, which is vital to maintain system security when faults or outagesoccur. In addition, wind turbines supply asynchronous power to the system whichcan impact the system’s voltage stability. However, it is possible to implement controlfeatures to emulate inertial response and mitigate the impact on voltage stability andthese will be necessary in order to increase the upper limit to the maximum penetrationof wind power on a system.1.3 Wind Power on the Irish Power SystemSituated on the western edge of Europe, Ireland is well positioned to benefit from strongAtlantic winds and consequently has one of the best wind resources in the world (SEAI,2010), as seen in Figure 1.3. The current installed wind capacity in Ireland stands atover 1.8 GW, which provides in excess of 10% of the electrical energy demand andanother 4 GW of proposed wind capacity is in various stages of planning. The growthof wind power in Ireland is also supported by ambitious Government targets (40% ofall electricity consumption to come from renewables by 2020) and competitive feedintariffs (REFIT in Republic of Ireland and ROCs in Northern Ireland). REFIT(Renewable Energy Feed-In Tariff) is guaranteed for up to 15 years (but not to extendbeyond 2024) and is paid to suppliers to encourage them to enter into power purchaseagreements with wind generators. The REFIT is linked to the Best New Entrant (BNE)generator and has averaged at e57/MWh for large scale wind and e59/MWh for smallscale wind in previous years. The ROCs (Renewable Obligation Certificate) schemein Northern Ireland is somewhat different in that an obligation to purchase renewablegeneration is mandated on suppliers.The Irish system is small and relatively electrically isolated: a 500 MW interconnec-

Chapter 1. Introduction 8Figure 1.3: Wind resource in Europe (Risø National Laboratory, 1989)tor is in place linking Northern Ireland and Scotland, however, trading arrangementslimit exports from Ireland to Great Britain to a maximum of 70 MW and with powerexchanges set one month ahead of time, Ireland infrequently exports power. Therefore,variations in power output and high penetrations of wind generation are manageddomestically by conventional generation rather than by exchanges to Great Britain.As such, record high instantaneous wind penetrations, in excess of 50%, have beenexperienced on the Irish system, as seen in Table 1.1. This is resulting in increasedcycling of conventional generation on the Irish system as seen in Table 1.2, which comparesthe annual number of plant start-ups for three CCGT units in 2008 and 2010.Consequently the Market Monitoring Unit (MMU) within the Energy Regulatory Authoritieshas identified power plant cycling as one of the foremost concerns of thermalpower generators operating in SEM (Single Electricity Market), the electricity marketof the Republic and Northern Ireland (MMU, 2010). However, MMU (2010) attributesthe intense plant cycling that some plants are experiencing to the introduction of SEM

Chapter 1. Introduction 9and the subsequent increase in competition rather than the increasing wind powerpenetration.Table 1.1: Wind penetration on the Republic of Ireland and Northern Ireland systemsRepublic of IrelandNorthern IrelandInstalled Wind (MW) 1455 355Maximum Output (MW) 1323 320Maximum Energy Penetration (%) 52.3 50Maximum Daily Energy Penetration (%) 37 29Table 1.2: Annual plant start-upsUnit 2008 2010Huntstown 1 23 63Tynagh 27 45Dublin Bay Power 7 37In anticipation of the challenges facing power system operation with such high windpenetrations, the Irish Governments commissioned a study entitled the ‘All IslandGrid Study’ (AIGS), published in 2008, to examine the ability of the 2020 Irish powersystem to handle various amounts of electricity from renewable sources. Various levelsof installed wind capacity ranging from 2000 MW to 8000 MW were examined in thisstudy, with an assumed peak demand of 9.6 GW assumed.The study was dividedinto several workstreams, the most relevant to this work being ‘Workstream 2B’, whichutilized a stochastic unit commitment and economic dispatch model to examine systemoperation under the various renewable scenarios (AIGS, 2008). The key results of thisworkstream, which are particularly relevant to the work of this thesis are as follows:“With increasing wind power capacity installed, extreme values and thestandard deviation of the variation of the net load (load minus windpower production in the actual hour) increases as well.Hence, thepower plant portfolio has to show enough flexible units (for examplewith sufficient ramp up and down rates as well as low start-up times)to be able to follow the net load.”“Generally, the bigger part of the electricity production in the All

Chapter 1. Introduction 10Island power system from conventional power plants is borne by coalfired plants and newer CCGTs. With increasing wind power capacityinstalled, the production and capacity factors of these units tends tobe decreased.... Coal fired units and newer CCGTs have a relative lownumber of start-ups and high number of online hours. The number ofstart-ups of these units tends to be increased with increasing wind powercapacity installed.”Following on from the All Island Grid Study, the tranmsission system operators ofNorthern Ireland (SONI) and the Republic of Ireland (EirGrid) conducted a comprehensivestudy to better understand the technical and operational implications associatedwith high shares of renewable energy called the ‘All Island TSO Facilitation ofRenewbles Studies’, which identified two key limitations to wind power penetration,namely (i) frequency stability after a loss of generation and (ii) frequency and transientstability after severe network faults. This study suggested that the maximumamount of ‘inertialess power’ (wind power and interconnector imports) that the systemcould cope with lies between 60% and 80%, but could be as low as 50% unless ROCOFrelays on distribution connected wind farms were disabled. Nonetheless, the studyfound that the limitations for instantaneous wind penetration did not fundamentallyconflict with the 2020 policy targets aiming at 40% electricity from renewables by 2020(EirGrid and SONI, 2010a).1.4 Thesis ObjectivesThe main objective of this research has been to investigate how the operation of thermalplant will be impacted by high penetrations of wind generation on a power system.Base-load coal and CCGT units in particular are examined, as these units, having beendesigned for maximum fuel efficiency, tend to have limited operational flexibility. Assuch, when subjected to cycling operation these units can accrue large levels of damageto plant components, leading to increased maintenance requirements and forced outagerates. In examining the operational impacts of high wind power penetrations on CCGT

Chapter 1. Introduction 11units, a novel operating strategy for these units was identified, which involved allowingCCGTs to switch between combined- and open-cycle mode when economically optimal.The potential benefits and impacts of this new multi-mode strategy are investigated inthis research.In an effort to improve power system flexibility and support integration of variablerenewable generation, various flexibility options such as storage, interconnection anddemand side management are commonly put forward in the literature. Analysis of theseoptions is typically concerned with their profitability in a system or their impact onsystem production costs, wind curtailment or emissions, while the impact on base-loadgeneration is typically over-looked. This research investigates how incorporating suchflexibility options (and others) into a power system will impact cycling of base-loadunits in a high wind power scenario.Finally, having identified that the operation of base-load plant will be significantlyimpacted as wind power penetrations increase, this research develops a unit commitmentformulation to allow cycling related costs to be modelled in a dynamic manner.The impact of accounting for cycling costs in a dynamic manner on plant dispatch isevaluated.1.5 Summary of Thesis ContributionsThe novel contributions emanating from this thesis can be categorised as (i) the identificationand investigation of a new operating strategy for CCGT units in a high windpower scenario, (ii) the investigation of how various power system sources of flexibilitywill impact the operation of base-load plant and (iii) the development of a new unitcommitment formulation to allow cycling costs to be modelled dynamically, such thatthey accumulate over time based on plant operation, reflecting increased wear to plantcomponents and reduced plant life-time.Examining the potential for running CCGT units in open-cycle mode, as well ascombined-cycle mode, revealed that a system can benefit from the additional fast-

Chapter 1. Introduction 12starting capacity. The increased replacement (non-spinning) reserve availability fromCCGT units in open-cycle mode also results in increased system security. Furthermore,open-cycle operation of CCGT units will displace production from conventional peakingunits, reducing the need for such units to be built and thus indicating a societal benefit.Sensitivity studies revealed how the usage of this multi-mode function will be dependenton the underlying level of flexibility present in the system. Optimizing the systemstochastically or allowing intra-day trading on interconnectors reduced the need forflexibility to be extracted from generators and consequently resulted in less frequentdeployment of the multi-mode function.The impact of various sources of power system flexibility, such as storage, interconnectionor demand side management, on the operation of base-load plant has beenexamined in this thesis. A side-by-side comparison reveals which are effective at reducingplant cycling, or alternatively which will aggravate plant cycling, in a high windpower context. The results are somewhat surprising as it was found that many ofthese options will in fact be in competition with base-load generation to provide energyand/or reserve to the system and so actually increase plant cycling.A novel unit commitment formulation was developed which utilises binary variablesto incur a dynamic incremental cost when cycling operation occurs. The types of operationwhich elicit a cycling related cost can be plant start-ups or ramping. The cyclingcost accumulates in tandem with plant operation such that it influences the dispatchdecisions. This formulation has particular applications for long term studies, such aswind integration studies, as it can reflect the depreciation of a plant and potentiallyshow how the merit order of generation can be altered over time. A case study inwhich this new formulation was implemented revealed that by modelling cycling costsdynamically, the burden of cycling operation will, over time, be distributed more evenlyacross the fleet of generators.1.6 Thesis OverviewThe remainder of this thesis is organised as follows:

Chapter 1. Introduction 13• Chapter 2 describes the effects of cycling operation on plant equipment and thedamage mechanisms involved. It describes the cost components which make upthe total cost of cycling a unit and the difficulties in calculating these costs.Various approaches which have been used to approximate these costs are alsodescribed.• Chapter 3 describes the stochastic unit commitment and economic dispatch modellingtool that was used in this thesis. A detailed description of the test systemis also provided.• Chapter 4 examines how the operation of base-load units, coal and CCGT unitsspecifically, will be impacted with an increasing wind penetration. Sensitivitiesare also conducted to examine the level of cycling these units would undergo inthe absence of pumped storage or interconnection on the system.• Chapter 5 examines the potential for multi-mode operation of CCGT units undervarious wind scenarios to determine if this new mode of operation can deliver benefitsto the power system, via increased flexibility, or the generators themselves,via increased generation opportunities.• Chapter 6 examines how various flexibility options, namely pumped storage, interconnection,demand side management (DSM), multi-mode operation of CCGTsand reduced minimum generating levels impact the operation of base-load units.A side-by-side comparison of these options reveals which are the most effectiveat reducing cycling of base-load plant.• Chapter 7 presents a novel formulation for modelling the cycling costs in a dynamicmanner within a Mixed Integer Programming (MIP) unit commitmentmodel. The formulation can be used to implement incrementing cycling costs forstarts or ramps for linear, piece-wise linear or step cycling cost functions.• The thesis is concluded in Chapter 8.

CHAPTER 2Cycling of Thermal Plant2.1 IntroductionAS discussed in Chapter 1, the increased variability and uncertainty that ariseswhen wind power is integrated into a power system can lead to more flexibleoperation or ‘cycling’ being demanded from conventional plants. In addition to windpower, the competitive markets in which these units operate are also a significant driverof plant cycling as generators are forced into more market-orientated, flexible operationto increase profits, while at the same time maintenance intervals are often lengthenedin order to minimize downtime and costs. An overcapacity of generation on a systemcan also exacerbate plant cycling as less efficient plant may be prematurely forced downthe merit order.Thermal plant can be broadly categorised as base-load, mid-merit or peaking. Midmeritunits follow the daily demand profile and shut down nightly whilst peaking units14

Chapter 2. Cycling of Thermal Plant 15are used to meet the extreme peaks in demand. Base-load thermal units, typically coal,Combined-Cycle Gas Turbine (CCGT) or nuclear, are those units which traditionallyrun on a continuous basis, at maximum efficiency, to supply the base electricity demandand therefore tend to have minimal operational flexibility. As such, the rapid changes intemperatures and pressures that occur during cycling operation will result in accelerateddeterioration of these units’ components through various degeneration mechanisms suchas fatigue, erosion, corrosion, etc. This in turn will lead to more frequent forced outages,reduced plant lifetime and significant costs for these units. As illustrated in Figure 2.1,the damage incurred from cycling operation is related to the temperature transientsin the plant’s components, with online ramping being the least damaging and coldstart-ups the most damaging.Figure 2.1: Cycling damage increases as plant temperature decreases (Lefton, 2004)This chapter discusses some of the common wear-and-tear effects that plants willexperience when undertaking cycling operation. The various cost implications of cyclingoperation are identified and the approaches used to quantify these costs are examined.2.2 Damage to Power Plants Due to CyclingFatigue damage is the most common problem for cycling units (EPRI, 2001b). Fatigueis caused by repeated exposure to large temperature and pressure transients, typicalof cycling operation (Lefton et al., 1997), and manifests as cracking or mechanical failureof structures (EPRI, 2001b). Traditionally, base-load units ran uninterrupted atfull production and as such were designed to operate under creep conditions (constant

Chapter 2. Cycling of Thermal Plant 16stress), with older design codes neglecting to consider fatigue (fluctuating stress) as adamage mechanism (EPRI, 2001b). Creep and fatigue can interact in a synergisticmanner in that creep will reduce fatigue life and likewise fatigue reduces creep life, asdepicted in Figure 2.2 (EPRI, 2001b). Therefore when a base-load unit which has beenoperating under creep conditions, switches to cycling operation, the creep-fatigue interactionrenders the unit highly susceptible to component failure (Lefton et al., 1995,1997). Creep-fatigue interaction is a particular concern for components such as superheater,reheater and economizer headers (MMU, 2010).Figure 2.2: Creep-fatigue interaction (EPRI, 2001b)Thick-walled components such as boilers, which are necessary to withstand theextreme temperature and pressure associated with base-load operation, can developthrough-wall temperature differences during cyclic operation. This results in differentialthermal expansion and ultimately places the component under high stress,causing cracks to initiate and grow (EPRI, 2001b). An example of this is shown inFigure 2.3. Rapid temperature transients will also cause differential thermal expansionand fatigue issues in components such as header ligaments or boiler tube ties (EPRI,2001b). Peak stresses typically occur in regions of discontinuity (Brown, 1994), andtherefore welded joints are highly stressed locations (King, 1996).Expansion related issues can also arise due to cycling operation. Thin-walled com-

Chapter 2. Cycling of Thermal Plant 17Figure 2.3: Cracking seen from inside economizer header (King, 1996)ponents, heat recovery steam generator (HRSG) ducts for example, will heat up rapidlyduring plant start-up, whilst the supporting steelwork remains cold, resulting in differentialthermal expansion and consequently high stress (Brown, 1994). Likewise, onstart-up, a typical large boiler will expand downward from its roof support by 250 mmwhich must be supported by the boiler support framework. If start-ups are occurringon a regular basis it can lead to failure of the boiler support framweork (MMU, 2010).Mechanical fatigue is also common during turbine run-up, when the rotor passesthrough a series of critical speeds where vibration levels are increased significantly. Repeatedstart-ups can subject components such as turbine blades to high cycle fatiguelevels (MMU, 2010).Thermal shocking of economizer headers occurs when cold feedwater is introducedto warm headers when a unit is re-starting following an overnight shut-down, for example(King, 1996), or alternatively when hot steam is admitted to cold superheaterheaders (EPRI, 2001b). If this is occurring on a regular basis it will lead to internalfatigue cracking (King, 1996). This is irreparable and must be monitored constantlyfor propagation. Start-ups and shut-downs can also cause oxide scales that have accumulatedin steam-side equipment to spall due to the differences in the coefficients ofthermal expansion between the oxide and the metal. The hard oxide particles becomeentrained in the steam and are carried through to the turbine causing erosion of theturbine blades (French, 1993).Increased frequency of shutdowns can contribute to infiltration of dissolved oxy-

Chapter 2. Cycling of Thermal Plant 18gen and other non-condensible gases, which will also lead to higher levels of erosionand corrosion. This can occur in cycling units when the condenser vacuum is notmaintained sufficiently during offline periods. In addition as a plant goes through variousmodes of operation and cycles, contaminant can be disturbed and disseminatedthroughout the steam-condensate cycle. Thus cycling units will need to employ continuouswater chemistry monitoring (Energy-Tech, 2004).Fatigue stresses during start-up and shutdown can also result in cracking of electricalequipment such as copper turns, as shown in Figure 2.4, and the resulting arcingand burning can cause short-circuits (Moore, 2006). Coils with shorted turns operateat lower temperatures than regular coils and the resulting temperature differencecan give rise to rotor bowing. This will cause unbalanced magnetic forces giving riseto rotor vibration. If the problem becomes severe enough forced outages can occur(Albright et al., 1999).Figure 2.4: Cracked copper turn and rotor bowing (Moore, 2006)2.3 Cycling CostsAny power generating company seeking to maintain profitable operation desires toknow the cost impact of cycling operation for their fleet of generators. However, quantifying,or even estimating, the magnitude of these cycling costs is challenging giventhe extensive range of components affected by cycling, as discussed in Section 2.2. Inaddition, the damage caused by cycling may not be immediately apparent and often itcan be several years before it manifests itself. Studies by Aptech Engineering Inc. (nowIntertek Aptech) suggest that it can take from 1 to 7 years for an increase in the failurerate to become evident after switching from base-load to cycling operation Lefton et al.

Chapter 2. Cycling of Thermal Plant 19(1998). The challenge of attributing costs to cycling operation is complicated furtherby the fact that normal base-load operation also results in some degree of damage toa units components and identifying the damage due to cycling from that associatedwith normal operation is also problematic. Considering these difficulties Aptech haveconcluded that utilities typically underestimate cycling costs by a factor of 3 to 30Lefton et al. (1998).Research related to the cost of generation cycling has been led by EPRI (ElectricPower Research Institute) and Aptech and the approaches employed can be categorizedas top-down (statistical analysis) or bottom-up (component modelling). EPRI carriedout a top-down study as part of its ‘Cycling Impacts Program’ which utilized multivariateregression models to analyze the operating regimes of 158 units from NERC (NorthAmerican Electric Reliability Corporation) GADS (Generating Availability Data System)and CEMS (Continuous Emission Monitoring) data, in an attempt to identifypatterns relating operation to capital expenditure. However, the inconsistency in accountingpractices between individual units complicated the modelling process and nocorrelation was found (EPRI, 2001a, 2002). Aptech employ a combination of top-downmodels based on historical operations, forced outage and cost data, as well as bottomupmethods which calculate operational stresses and the life expenditure of criticalcomponents using physical models fine-tuned with real plant data, in order to determinecycling costs for individual generating units (Lefton, 2004). Aptech have analyzedcycling costs for over 300 generating units and found that the cost of cycling a conventionalfossil-fired power plant can range from $2,500-500,000 per start/stop cycledepending on unit age, operating history and design features, and are often grosslyunderestimated by utilities (Lefton, 2004; Lefton et al., 1998). Babcock Energy Ltd.also developed a methodology for determining the long-term damage that arises fromtwo-shift operation in order to optimize operating procedures and minimize damage.This involved identifying components most susceptible to creep-fatigue damage usingdata from thermocouples and modelling these components using finite elements so thatoperational events could be related to induced stresses (Brown, 1994).The factors which contribute to the total cost of cycling are: (i) increased fuel con-

Chapter 2. Cycling of Thermal Plant 20Figure 2.5: Impact of cycling on forced outage rate (Lefton, 2011)sumption due to increased plant start-ups and operation at part-load levels (and thereforereduced efficiency), (ii) increased fuel consumption due to loss of plant efficiencyarising from increased wear to components, (iii) increased operations and maintenance(O&M) costs due to increased wear-and-tear to plant components, (iv) increased capitalcosts resulting from component failures, (v) increased environmental costs resultingfrom increased emissions, and (vi) loss of income due to longer and more frequent forcedoutages. Figure 2.5, provides an example of how the forced outage rate of a plant canincrease as a result of cycling operation, however, capital expenditure on plant upgradescan help combat this. Of the studies undertaken to date, the magnitude of these cyclingcosts has been significant. For example, a recent study by Aptech on Excel Energy’sHarrington coal plant suggested that for each additional hot start the unit performed,the maintenance related cycling costs the unit would incur were $87k, more than 5times greater than the cost of the start-up fuel consumed (Xcel Energy, 2010). Thiswould indicate the importance of having a good understanding of cycling costs in orderto maintain profitable operation in the long term.However, in reality generators will often under-value these costs in order to keeptheir short-run costs down in a competitive marketplace, the consequence of which isthat the generator will subsequently be scheduled to cycle more often. Or in somesituations generators will take advantage of the uncertainty surrounding these cycling

Chapter 2. Cycling of Thermal Plant 21costs in order to exercise market power. For example, a generator may increase its startupcosts excessively in order to avoid shut-down, although this strategy may result inthem being left offline following a trip or scheduled shut-down because of their excessivestart-up cost. In any case in most markets at present it is unclear how these costs shouldbe represented in a generator’s bid. Generators in SEM, the Irish electricity market,are directed to include cycling costs in their start-up costs, however cycling costs couldalso be included in shut-down, no-load or energy costs, or even defined as a new marketproduct such as ramping costs (Flynn et al., 2000).2.4 Next Generation Thermal PlantWith increasing penetrations of variable renewables and competitive electricity marketsbecoming the norm worldwide, power plant manufacturers are recognising the need forgreater operational flexibility (Probert, 2011). Siemens, for example, have outlinedareas where CCGT plant can be upgraded with new features such as a stress andfatigue monitoring system for the HRSG, a piping warm-up system, attemperatorsin the steam lines to maintain required temperatures in order to make them morecapable of frequent cycling (Siemens, 2008b). General Electric (GE) meanwhile havelaunched their ‘FlexEfficiency CCGT’ which offers faster ramp rates, shorter start-uptimes, lower turndown and fuel flexibility whilst achieving an efficiency of 61%. Nextgeneration thermal plant can also avail of new materials which have been developed suchas the high strength P91 steel which allows for high-pressure components to be madethinner (EPRI, 2001b). Thinner components will reach thermal equilibrium quickerand therefore are less susceptible to cracking. Improvements in instrumentation willalso allow for easier start-up and shut-downs and part-load operation (Energy-Tech,2004). Online monitoring systems can help to protect critical components from thermalstresses. However, although next generation thermal plant may be more suited tocycling operation, current generation will still be in operation for decades more. Thuscycling poses serious difficulties for generators seeking to remain in profitable operationand system operators who must maintain a stable system in spite of increasing forcedoutages.

CHAPTER 3Unit Commitment with High Wind Power Penetrations3.1 IntroductionPRIOR to the large-scale deployment of renewables, uncertainty in power systemswas limited to load forecast error and the unplanned outages of generators ortransmission lines. In order to maintain a secure system, adequate levels of spinningand non-spinning reserve were maintained to cover this error. Incorporating variablerenewable generation adds an additional source of uncertainty given the unpredictablenature of renewable power sources. With low levels of renewables on power systems,additional reserve is needed to cover the additional uncertainty associated with renewables.However, as the penetration of renewables grows, it becomes increasingly inefficientto rely on reserves alone to cover the uncertainty related to renewables. Rathermore robust schedules are required through stochastic scheduling, which considers multiplescenarios corresponding to multiple values of the stochastic variable, in this casethe power output from the renewable generation (Monteiro et al., 2009). In addition,22

Chapter 3. Unit Commitment with High Wind Penetration 23to make the most efficient use of the renewable generation, forecasts need to be utilized.As the accuracy of these forecasts increases as the forecast horizon decreases,it is important that updated forecasts are used to update the commitment decisionsthrough a rolling unit commitment mechanism (Kiviluoma and Meibom, 2011). Thiscan in turn lead to a reduced reserve requirement (AIGS, 2008).3.2 The Wilmar Planning ToolThe Wilmar Planning Tool is an output of a collaborative research effort supported bythe European Commission to develop a tool to analyse the integration of wind power inlarge liberalised electricity systems (Meibom, 2006). The original model was developedfor two power pools: NordPool and the European Power Exchange, (i.e. Germany,Denmark, Norway, Sweden and Finland). It was later adapted to the Irish systemas part of the All Island Grid Study (AIGS, 2008; Meibom et al., 2011; Tuohy et al.,2009). Wilmar is an advanced stochastic, mixed integer unit commitment and economicdispatch model, the main functionality of which is embedded in the Scenario Tree Tooland the Scheduling Model.3.2.1 The Scenario Tree ToolThe Scenario Tree Tool (STT) generates scenarios trees which feed into the SchedulingModel. Each branch of the scenario tree represents a realistic forecast scenario of load,wind power output and demand for replacement reserve (activation time > 5 minutes).The STT also produces a forced outage time series for each generating unit.The STT utilizes knowledge of historical wind speed forecast accuracy and knowledgeof the correlation between wind speed forecast errors in neighbouring areas, as well ashistorical load data and load forecasts, to identify an Auto Regressive Moving Average(ARMA) series, based on the methods described in (Söder, 2004). The parameters ofthe ARMA series are determined by minimizing the difference between the standard

Chapter 3. Unit Commitment with High Wind Penetration 24deviation of the historical forecast error and the standard deviation of the forecast errorproduced by the ARMA series for each hour. The ARMA series is used to simulate loadand wind speed forecast errors for various time horizons. These simulated load and windspeed forecast errors are paired in a random way before a scenario reduction technique,following the approach of (Dupacova et al., 2003), is applied. The resulting load andwind speed forecast error scenarios are combined with scaled-up load and wind speedtime series to produce load and wind speed forecast scenarios. Finally, the wind speedforecast scenarios are transformed to wind power forecast scenarios using an aggregatedwind power curve following the approach of (Norgaard and Holttinen, 2004). For eachscenario the demand for replacement reserve (activation time >5 minutes) is calculatedbased on a comparison of the hourly power balance considering perfect forecasts and noforced outages with the power balance considering scenarios of wind and load forecasterrors as well as forced outages. A percentile of the deviation between the comparedpower balances must be covered by replacement reserves; in this case the 90 th percentileis chosen based on current practice (Meibom et al., 2011). A forced outage time seriesfor each unit is also generated by the STT using a semi-Markov process based onhistorical plant data of forced outage rates, mean time to repair and scheduled outages.3.2.2 The Scheduling ModelThe Scheduling Model minimizes the expected costs for all scenarios, subject to systemconstraints for reserve and minimum number of units online (in this case 6 units mustbe online in all time periods in the Republic of Ireland and 2 units in Northern Ireland).A minimum number of online units are maintained to ensure a sufficient level of systeminertia. These costs include fuel, carbon and start-up fuel costs (always assumed tobe hot starts). In addition to replacement reserve, one category of spinning reserve,namely tertiary operating reserve (TR1), is modelled, which has a response time of 90seconds to 5 minutes and can only be supplied by online units. Wind generators, whencurtailed, are assumed to be capable of contributing to spinning reserve requirements.Sufficient spinning reserve must be available to cover an outage of the largest onlineunit occurring concurrently with a fast decrease in wind power production over the

Chapter 3. Unit Commitment with High Wind Penetration 25TR1 time frame, as described in (Doherty and O’Malley, 2005). Generator constraintssuch as minimum down times (the minimum time a unit must remain offline followingshut-down), synchronization times (time taken to come online), minimum operatingtimes (minimum time a unit must spend online once synchronized) and ramp ratesmust also be obeyed.Rolling planning is employed to re-optimize the system as new wind generation andload information become available. Starting at noon each day, the system is scheduledover 36 hours until the end of the next day. The model steps forward with a three hourtime step and in each planning period a three-stage, stochastic optimisation problemis solved. This involves a deterministic first-stage covering three hours, a stochasticsecond stage with three scenarios covering three hours and a stochastic third stage withsix scenarios covering a variable number of hours, depending on the planning periodin question, as seen in Figure 3.1 (AIGS, 2008). The structure of the scenario treeassumes perfect knowledge of load and wind power output in the first three hours anduncertainty in subsequent hours, and the opportunity to revise the planned commitmentevery three hours based on information from new forecasts. The model produces a yearlongdispatch at an hourly time resolution for each individual generating unit.The model can also be run in deterministic and perfect foresight modes wherebyonly one wind generation and load scenario are planned for. In deterministic mode, thisscenario is the expected value of wind and load. The expected value of wind is found bysumming, for all (post-reduction) scenarios, the product of the wind power forecasts andtheir probability of occurrence. The expected value of load and replacement reserve isfound similarly (Tuohy et al., 2009). Consequently, the scenario planned for will differfrom the realized scenario. This mode is typical of the scheduling process currentlypracticed by most system operators, i.e. only one scenario is planned for and it willcontain some level of forecast error. Perfect foresight mode contains no forecast errorfor wind generation or load but forced outages still occur, as with all other modes.Further detail on the model and formulation of the unit commitment problem canbe found in (Meibom et al., 2011). The Generic Algebraic Modeling System (GAMS)

Chapter 3. Unit Commitment with High Wind Penetration 26Figure 3.1: Illustration of rolling planning and decision structure in Wilmaris used to solve the unit commitment problem using the mixed integer feature of theCPLEX solver (version 12). For all simulations in this study the model was run with aduality gap of 0.5%. A year-long simulation takes > 3 hours when run in deterministicmode or > 24 hours in stochastic mode, on an Intel core quad 3 GHz processor with 4GB of RAM.3.2.2.1 Modelling DSMLater versions of the Wilmar planning tool included add-ons to model demand sidemanagement (DSM), which is utilised in Chapter 6.

Chapter 3. Unit Commitment with High Wind Penetration 27DSM units can be either peak clipping or peak shifting units. Peak clipping unitsallow demand to be reduced at a cost to the system without increasing demand atanother time. They are modelled as flexible gas turbines with a variable operatingcost and no fuel or start-up costs. Peak shifting units allow demand to be reducedand reallocated in time at a cost to the system, without reducing the overall energydemand. They are modelled as storage units with 100% efficiency. The constraintsimplemented in Wilmar to model DSM ensure that (i) the DSM units are scheduledday-ahead and their dispatch cannot be revised intra-day, (ii) the DSM units cannotprovide non-spinning reserve, (iii) all demand shifted over a day must must be addedto demand at another point in that day, and (iv) the amount of demand clipped by apeak clipping unit cannot exceed a defined energy limit.3.2.2.2 Improved Modelling of Plant Start-upsMore detailed modelling of plant start-ups was implemented to improve the validityof results. In the original version of the Wilmar Planning Tool, units remained atzero production over the course of their start-up period. Here, units are block loadedfrom zero to minimum output over the course of the start-up process, following theformulation given in (Arroyo and Conejo, 2004).The start-up and shut-down binary variables are set appropriately by Equation 3.1.Power output levels, P U (i), are defined for each interval, i, of the units’ start-up process.Equation 3.2 sets the minimum allowable power output for a unit equal to P U (g,i)when the unit is in the ith interval of the start-up process, or equal to its minimumstable operating level when the unit is online and not in its start-up process. Likewise,Equation 3.3 sets the maximum allowable power output for a unit equal to P U (g,i)when the unit is in the ith interval of the start-up process, or equal to its maximumoperating level when the unit is online and not in its start-up process. Equation 3.4 isneeded for the commitment logic (Arroyo and Conejo, 2004).V Starts,t,g− V Shuts,t,g= V Onlines,t,g− V Onlines,t−1,g (3.1)

Chapter 3. Unit Commitment with High Wind Penetration 28UD∑ gp(s, t, g) ≥ Pgmin [Vs,t,gOnline − Vs,t−i+1,g] Start +i=1UD g∑i=1P U (g, i)V Starts,t−i+1,g (3.2)p(s, t, g) ≤UD g∑i=1P U (g, i)V Starts,t−i+1,gUD∑ g+ Pgmax [Vs,t,gOnline − Vs,t−i+1,g] Start (3.3)i=1V Onlines,t,g≥UD g∑i=1V Starts,t−i+1,g (3.4)3.3 Other Unit Commitment ModelsMany approaches are available for solving the unit commitment problem, as discussed in(Padhy, 2004; Salam, 2007; Sen and Kothari, 1998), ranging from heuristic approachessuch as priority list to mathematical programming approaches such as dynamic programming,Lagrangian relaxation or mixed integer programming. Dynamic programmingwas the first optimization based method to be applied to the unit commitmentproblem and is used worldwide (Padhy, 2004), however it suffers from the curse of dimensionalityas it evaluates the complete decision tree, and thus for larger systems thesolution time can become impractical (Sen and Kothari, 1998). Simplifications suchas truncation or fixed priority ordering have been implemented to reduce the searchspace but this can lead to suboptimal schedules (Salam, 2007). Lagrangian relaxation,which involves decomposing the primal problem into sub-problems which are linked byLagrangian multipliers, is one of the most commonly used unit commitment formulationsin electricity markets worldwide. However, it is well understood that given thenature of how it works it will generally produce sub-optimal solutions and not producea global optimal solution (EirGrid and SONI, 2010b).One of the key advantages to using MIP models (such as the Wilmar model), in additionto global optimality is the ease of adding constraints. As noted in Streiffert et al.(2005), MIP models do not require complex algorithmic development to implement sim-

Chapter 3. Unit Commitment with High Wind Penetration 29ple constraints unlike Lagrangian relaxation models for example which would requirethe addition of new Lagrangian multipliers. (Streiffert et al., 2005) also notes thatmore accurate modelling of combined-cycle plant is more challenging for a Lagrangianrelaxation model compared to a MIP model; a topic that is dealt with in this thesis.MIP models have also benefited in recent years from improvements in the solutionmethods. Traditionally MIP models were solved using the branch and bound technique,however, more recently other techniques such as node pre-solve, heuristics andcutting planes have been implemented to improve the solution and the optimizationtime. The commercial solver CPLEX (which was used in this work to solve the Wilmarmodel) employs these techniques to reduce the upper (heuristics and node presolve) andlower (cutting planes and node presolve) bounds of the objective function (Bixby et al.,2000). Implementing a combination of solution techniques has been found to yield adramatic reduction in optimization time. Branch and bound algorithms have an additionaladvantage of being suitable for parallel processing (Streiffert et al., 2005). Manysystems such as CAISO, PJM and the Irish system are now using or testing MIP unitcommitment models (EirGrid and SONI, 2010b).3.4 The Irish 2020 Test SystemThe test system used in the following chapters is the Irish 2020 system, based onportfolio 5 from the All Island Grid Study (AIGS, 2008; CER, 2010). Table 3.1 showsthe number of units, installed capacity and average operating cost (fuel) by generationtype for this test system. The peak demand from AIGS (2008) was 9.6 GW peakand the total demand was 54 TWh. More recent long term forecasts (EirGrid, 2009)however, have indicated a considerably lower peak demand for 2020, resulting fromthe current economic depression. Thus a revised test system has also been studied, inwhich the demand profile is scaled down to a 7.55 GW peak and a total demand of42 TWh. In this revised system four 103.5 MW OCGT units were also removed fromthe original grid study portfolio (which contained 8 OCGT units as seen in Table 3.1),as recent generation adequacy reports would indicate they are unlikely to be built by

Chapter 3. Unit Commitment with High Wind Penetration 302020 (EirGrid, 2009).Table 3.1: Generation Mix of Test SystemGeneration Type Capacity No. Units Avg. Operating(MW)Cost (e/MWh)Wind power 2000/4000/6000 0CCGT 4012 10 39.79Coal 1324 5 18.45OCGT 828 8 61.16Gasoil 383 8 121.26Other renewables 360 10Peat 343 3 36.32Pumped storage 292 4 0Hydro 216 15 0Legacy CCGT 215 2 47.97CHP 166 2 37.94ADGT 111 1 47.85Tidal 72 0For both of the test systems, three different levels of installed wind power were examined:2000, 4000 and 6000 MW, which supply 11%, 23% and 34% or 15%, 29% and 43%of the total energy demand, on the 9.6 GW peak and 7.55 GW peak systems respectively.The wind power data used to generate the scenario trees, used in AIGS (2008)and this thesis, was 2004 data from 11 onshore regions across Ireland and NorthernIreland and 10 offshore regions. The wind power time series collected from each regionwere smoothed to account for wind correlation effects. To simulate wind speed forecasterrors, required for generating the scenario trees, wind speed forecasts for 6 locationswere used. However, forecast results were only available for time horizons greater than5 hours so in order to generate forecast errors for the first 5 hours persistence forecastswere assumed for the 6 locations. Figure 3.2 shows the day-ahead wind power forecasterror probability function (mean absolute error is 9.6%). It is evident that wind poweris more frequently over-forecast on the test system but the largest forecast errors wereunder-forecasts. The additional amount of spinning reserve that must be carried tocover wind power uncertainty was determined in Doherty and O’Malley (2005) for theIrish 2020 test system and is shown in Table 3.2.

Chapter 3. Unit Commitment with High Wind Penetration 31Figure 3.2: Day-ahead wind forecast errorTable 3.2: Additional spinning reserve requirement due to wind generationWind Generation (MW)TR1 (MW)0-1000 51000-2000 182000-3000 373000-4000 634000-5000 945000-6000 131The pumped storage units, with a round-trip efficiency of 75% and a maximumpumping capacity of 70 MW each, are large providers of spinning reserve to the system,however at least 50% of the spinning reserve target has to be provided by conventionalunits (excluding pumped storage and wind generation). The 2 CHP units have ‘mustrun’status as they provide heat for industrial purposes. The outputs for hydro andtidal units are inputted to the scheduling model as a time series and these units arealso not dispatchable. Sewage gas, landfill gas, biogas and biomass generation make upthe ‘other renewables’ category. Fuel prices are as given in Table 3.3. Base-load gasgenerators (i.e. CCGTs and CHP) are assumed to have long-term fuel contracts andtherefore pay a cheaper fuel price compared to mid-merit gas generators (i.e. OCGTs,

Chapter 3. Unit Commitment with High Wind Penetration 32ADGTs and legacy CCGTs). Differences in the fuel price for coal and gasoil in theRepublic of Ireland and Northern Ireland reflect varying delivery costs.FuelTable 3.3: Fuel Prices by Fuel TypeFuel Price (e/GJ)Renewables 0Coal - Republic of Ireland 1.75Coal - Northern Ireland 2.11Peat 3.71Base-load gas 5.91Mid-merit gas 6.12Gasoil - Northern Ireland 8.33Gasoil - Republic of Ireland 9.64The test system assumes that there is 1000 MW of HVDC interconnection in placebetween Ireland and Great Britain and it is scheduled on an intra-day basis, i.e. it canbe rescheduled in every 3 hour rolling planning period. It is assumed that the total1000 MW can be exported from Ireland to Britain, however, when Ireland is importingfrom Britain 100 MW of capacity is maintained to provide spinning reserve. In additionanother 50 MW of spinning reserve is assumed to be available from interruptible load.A simplified model of the British power system is included, with aggregated units, nointeger variables for generators and where wind generation and load are assumed to beperfectly forecast. The total demand in Britain is assumed to be 370 TWh with a peakof 63 GW and the installed wind capacity is assumed to be 14 GW. A carbon price ofe30/ton was assumed.The 2020 Irish system serves as an interesting test system to study issues arisingfrom large-scale wind power. Being a small island system, with limited interconnectionto Great Britain integration issues arise and become more obvious at lower levels ofwind power and can indicate future issues for other power systems pursuing large-scalewind power. The large proportion of base-load units on the Irish system, most of whichare CCGTs, combined with the high wind penetration deem it useful for studying plantcycling and investigating means of limiting the extent of this cycling operation. Thus,the findings in this thesis bear relevance to other gas and wind-dominated systems, forexample the ERCOT system.

CHAPTER 4Cycling of Base-load Plant on the Irish Power System4.1 IntroductionCERTAIN developments in the electricity sector may result in suboptimal operationof base-load generating units in countries worldwide. Despite the fact thatthey were not designed to operate in a flexible manner, increasing penetration of variablepower sources, such as wind generation, coupled with increased competition inthe electricity sector can lead to these base-load units being shut down, ramped oroperated at part-load levels more often. An overcapacity of generation on a system canalso exacerbate plant cycling as less efficient plant may be prematurely forced downthe merit order.Although all conventional units will be impacted to some degree by the integrationof wind generation, it is cycling of base-load units that is particularly concerning forsystem operators and plant owners. As these units are designed for maximum efficiency,33

Chapter 4. Cycling of Base-load Plant 34they typically have limited operational flexibility, and as such cycling these units willresult in accelerated deterioration of plant components through various degenerationmechanisms such as fatigue, erosion, corrosion, etc. This will lead to more frequentforced outages and loss of income, as discussed in Chapter 2. Start/stop operation andvarying load levels result in thermal transients being set up in thick-walled componentsplacing them under stress and causing them to crack. Cycling interrupts plant operationwhich can in turn disrupt the plant chemistry resulting in higher amounts of oxygen andother ionic species being present, and therefore leading to corrosion and fouling issues.Thus, excessive cycling of base-load units can potentially leave these units permanentlyout of operation prior to their expected lifetimes.The severity of plant cycling, will be dependent on the generation mix and thephysical characteristics of the power system. It is widely reported that the availabilityof interconnection and storage can assist the integration of wind on a power system(IEA, 2008; EWEA, 2011a). Interconnection can allow imbalances from predicted windpower output or variations in net load to be compensated via imports/exports, whilstsome form of energy storage can allow excess wind to be more easily absorbed bycharging (and thereby increasing demand) during these periods. This should relievecycling duty on thermal units as the onus on them to balance fluctuations is relieved.This chapter examines the effect that an increasing penetration of wind power willhave on the operation of base-load units. The role that interconnection and storageplay in alleviating or aggravating the cycling of base-load units is also investigatedacross different wind penetration scenarios.4.2 Scenarios ExaminedThe 2020 Irish system, as described in Chapter 3, was chosen as a test case for this studybecause its unique features make it suitable for investigating base-load cycling. It is asmall island system, with limited interconnection to Great Britain, a large portion ofbase-load plant and significant wind penetration. Thus, potential issues with cycling of

Chapter 4. Cycling of Base-load Plant 35base-load units may arise on this system at a lower wind energy penetration, comparedto a larger, more interconnected or more flexible system. Two versions of the 2020Irish system are discussed in Chapter 3, one with a 7.55 GW peak demand and theother with a 9.6 GW peak demand. Both versions are examined in this chapter. Foreach of the demand scenarios, three levels of installed wind generation, namely 2000,4000 and 6000 MW, were examined. As seen in Chapter 3, the remaining generationis primarily thermal generation, with a small portion of inflexible hydro capacity whilethe base-load is composed of coal and combined-cycle gas turbine (CCGT) generation.The characteristics of a typical base-load CCGT and coal unit on the test systems areshown in Table 4.1.Table 4.1: Characteristics of a Typical CCGT and Coal Unit on the Test SystemCharacteristic CCGT CoalMaximum Power (MW) 400 260Minimum Power (MW) 200 105Maximum Efficiency (%) 57.6 36.9Hot Start-up Cost (e) 13,280 5,320Full Load Cost (e/hour) 15,900 4,880Maximum Spinning Reserve Contribution(% of Max Power) 9 13Minimum Down Time (Hour) 1 5Start-up time (Hour) 2 5The Wilmar model was run deterministically (i.e. the expected value of wind andload is planned for), for one year, for each of the three wind cases, and for both levelsof peak demand in order to examine the effect that increasing wind power penetrationwill have on the operation of base-load units. These are the units with the most limitedoperational flexibility, and as such, will suffer the greatest deterioration from increasedcycling. A sensitivity analysis was conducted to investigate the role that storage andinterconnection play in altering the impact of increasing wind penetration on base-loadoperation. This involved running the model deterministically for one year, for each ofthe three wind cases, first, without any pumped storage on the system, and second

Chapter 4. Cycling of Base-load Plant 36without any interconnection on the system. In order to fairly compare systems withoutstorage/interconnection to the systems with storage/interconnection, the systems mustmaintain the same level of reliability. Thus it was necessary to replace the pumpedstorage units and interconnection with conventional plant. The 292 MW of pumpedstorage was replaced with three 97.3 MW open cycle gas turbine (OCGT) units whilethe 1000 MW of interconnection was replaced with nine 100 MW OCGT units (as 100MW is always used as spinning reserve, the maximum import capacity is 900 MW). Thecharacteristics of these substitute units were set such that they could deliver the sameamount of generation over the same time period as the interconnection/storage unitsthat they replaced. The OCGT units which replaced the storage units were capable ofdelivering the same amount of spinning reserve (132 MW in total). The OCGT unitsthat replaced the interconnection did not contribute to spinning reserve but instead100 MW was subtracted from the demand for spinning reserve in each hour. This isthe assumption used when the interconnector is in place.The cost per MWh from the OCGT units is generally greater than the cost ofimports or production from the storage units, thus the production previously providedfrom storage/interconnection is not shifted directly to these OCGT units. This isadvantageous in this type of study, as the operation of other units on the systemwithout storage/interconnection can be observed, whilst the system adequacy is notundermined by the reduced capacity, thus facilitating the sensitivity analysis. Forexample, had CCGT capacity been used to replace the interconnector, it would likelyprovide the energy that had been previously delivered by the interconnector, but thiswould not allow examination of how the existing units on the system are affectedin the absence of interconnection. The results from the systems without storage andinterconnection were compared to the base case (i.e. with storage and interconnection).To examine the results, the base-load units were categorized as coal or CCGT. Theresults for the individual units in each group were normalized by their capacity toobtain the result per MW for each unit. The average result per MW was then obtainedand this was multiplied by the capacity of a typical coal or CCGT unit (chosen to be260 MW and 400 MW respectively) to give the result for a typical coal or CCGT unit

Chapter 4. Cycling of Base-load Plant 37as shown below:∑ ni=1 (x i/c i )∗ T ypical Unit Size (4.1)nwhere x i is the result for the i th unit, c i is the capacity of the i th unit and n is thenumber of units4.3 Results4.3.1 Increasing Wind Penetration and the Operation of Base-LoadUnitsAs the penetration of wind generation on a power system is increased, large fluctuationsin the net load (load minus wind generation) will occur more frequently, as seen in Table4.2 and Table 4.3, which shows the annual number of hourly net load ramps whichexceed 1000 MW on the 7.55 and 9.6 GW peak demand test systems. (The probabilitydistribution for net load ramps can also be found in Appendix A and shows larger netload ramps occur more frequently on the 9.6 GW peak demand system relative to the7.55 GW peak demand system.)Table 4.2: No. hours when net load changes by >1000 MW from previous hour on the7.55 GW peak demand systemWind energy penetration 15% 29% 43%7.55 GW peak system 90 135 211Table 4.3: No. hours when net load changes by >1000 MW from previous hour on the9.6 GW peak demand systemWind energy penetration 11% 23% 34%9.6 GW peak system 277 342 454In addition, as wind generation is modelled as having zero operating costs, produc-

Chapter 4. Cycling of Base-load Plant 38Table 4.4: Number of thermal units online with increasing wind penetration (averagedat each hour shown over the year)Time (Hour) 00 03 06 09 12 15 18 2115% wind energy penetration 12.9 11.1 12.3 16.1 16.5 16.1 17.2 15.529% wind energy penetration 12.3 10.8 11.6 14.6 15.2 14.9 15.7 14.643% wind energy penetration 11.6 10.5 11.1 13.9 14.3 13.8 14.6 13.6tion from thermal units is increasingly displaced, thus the number of units online willdecrease. This is shown for the 7.55 GW peak demand system, in Table 4.4. Therefore,with less units online to manage growing fluctuations in net load, the onus on thermalunits becomes more demanding with increasing wind penetration.Figure 4.1 and Figure 4.2 show the annual number of start-ups and capacity factorfor an average sized CCGT (400 MW) and coal unit (260 MW), as wind penetrationincreases on the 7.55 and 9.6 GW peak demand systems respectively. The capacityfactor is defined as the ratio of actual generation to maximum possible generation ina given time period (in this case over the test year). As the wind energy penetrationgrows and the variability and unpredictability involved in system operation is increased,the operation of a base-load CCGT unit is severely impacted. Moving from 15% to 43%wind energy penetration the annual start-ups for a typical base-load CCGT unit risefrom 67 to 107, an increase of 60%. On the 9.6 GW peak demand system, annual CCGTstart-ups increase from 31 to 86, a 177% increase, as wind energy penetration increasesfrom 11% to 34%. This increase in CCGT start-ups corresponds to a plummetingcapacity factor for the units as seen in Figure 4.1 and Figure 4.2, as increasing levels ofwind power will displace production from CCGT units and force them closer to midmerittype operation. The start-ups are higher and the capacity factor is lower for atypical CCGT unit on the 7.55 GW peak demand system relative to the 9.6 GW peakdemand system, as CCGTs will more frequently be the marginal units on the systemwith less demand. (Not shown in Figures 4.1 and 4.2 are those CCGT units on thesystem, originally built for base-load operation, but having over time been displacedinto mid-merit operation. With increasing penetration of wind power, such units also

Chapter 4. Cycling of Base-load Plant 39Figure 4.1: Annual number of start-ups and capacity factor for an average CCGT andcoal unit with increasing wind penetration on the 7.55 GW peak demand systemtend to have a decreasing capacity factor, however their annual number of start-ups,which are much larger than a typical base-load CCGT shown in Figure 4.1 and 4.2,actually reduce as they are forced from mid-merit into peaking operation.)As wind generation on the system increases, the timing and predictability of whenCCGT units will be started is also impacted, as seen in Table 4.5, which shows thepercentage of total CCGT starts that occur during each two hour interval over theyear, on the 7.55 GW peak demand system. With just 2000 MW installed wind power(15% energy penetration) CCGT start-ups are seen to be concentrated around 6-7am.However moving to 6000 MW installed wind power CCGT start-ups are now morewidely distributed throughout the day. This will have repercussions for plant personnelwho are responsible for plant start-ups and would indicate that stringent start-up procedureswill need to be put in place for these units, to minimize the risk of difficultiesarising during the start-up process. In the UK market, for example, a generating unitmust be synchronized within a +/- five-minute window when delivering power to thegrid. If late, the grid operator may not accept the power at all, regardless of the fueland production costs already incurred (OSIsoft, 2007).Unlike a CCGT unit, the annual number of start-ups for a typical coal unit on the

Chapter 4. Cycling of Base-load Plant 40Figure 4.2: Annual number of start-ups and capacity factor for an average CCGT andcoal unit with increasing wind penetration on the 9.6 GW peak demand systemTable 4.5: Percentage of total start-ups occurring during each two-hour interval overthe yearTime (Hour) 00-01 02-03 04-05 06-07 08-09 10-112000 MW wind power 1.09 0.82 1.63 46.74 22.01 11.686000 MW wind power 1.03 2.07 7.76 28.79 23.45 9.66Time (Hour) 12-13 14-15 16-17 18-19 20-21 22-232000 MW wind power 2.17 2.45 8.15 2.44 0.27 0.546000 MW wind power 3.10 6.38 14.14 1.21 1.72 0.697.55 GW peak demand system decreases somewhat as the wind energy penetrationincreases, as seen in Figure 4.1. On the 9.6 GW peak demand system start-ups for acoal unit increase with wind energy penetration up to 23% (albeit not as drastically asa CCGT unit). However, at wind energy penetrations greater than 23%, this patterndiverges and the start-ups for a coal unit begin to decrease, as seen in Figure 4.2. Aswind energy penetration grows, the demand for spinning reserve will increase. Due tohigh part-load efficiencies, coal units are the main thermal providers of spinning reserveon this system. As CCGT units are taken offline more frequently with increasingwind penetration, the requirement on coal units to provide reserve to the system isdriven even higher. Coal units also have lengthy start-up times; once taken offline it

Chapter 4. Cycling of Base-load Plant 41is a minimum of ten hours (minimum down time plus synchronization time as seenin Table 4.1) before the unit can be online and generating again. Thus on a systemwith a high wind energy penetration, coal units are even less likely to be cycled offlineto avoid shortfalls in spinning reserve. This would indicate that the units with themost limited operational flexibility may actually be rewarded at high levels of wind fortheir inflexibility and suggests that some form of incentive may be needed to secureinvestment in flexible plants (for example OCGTs), which are commonly reported asbeing beneficial to system operation with large amounts of wind (Kirby and Milligan,2008; Strbac et al., 2007). Coal units do, however, have low minimum outputs so attimes of high wind power penetration more coal units can remain online to meet theminimum units online constraint, thus minimizing wind curtailment. CCGT units, onthe other hand, are typically restricted by high minimum outputs because of emissionsrestrictions as opposed to physical limitations. When running base-load, CCGTsachieve high firing temperatures which allows CO (carbon monoxide) to be oxidizedinto CO 2 . However, at part load levels, when the firing temperature is lower, the COto CO 2 oxidation reaction is quenched by cool regions near the walls of the combustionliner resulting in increased levels of CO (Siemens, 2008a).It would appear from examination of capacity factors in Figure 4.1 and Figure 4.2that a crossover point exists when coal units become the most base-loaded plant onthe system. It is clear from Table 4.1, that coal generation is cheaper than generationfrom CCGT units and so these units are in fact the most base-loaded plant at all windenergy penetrations examined, however they are modelled as having more frequentoutages compared to the CCGT plant, thus yielding relatively lower capacity factors(at some wind energy penetrations) compared with the CCGT units.Figures 4.3 and 4.4 show the utilization factor for an average base-load coal andCCGT unit, and the number of hours they perform severe ramping as wind penetrationincreases. The utilization factor is the ratio of actual generation to maximum possiblegeneration during hours of operation in a given period. Severe ramping is defined hereas a change in output greater than half the difference between a unit’s maximum andminimum output over one hour. Periods when the unit was starting up or shutting down

Chapter 4. Cycling of Base-load Plant 42Figure 4.3: Utilization factor and annual number of hours where severe ramping isperformed for an average CCGT and coal unit with increasing wind penetration on the7.55 GW peak demand systemwere not included. Although coal units, as the most base-loaded thermal generation,will avoid heavy start-stop cycling as wind levels grow they do experience increasedpart-load operation. This is indicated by a drop in utilization factor from 0.90 to 0.81,or 0.92 to 0.88, as wind energy penetration increase from 15% to 43%, or 11% to 34%,on the 7.55 and 9.6 GW peak systems respectively, as seen in Figures 4.3 and 4.4. Theutilization factor for a CCGT unit is also seen to decrease with increasing levels of windpower, however, it remains high in comparison with a coal unit, indicating the relativelysmaller contribution to spinning reserve it provides to the system and correspondinglythe infrequent periods of part-load operation. As seen in both Figures 4.3 and 4.4,both types of unit experience a dramatic increase in periods where severe rampingis required as wind energy penetration increases. For both test systems CCGT unitsexperience more ramping as they are more frequently the marginal units on the system,however the rate of increase in ramping is higher for the coal units as the number ofoperating hours exceeds that for a CCGT as the wind energy penetration increases.Such increases in part-load operation and ramping can lead to cycling damage suchas fatigue damage, boiler corrosion or cracking of headers, as discussed in Chapter 2.A recent study of the impacts of ramping on three of Xcel Energy’s coal plants, for

Chapter 4. Cycling of Base-load Plant 43Figure 4.4: Utilization factor and annual number of hours where severe ramping isperformed for an average CCGT and coal unit with increasing wind penetration on the9.6 GW peak demand systemexample, predicted a 200%-500% increase in variable O&M costs and capital expenses(Danneman and Beuning, 2011).The results reported here are for “average sized” CCGT and coal units. In order toshow how these results correspond to the actual results for the real units modelled, themaximum, minimum, average and standard deviation of the number of start-ups andcapacity factor for the modelled CCGT and coal units are given in Appendix B.4.3.2 Sensitivity AnalysisThe previous section showed the serious impact increasing levels of wind power willhave on the operation of base-load units. The extent of this impact will be determinedby the generation portfolio and the characteristics of the system. This section providesa sensitivity analysis of the effect of the portfolio on the results, by examining theoperation of the base-load units with increasing levels of wind power when storage andinterconnection are removed from the system.

Chapter 4. Cycling of Base-load Plant 444.3.2.1 No Storage CaseFigure 4.5 shows the number of hours online for an average CCGT and coal unit,for increasing wind energy penetrations on the 7.55 GW peak demand system, withand without pumped storage. Although storage will typically charge overnight whenprices are low, thus raising the base-load, Figure 4.5 reveals that base-load units infact spend more hours online on the system without pumped storage, compared to thesystem with storage. Pumped storage units can provide spinning reserve to the systemwhen pumping and when generating, and as such they are large providers of spinningreserve to the system. Their typical mode of operation is to charge at night, althoughthis is typically seen to be concentrated over a small number of hours, and generateat minimum load, providing the maximum amount of spinning reserve possible to thesystem (a maximum of 50% of the total spinning reserve demand can come from storageunits), throughout the day. On the system without pumped storage, this spinningreserve must now be provided by conventional plant. The increased requirement onbase-load units to provide spinning reserve in the absence of storage is evident in Table4.6, which shows the total amount of spinning reserve provided by a CCGT or coalunit on the 7.55 GW peak demand system, with and without pumped storage. Thuson occasions when CCGT or coal units may have been cycled offline on the system withpumped storage, they will now be more likely to be kept online on the system withoutpumped storage.Table 4.6: Total contribution to spinning reserve (MWh) from typical CCGT and coalunit on the 7.55 GW peak demand systemInstalled wind capacity (MW) 2000 4000 6000With storageCCGT 150,001 156,557 151,965coal 166,069 167,295 173,303Without storageCCGT 238,609 238,473 217,015coal 223,884 228,462 225,772As such, Figure 4.6, which shows the number of start-ups for a typical base-loadCCGT and coal unit on a system with and without pumped storage as wind penetration

Chapter 4. Cycling of Base-load Plant 45Figure 4.5: Number of hours online for an average CCGT and coal unit with/withoutstorage and an increasing wind penetration on the 7.55 GW peak demand systemincreases on the 7.55 GW peak demand test system reveals that without storage on thesystem, both CCGT and coal units have reduced start-ups (although the difference issmall). However, although base-load units may benefit from less start-ups and morehours online on the system without storage, the increase in reserve provision from theseunits implies increased part-load operation, which has been shown to cause componentdegradation in base-load plant. The HRSGs in CCGTs in particular can be affected byflow instability, which is associated with part load operation (Wambeke, 2006). (Similarresults were obtained for the 9.6 GW test system and have been included in AppendixC.)4.3.2.2 No Interconnection CaseFigure 4.7 compares the number of hours spent online by a typical CCGT and coalunit on the 7.55 GW peak demand system with and without interconnection, as windenergy penetration is increased. The base-load units are seen to spend significantlymore hours online on the system without interconnection compared to the system withinterconnection. (A similar result was found for the 9.6 GW test system and this hasbeen included in Appendix C.) Due to a large portion of base-load nuclear plant and

Chapter 4. Cycling of Base-load Plant 46Figure 4.6: Number of start-ups for an average CCGT and coal unit with/withoutstorage and an increasing wind penetration on the 7.55 GW peak demand systemcheaper gas prices compared with Ireland, the market price for electricity tends to becheaper in Great Britain. As a consequence Ireland tends to be a net importer ofelectricity from Great Britain and as such will often favour importing electricity beforeturning on domestic units. Thus interconnection to Great Britain displaces conventionalgeneration on the Irish system, forcing units down the merit order and exacerbatingplant cycling. Without the option to import electricity, as shown in Figure 4.7, alldemand must be met by domestic units, requiring more units to be online generatingmore often. Thus, a typical CCGT and coal unit are seen in Figure 4.7 to spend moretime online without interconnection, particularly the CCGT unit which is closer tobeing the marginal unit and therefore its production is displaced ahead of productionfrom a coal unit. Likewise interconnection is seen to displace more base-load productionon the 7.55 GW peak demand system compared to the 9.6 GW peak demand system,and at higher wind energy penetrations compared to lower wind energy penetrations,as with less demand less conventional generation will be online and therefore base-loadunits are closer to being the marginal units on the system.On the 9.6 GW peak demand system, removing interconnection is seen in Figure 4.9to also reduce plant cycling as domestic units were required to stay online. However,

Chapter 4. Cycling of Base-load Plant 47Figure 4.7: Number of hours online for an average CCGT and coal unit with/withoutinterconnection and an increasing wind penetration on the 7.55 GW peak demandsystemas the wind energy penetration is increased, the electricity price in Ireland undercutsBritish prices more often making exports economically viable and eventually a crossoverpoint is reached when the system with interconnection can deal with large fluctuationsin the wind power output via imports/exports more favourably and avoid plant shutdownsrelative to the system without interconnection. Coal units, being the mostbase-loaded units on the system, are first to benefit from increased exports on thesystem and thus a crossover point can be observed for the coal units in Figure 4.9 at34% wind energy penetration.Likewise, for the 7.55 GW peak demand test system electricity prices in Irelandtend to be lower than British prices so exports to Britain are up to four times greaterthan on the 9.6 GW peak demand system. Thus coal units are seen in Figure 4.8 tobenefit from reduced start-ups relative to the case without interconnection. However,similar to the 9.6 GW peak demand system, at lower wind energy penetrations theCCGT units experience less cycling on the system without interconnection, until againa crossover point occurs, this time at 35% wind energy penetration, beyond which thesystem with interconnection benefits from reduced CCGT cycling.

Chapter 4. Cycling of Base-load Plant 48Figure 4.8: Number of start-ups for an average CCGT and coal unit with/withoutinterconnection and an increasing wind penetration on the 7.55 GW peak demandsystemA further sensitivity was conducted to examine base-load cycling when CCGT generationwere the most base-loaded plant on the system. This was conducted for the7.55 GW peak demand system with 6000 MW installed wind capacity and with no interconnection,so that changes to plant operation could be directly attributable to thechange in the merit order rather than a change in the operation of the interconnector.In this sensitivity analysis the cost of coal generation was increased such that it wasmore expensive than generation from base-load CCGT units (but still less expensivethan mid-merit CCGTs). Such a scenario is not unrealistic given the current trend forlow gas prices and the rising cost of coal generation due to environmental restrictions(Carrino and Jones, 2011). As seen in Table 4.7, the results showed drastic increases incycling, not only for coal plant, but CCGT plant also. During periods of very low netload (i.e. high wind generation) it becomes difficult to meet the minimum number ofunits online constraint while also avoiding curtailment of wind generation. As CCGTunits have high minimum loads relative to coal units, they cannot reduce their outputsufficiently to accommodate the wind power, forcing them to be cycled off-line andrequiring other units, in this case the coal units, to be started. The result is greatlyincreased cycling for both types of units.

Chapter 4. Cycling of Base-load Plant 49Figure 4.9: Number of start-ups for an average CCGT and coal unit with/withoutinterconnection and an increasing wind penetration on the 9.6 GW peak demand systemTable 4.7: Annual start-ups for a typical CCGT and coal unit on 7.55 GW peak demandsystem with 6000 MW installed wind power and no interconnectionCoal mostbase-load generationCCGT mostbase-load generationCCGT starts 130 204Coal starts 47 214Total 117 4184.3.3 Effect of Modelling AssumptionsThe results in this chapter were produced by running the Wilmar model in deterministicmode, i.e. the model planned for the expected values of load, wind generation anddemand for replacement reserve. It is considered that this mode is most representativeof current practice. However, in the future as higher wind energy penetrations arereached, stochastic scheduling is likely to be implemented to provide schedules that aremore robust, thus maintaining reliable system operation when large forecast errors mayoccur. To determine the impact that stochastic scheduling will have on the operationof base-load plant further simulations were run using the Wilmar model in stochasticmode, for the 7.55 GW and 9.6 GW peak demand systems, each with 2000, 4000 and

Chapter 4. Cycling of Base-load Plant 506000 MW installed wind power. Figure 4.10 shows the difference in annual start-upsthat was found for a typical CCGT and coal unit, at each of the wind penetrations,when optimized deterministically and stochastically on the 7.55 GW peak demandsystem.As can be seen there is relatively little difference between the two optimizationmethods at the lower wind energy penetrations, but at 43% wind energy penetration(6000 MW installed) the system optimized stochastically has slightly increased starts(+11 for CCGT, +2 for coal). When optimized stochastically, the model must find asolution that satisfies multiple scenarios, covering high and low net loads. Units withlong start-up times, i.e. base-load units, are therefore more often committed when thesystem is optimized stochastically, because if they are required for any of the scenariosthey will have to be committed in advance. No decision has to be made in advance forfast start units, on the other hand, as these can be started in a given hour, when theload and wind generation are known. At lower wind penetrations, as compared to highwind penetrations, it is more likely that the committed base-load generation will berequired as the net demand will simply be higher. However, with a high wind energypenetration, base-load generation that has been committed to meet a high net demandscenario, may in fact not be needed, if the net demand that is realised is low. This maylead to these units being shut-down more frequently, as seen in Figure 4.10, althoughthe difference is relatively small.4.4 SummaryIncreasing penetration of wind generation on a power system will lead to changes in theoperation of the thermal units on that system, but most worryingly to the base-loadunits. The base-load units are impacted differently by increasing levels of wind, dependingon their characteristics. CCGT units see significant increases in start-stop cycling,plummeting capacity factors and are essentially displaced into mid-merit operation. Onthe test systems examined coal units are the main thermal providers of spinning reserveto the system and also are highly inflexible and as a result avoid start-stop cycling but

Chapter 4. Cycling of Base-load Plant 51Figure 4.10: Annual start-ups on the 7.55 GW peak demand system, optimized stochasticallyand deterministicallysee increased part-load operation and ramping. This increase in cycling operation canover time lead to increased forced outages and plant depreciation.Certain power system assets are widely reported to assist the integration of windpower. This chapter examined if pumped storage and interconnection reduced cyclingof base-load units by comparing a system with storage and interconnection to a systemwithout either, across a range of wind penetrations. It was found that in the absenceof storage there was a greater requirement to keep base-load units online to meet thesystem’s spinning reserve requirement. Thus, base-load units were seen to be cycledless on a system without pumped storage, compared to a system with pumped storage.For a system with a high electricity price relative to its neighbours, interconnection wasfound to displace generation from domestic units. As such, base-load units were alsoseen to be cycled less on a system without interconnection compared to a system withinterconnection.In the long term if power systems are to include large portions of variable windpower, a flexible plant portfolio will be needed. As shown in this chapter, a unitthat is highly inflexible but provides a large portion of spinning reserve to the systemwill benefit from its inflexibility by being kept online more. It is also possible thatgenerators that are repeatedly cycled would alter the technical characteristics of theplant which are bid into the market, such as minimum down time or ramp rates,

Chapter 4. Cycling of Base-load Plant 52in an attempt to avoid or minimise cycling. This would indicate that in order toincentivise new plant to be flexible the revenue streams available to a unit may needto be adjusted to reflect the value of flexibility. Some markets include a capacitypayment in order to incentivise generators to be available as much as possible. Aspower systems evolve to include greater penetrations of wind, these payments could berestructured in order to incentivise generator performance, such that new plant is moreadequately designed to deal with cycling. New ancillary services could also be definedand increasing the ancillary services fund could also incentivise operational flexibility.For example, ramping payments to generators for providing ramping service has beenproposed for the Ontario power system (APPrO, 2006).

CHAPTER 5Multi-mode Operation of Combined-Cycle Gas Turbines5.1 IntroductionCOMBINED -cycle gas turbines (CCGTs) are a type of power generating unitthat achieve high efficiencies (up to 61%) by capturing the waste heat froma gas turbine in a heat recovery steam generator (HRSG) and using it to producesuperheated steam to drive a steam turbine (Kehlhofer et al., 2009).The high efficienciesachieved, combined with their ease of installation, short-build times and relativelylow gas prices have made the CCGT a popular technology choice (Watson, 1996;Colpier and Cornland, 2002). In the Republic of Ireland, for example, 43% of the installedthermal capacity is CCGT technology, whilst in the markets of Texas (ERCOT)and New England (NEPOOL) CCGTs represent 37% of the total installed capacity.The operational flexibility of a CCGT unit is limited by the steam cycle, whichcontains many thick-walled components, necessary to withstand extreme temperatures53

Chapter 5. Multi-mode Operation of Combined-Cycle Gas Turbines 54Figure 5.1: Schematic of CCGT in open- and combined-cycle mode (Eskom, 2007)and pressures (Shibli and Starr, 2007; Starr, 2003). To avoid differential thermal expansionacross these components and the subsequent risk of cracking, these componentsmust be brought up to temperature slowly, resulting in slower start-up times and ramprates for the unit overall (Anderson and van Ballegooyen, 2003). Although, as CCGTunits were traditionally base-loaded, this was not a major concern for plant operators.However, by incorporating a bypass stack upstream of the HRSG at the design stage,as shown in Figure 5.1, a CCGT unit has the option to bypass the steam cycle andrun in open-cycle mode, whereby exhaust heat from the gas turbine is ejected directlyinto the atmosphere via the bypass stack (Anderson and van Ballegooyen, 2003). Thisreduces the power output and efficiency of the plant but offers greater operational flexibility.Running in open-cycle mode, the gas turbine has a short start-up time of 15to 30 minutes and is capable of changing load quickly. However, bypass stacks are notalways incorporated because they can potentially lead to leakage losses, thus reducingplant efficiency, while also introducing additional capital costs (Kehlhofer et al., 2009).As discussed in Chapter 1, international energy policy is driving ever greater penetrationsof renewable energy and thus wind power is set to represent a larger portionof the future generation mix (Bird et al., 2005). This is driving a greater demand for

Chapter 5. Multi-mode Operation of Combined-Cycle Gas Turbines 55flexibility within power systems in order to deal with high penetrations of variable anddifficult to predict energy sources (IEA, 2008; Van Hulle and Gardner, 2008). Storage,interconnection and responsive demand are commonly cited as flexible options for dealingwith variability issues (Brown et al., 2008; Göransson, 2008; Hamidi and Robinson,2008) however these options have considerable costs associated with them. Facilitatingopen-cycle operation of CCGT units that have the technical capability to run in opencyclemode (i.e. those with a bypass stack) can also deliver much needed flexibility to asystem with a high wind penetration. This resource is often technically available, butinaccessible due to market arrangements.For example, in SEM (Single Electricity Market), the electricity market of NorthernIreland and the Republic of Ireland, generators submit technical (operating characteristics)and commercial (cost characteristics) data day-ahead and the cheapest generatorsare dispatched on the trading day until the demand is met (EirGrid and SONI, 2010b).The current market rules do not facilitate multiple bids from CCGT units which arecapable of open-cycle operation. Instead these units can bid into the market day-aheadeither their combined-cycle or open-cycle characteristics, but not both at the sametime.In order to derive the greatest benefits from a CCGT unit that can run in open-cyclemode, it is necessary for the scheduling algorithm to explicitly consider both modesof operation for the unit, i.e. open-cycle and combined-cycle (Lu and Shahidehpour,2004). These will have greatly different technical and cost characteristics and so needto be declared individually. Currently most markets do not facilitate CCGT units tosubmit multiple bids representing different modes of operation, thus presently opencycleoperation of a CCGT unit is typically limited to periods when the steam sectionis undergoing maintenance. However, some US systems have begun addressing thisissue to varying degrees, with ERCOT and CAISO seeking to implement configurationbased modelling of CCGTs (Blevins, 2007; CAISO, 2010b).The option to run in open-cycle mode could also provide benefits for the generators.Renewable integration studies have shown that CCGT units will experience signifi-

Chapter 5. Multi-mode Operation of Combined-Cycle Gas Turbines 56cant decreases in running hours and thus will receive less revenue from the marketas they are displaced by greater levels of wind generation which has an almost zeromarginal cost (CAISO, 2010a; Göransson and Johnsson, 2009; NREL, 2010; NYISO,2010; Troy et al., 2010). Due to their high minimum loads CCGTs are shut downfrequently with high wind penetrations as they cannot reduce output sufficiently toaccommodate the wind power output (Troy et al., 2010). By facilitating CCGT unitsto operate in open-cycle mode, these units may have a new opportunity to capturerevenue from increased operation during periods when they might otherwise be offline.For example, if a CCGT unit has been forced offline by high wind generation on thesystem, it may have the opportunity to run as a peaking unit.Multi-mode operation may also lead to a reduction in plant cycling. Online CCGTunits which have bypass stacks can instantaneously switch to open-cycle operation,while remaining online, by opening the bypass damper to release exhaust gases throughthe bypass stack. This could allow the gas turbine to remain online during periods whenthe CCGT would otherwise be shut-down for minimum load reasons, thereby reducingstart-ups for the gas turbine. Likewise, offline CCGT units with bypass stacks canstart-up in open-cycle mode and the steam unit can be warmed slowly to be broughtinto operation at a later point.This chapter examines if a power system with a high wind penetration can benefitfrom the additional flexibility introduced, or if the CCGT units themselves benefit,when they are facilitated to operate in open-cycle mode when technically feasible andeconomically suitable. As discussed in Chapter 3, the all-island Irish 2020 system(AIGS, 2008) is expected to contain both a large share of wind power and CCGT units(50% of which include a bypass stack) and thus provides an appropriate test system.5.2 MethodologyIn order to examine the potential for multi-mode operation of CCGT units some changeswere made to the Wilmar model. A set, ‘ccgt’, of all CCGT units capable of prolonged

Chapter 5. Multi-mode Operation of Combined-Cycle Gas Turbines 57open-cycle operation, i.e. those with bypass stacks, was defined. The set ‘ccgt opena ’ correspondsto these CCGT units when run in open-cycle mode. CCGT units comprisedof two or more gas turbines will have multiple ‘ccgt opena ’ units, as indicated by index‘a’. The relation ‘multi-mode’ is defined to pair each member of ‘ccgt’ with the correspondingmember(s) of ‘ccgt opena ’. To ensure the mutually exclusive operation of these‘ccgt’ units and the corresponding ‘ccgt opena ’ units, the constraint shown in (5.1) wasadded to the model, where V Online is the state binary variable which describes the onlinestatus of the unit. This allows the model to dispatch, when economically optimal,either the ‘ccgt’ (combined-cycle mode) or any/all of the corresponding ‘ccgt opena ’ units(open-cycle mode), for all scenarios ‘s’ and time steps ‘t’, but not both simultaneouslyas they are in reality the same unit.V Onlines,t,ccgt+ V Onlines,t,ccgt opena≤ 1,∀ s, t, multi − mode(ccgt, ccgt opena )(5.1)Equation (5.2), taken from (Arroyo and Conejo, 2004), sets the state binary variablesV Starts,t,irespectively.or V Shuts,t,iequal to 1 for all units ‘i’, when a unit is started up or shut downV Starts,t,i− V Shuts,t,i= V Onlines,t,i− V Onlines,t−1,i (5.2)When modelling multi-mode operation of CCGT units two new circumstances arisewhen calculating the start-up fuel consumption, Fuel Starts,t,i , which must be explicitlyrepresented. Firstly, when a ‘ccgt’ unit transitions from conventional combined-cycleoperation into open-cycle operation no start-up fuel is consumed by the ‘ccgt open ’ unit asrepresented by inequality (5.3), where Startfuel i is the start-up energy used by each unit(measured in MWh). When the ‘ccgt open ’ unit starts from zero production (V Starts,t,ccgt opena= 1 and V Shuts,t,ccgt = 0), the first term on the right hand side of inequality (5.3) determinesthe fuel used by the unit whilst the second term equals zero. Alternatively, when the

Chapter 5. Multi-mode Operation of Combined-Cycle Gas Turbines 58unit switches from combined-cycle to open-cycle operation (V Starts,t,ccgt open = 1 and V Shutas,t,ccgt= 1) the second term causes the right hand side of (5.3) to equal zero. Setting Fuel Starts,t,ias a positive variable and using an inequality condition ensures that when a ‘ccgt’ unitis shutting down and the corresponding ‘ccgt open ’ unit is not starting up Fuel Starts,t,ccgt openawill be 0.F uel Starts,t,ccgt opena≥ (Startfuel ccgtopena− (Startfuel ccgtopena∗ V Starts,t,ccgt opena)∗ V Shuts,t,ccgt)(5.3)The second circumstance relates to the unit transitioning from open-cycle to combinedcycleoperation. In this case the start-up fuel consumed is less than the start-up fuelused in bringing the CCGT online from zero production, as some of this start-up fuelhas already been used to bring the unit online in open-cycle mode and the gas sectionof the plant is in a hot state. As an approximation, the start-up fuel used to bring theunit into combined-cycle operation from open-cycle operation is the difference betweenthe start-up fuel for the ‘ccgt’ and a fraction, α, of the start-up fuel for the ‘ccgt open ’,as seen in (5.4).Based on the operating experience of generators, α was chosen tobe 0.5 here. When the ‘ccgt’ unit is started from zero production (V Starts,t,ccgt aV Shuts,t,ccgt opena= 1 and= 0), the first term on the right hand side of (5.4) provides the start-upfuel consumed whilst the second term equals zero. When the unit switches from opencycleto combined-cycle operation the second term is included, thus approximating thestart-up fuel consumed in this situation.F uel Starts,t,ccgt ≥ (Startfuel ccgt ∗ V Starts,t,ccgt)− (Startfuel ccgtopena∗ V Shuts,t,ccgt opena∗ α)(5.4)In the Wilmar model any unit can contribute to the target for replacement (nonspinning)reserve, provided that an offline unit can come online in time to provide

Chapter 5. Multi-mode Operation of Combined-Cycle Gas Turbines 59reserve for the hour in question and the reserve available from an online unit is notneeded to meet spinning reserve targets. In Wilmar, the contribution from online andoffline units to the replacement reserve target, P Offs,t,i(MW), are calculated individually.In this case the ‘ccgt’ units cannot provide offline replacement reserve as they havelong start-up times, but the corresponding ‘ccgt open ’ units can, given their fast start-uptimes.The constraints shown in (5.5) and (5.6), where P maxiis a unit’s maximumcapacity (MW), ensure that if either the ‘ccgt’ unit or the ‘ccgt open ’ unit is online,then the ‘ccgt open ’ unit cannot contribute to the portion of replacement reserve that isprovided from offline units. This is necessary to avoid the situation where a ‘ccgt’ unitis online and the model allows the corresponding ‘ccgt open ’ unit to contribute to offlinereplacement reserve.P Offs,t,ccgt opena≤ P maxccgt opena∗ (1 − V Onlines,t,ccgt a) (5.5)P Offs,t,ccgt opena≤ P maxccgt opena∗ (1 − V Onlines,t,ccgt opena) (5.6)When the bypass stack is utilized to switch from combined-cycle to open-cycle operation,the transition is automatic and occurs without shutting down the gas turbine orreducing its power output. However, the transition from open-cycle to combined-cycleoperation is dependent on the temperature state of the boiler. Therefore, if the CCGTunit has been operating for a period of time in open-cycle mode and is then scheduledto switch to combined-cycle mode, its output must adjust in order to achieve the correctHRSG inlet temperature, as depicted in Figure 5.2. This was implemented by settingthe allowable power output (P U (i) from (Arroyo and Conejo, 2004)) for each intervalof the CCGT’s start-up process, which begins at hour 0 in Figure 5.2, such that theappropriate soak time is achieved.Scheduled outages for each unit, determined from historical experience (AIGS, 2008),

Chapter 5. Multi-mode Operation of Combined-Cycle Gas Turbines 60Figure 5.2: CCGT start-up from open-cycle modeare inputted in time-series format to the Wilmar model. In this case, CCGT units withthe capability to operate in open-cycle mode are considered to be available to run inopen-cycle mode for a portion of their scheduled outage. Given that gas turbine equipmentis more accessible and compact in comparison with the steam turbine equipment,it was assumed that one third of the maintenance period was sufficient for the gasturbine.5.3 Test SystemThe test system used was the 7.55 GW peak demand test system as set out in Chapter3. Five (of the ten) CCGT units on the Irish system include bypass stacks andtherefore can run in open-cycle mode. Each of these units is currently installed andoperational. The characteristics of these units in combined-cycle mode are given inTable 5.1. Limited data was available for these units in open-cycle mode so each wasgiven characteristics similar to a typical open-cycle gas turbine (OCGT) unit, as shownin Table 5.1. As CCGT 2 and CCGT 5 are comprised of two gas turbines connected toone steam turbine (2+1 configuration), these units were modelled as having two iden-

Chapter 5. Multi-mode Operation of Combined-Cycle Gas Turbines 61tical open-cycle units available for dispatch when the CCGT is operated in open-cyclemode. CCGTs 2 and 3, located in Northern Ireland and CCGTs 1, 4 and 5, locatedin the Republic of Ireland contribute to the minimum units online constraint which ismodelled in Wilmar (as discussed in Chapter 3), for their respective regions.Table 5.1: Characteristics of CCGT units (capable of multi-mode operation) incombined- and open-cycle modesCCGT 1 2 3 4 5Configuration 1+1 2+1 1+1 1+1 2+1Characteristics in combined-cycle modeMax output (MW) 445 480 404 343 480Min output (MW) 240 232 260 220 280Max efficiency (%) 57.6 58.9 53.9 52.9 52.3Min up time (Hours) 4 4 6 4 4Min down time (Hours) 1 2 4 4 2Start-up time (Hours) 2 1 1 2 4Hot start-up fuel (GJ) 2600 2000 1080 1732 2000Max spinning reservecontribution (MW) 42 37 40 25 40Efficiency at maxspinning reserve (%) 57.4 58.1 52.8 52.2 51.3Characteristics in open-cycle modeMax output (MW) 280 160 256 265 160Max efficiency (%) 39.5 38 39.3 39.3 38Min up time (Hours) 0 0 0 0 0Min down time (Hours) 0 0 0 0 0Start-up time (Hours) 0 0 0 0 0Hot start-up fuel (GJ) 14 8 13 13 8Max spinning reservecontribution (MW) 20 20 20 20 20Efficiency at maxspinning reserve (%) 39.3 37.5 39.1 39.2 37.55.4 ResultsA number of model runs were conducted to investigate the potential for multi-modeoperation of CCGT units. The Wilmar model was run in deterministic mode as this ismore representative of current scheduling practice. A year long dispatch was produced

Chapter 5. Multi-mode Operation of Combined-Cycle Gas Turbines 62for each of the three wind power penetrations outlined in Section III, when (i) multimodeoperation of CCGT units is not allowed and (ii) when multi-mode operation ofCCGT units is allowed.5.4.1 Utilization of the Multi-mode FunctionThe average number of times a CCGT unit with multi-mode capability was run inopen-cycle mode and the average production from a CCGT in open-cycle mode overthe year, at each of the wind penetrations examined, is shown in Figure 5.3. Despiteincreasing wind penetration being correlated with an increased demand for flexibility,be it fast starting or ramping, Figure 5.3 shows the multi-mode function is used lessfrequently as wind penetration on the system increases.As more wind power, with an almost zero marginal cost, is added to a system,the production from thermal plant is increasingly displaced and as such there is anincreased likelihood of generators operating at part-load. To illustrate this, Table 5.2gives the annual utilization factor (ratio of actual generation to maximum possiblegeneration during hours of operation) averaged for the coal, CCGT and peat units onthe system with 2000, 4000 and 6000 MW wind power. Therefore, as wind penetrationincreases, online part-loaded units are more often available to ramp up their output tomeet unexpected shortfalls in production, avoiding the need to switch on fast-startingunits, such as the CCGTs in open-cycle mode.Table 5.2: Average utilization factors with increasing wind penetrationInstalled Wind 2000 MW 4000 MW 6000 MWCoal 0.90 0.87 0.82CCGT 0.83 0.79 0.80Peat 0.75 0.55 0.51The trend seen in Figure 5.3 is consistent with the production from peaking plantsas wind penetration increases. Table 5.3 shows the drop in production from the mostutilized OCGT unit, with increasing wind penetration when multi-mode operationis, and is not, allowed.Reduced production from peaking plants due to increased

Chapter 5. Multi-mode Operation of Combined-Cycle Gas Turbines 63Figure 5.3: Average production from a CCGT in open-cycle mode (line) and averagenumber of instances generators utilized open-cycle operation (grey column), shown forvarious levels of installed wind capacitywind penetration has also been observed in other wind integration studies, such asNYISO(2010), however, it is also likely that systems with base-load units that haveslower ramp rates than those examined in this study will rely on fast-starting units(such as CCGTs in open-cycle mode) more often as wind penetration increases. (Allunits on the test system are assumed to be capable of ramping from minimum tomaximum output in one hour or less.) The average production from the CCGT unitsin open-cycle mode, as seen in Figure 5.3, is comparable with average production levelsfrom dedicated OCGT peaking plants on the system when multi-mode operation ofCCGTs is not enabled.Table 5.3: OCGT production (GWh) with increasing wind penetrationInstalled Wind 2000 MW 4000 MW 6000 MWMulti-mode not allowed 8.5 3.9 3.4Multi-mode allowed 2 0.2 0.3As wind penetration increases so too will the demand for replacement reserve, due tothe increased forecast error. The replacement reserve target can be met by fast-startingoffline units or from excess spinning reserve, if available. If sufficient excess spinningreserve is not available to meet the replacement reserve target, the model must ensure

Chapter 5. Multi-mode Operation of Combined-Cycle Gas Turbines 64a number of fast-starting units are offline and available for operation to maintain asecure system. Consequently, as a result of maintaining the replacement reserve target,production from fast-start units (such as the multi-mode units in open-cycle mode) isreduced. Additional simulations were conducted for the various wind penetrations withno replacement reserve target, to investigate the extent that maintaining replacementreserve suppressed the multi-mode units from running in open-cycle mode. For manysystems, such as the Irish system, this is more representative of current practice, whereno replacement reserve target formally exists. Table 5.4 shows the difference in theaverage open-cycle production from multi-mode units that results when no replacementreserve targets are enforced.Table 5.4: Difference in Open-cycle production (GWh) from multi-mode units with noreplacement reserve target enforcedInstalled Wind 2000 MW 4000 MW 6000 MW△ Production 16.9% 7.2% -0.5%As seen, in the absence of a target for replacement reserve, open-cycle productionfrom the multi-mode units is utilized substantially more for the 2000 MW and 4000MW wind power scenarios. However, with 6000 MW wind power, due to more frequentpart-loading of units, there is more frequently an excess of spinning reserve on thesystem, as well as off-line fast-starting units (as per Table 5.3) which can contributeto the replacement reserve target. Thus with 6000 MW wind power, the replacementreserve target has little effect on the open-cycle operation of multi-mode units. Table5.5 shows the average surplus spinning reserve available and the average replacementreserve target per hour for each of the wind cases examined.Table 5.5: Average hourly surplus spinning reserve (MW) available and replacementreserve target (MW)Installed Wind 2000 MW 4000 MW 6000 MWSurplus spinning reserve 65 120 240Replacement reserve target 500 580 700Similarly, if additional peaking capacity, lower in merit relative to the CCGT units

Chapter 5. Multi-mode Operation of Combined-Cycle Gas Turbines 65Figure 5.4: Combined-cycle capacity factor (dashed line) and open-cycle production(solid line) for each CCGT with multi-mode capability for the 2000 MW wind powersystemin open-cycle mode, is added to the system, open-cycle operation from the multimodeCCGTs increases as the new peaking plants are now kept offline to meet thereplacement reserve target instead of the CCGTs in open-cycle mode. To demonstratethis, 4 new OCGT units were added to the test system and the model was run forthe 2000 MW installed wind power scenario. The results showed a 32% increase inopen-cycle production from multi-mode CCGTs.Figure 5.4 shows the capacity factor for each CCGT in combined-cycle mode andits production over the year in open-cycle mode for the 2000 MW wind power scenario.An inverse relationship is evident between the open-cycle production from a CCGT andthe capacity factor of the CCGT, which indicates that usage of the multi-mode functionis related to the amount of time the CCGT is offline. The more often a CCGT is not inoperation but available for dispatch, the more opportunities it has to run in open-cyclemode, and this relationship would be expected regardless of the plant portfolio.The percentage change in total production (combined-cycle plus open-cycle) thatresults when multi-mode operation of CCGTs is enabled is shown in Table 5.6, foreach of the wind penetrations examined. Multi-mode operation increased productionfor CCGT5, the lowest merit CCGT, which was seen to utilize the function most

Chapter 5. Multi-mode Operation of Combined-Cycle Gas Turbines 66Table 5.6: Percentage change in total production when multi-mode is enabled, shownfor each wind penetrationInstalled 2000 4000 6000Wind MW MW MWCCGT1 5.5 5.4 -7.5CCGT2 0 0.1 -0.1CCGT3 -3.3 5 -2.5CCGT4 -1.4 -7.3 -37.1CCGT5 13.3 38.5 11.1frequently across all the wind penetrations examined. Total production from CCGT3and CCGT4, which are mid-merit CCGTs, is reduced in all cases but one. There is arisk (particularly for CCGTs that are frequently the marginal unit on the system, suchas CCGT3 and CCGT4) when offering open-cycle operation, of being dispatched fromcombined-cycle to open-cycle operation at times of low net demand (demand minuswind generation) to alleviate minimum load issues and then losing out to anothergenerator that can come online faster/cheaper, when the net demand increases again.However, it is also likely that in a market environment, generators would strategisewhen they would offer this multi-mode capability to avoid losing out on production.CCGT1, the highest merit CCGT, benefits from increased production when multi-modeoperation is enabled on the system with 2000 MW and 4000 MW installed wind power.This is due to increased exports and reduced production from the other CCGTs, asopposed to increased production in open-cycle mode.5.4.2 Benefits Arising from Multi-mode OperationThe efficiencies of the OCGT peaking units on the system are comparable with theCCGT units in open-cycle mode.However, the CCGT units running in open-cycleoperation are assumed to have a lower gas price, to reflect the advantage of long-termcontracts. Their open-cycle capacity (as seen in Table 5.1) is also larger than the capacityof the OCGTs (103.5 MW each) and they benefit from avoided start-up costs whentransitioning from combined-cycle mode. Thus, when multi-mode operation of CCGTswas enabled, production from OCGT peaking plant tended to be substituted by pro-

Chapter 5. Multi-mode Operation of Combined-Cycle Gas Turbines 67Figure 5.5: Average production from OCGT peaking units in each wind power scenario,with multi-mode operation of CCGTs not allowed (light grey) and allowed (dark grey)duction from the CCGTs in open-cycle mode. Figure 5.5, which shows the averageproduction from OCGTs for each wind penetration level when multi-mode operation ofCCGTs is allowed and not allowed, illustrates this point. Assuming open-cycle productionfrom CCGTs is more economic than production from OCGTs, as is the case here, itis possible that by enabling multi-mode operation of CCGTs sufficient flexibility couldbe extracted from a system’s portfolio of plant to avoid building additional peakingunits, or equally that OCGT units would no longer be able to cover their costs andso would be forced to retire from service. Both situations may then lead to increasedproduction from CCGTs in open-cycle mode.Table 5.7 shows the total shortfall in replacement reserve over the year and thenumber of hours in which this occurred, for each of the wind penetrations examined,when multi-mode operation of CCGTs is, and is not, allowed. The additional faststartinggeneration available to the system when multi-mode operation of CCGT unitsis allowed significantly reduces the shortfall in replacement reserve. This contributesto a more secure system by preventing capacity shortfalls when wind forecasts proveto be overly optimistic and also indicates that, depending on the market structure,the generators may benefit from an additional revenue stream, via ancillary servicepayments for the replacement reserve provided.

Chapter 5. Multi-mode Operation of Combined-Cycle Gas Turbines 68Table 5.7: Magnitude and frequency of replacement reserve shortfall, shown for variouslevels of installed windInstalled Multi-mode CCGT Multi-mode CCGTWind not allowed allowedMW MWh No. hours MWh No. hours2000 1688.7 13 861.4 34000 2972.9 17 880.2 56000 609.9 13 7.6 1In addition to enhanced system security, the additional flexibility available to thesystem when multi-mode operation of CCGT units is allowed will also yield operatingcost savings. Table 5.8 shows the total system production cost savings achieved byenabling multi-mode operation of CCGTs. The total system cost is made up of fuel,carbon and start-up costs for the Irish and British system combined, as they are cooptimized.In this case, these savings were achieved at no additional cost as each ofthe CCGTs is currently capable of multi-mode operation.Table 5.8:CCGTsTotal system cost saving (Me) resulting from multi-mode operation ofInstalled Wind 2000 MW 4000 MW 6000 MWReduction in costs 1.55 0.51 2.65The availability of less expensive peaking capacity when multi-mode operation ofCCGTs is enabled will tend to reduce price spikes. In addition, the model includes costpenalties (these are not included in the system production costs) if demand, spinningreserve or replacement reserve targets are not met. There were no hours when demandwas not met. However, the reduction in hours when the replacement reserve target isnot met, achieved by enabling multi-mode operation, as seen in Table 5.7, consequentlyreduces the number of hours when this cost penalty (e10,000 for the given hour) isincurred. This is seen in Table 5.9 which shows the average electricity price (excludinghours with a cost penalty imposed) and the number of hours when the electricity priceexceeded e500/MWh (including hours with a cost penalty imposed) when multi-modeoperation of CCGTs is allowed, and not allowed, for each of the wind penetration levels.

Chapter 5. Multi-mode Operation of Combined-Cycle Gas Turbines 69Table 5.9: Average price and frequency of price spikes (>e500/MWh)Installed Multi-mode CCGT Multi-mode CCGTWind not allowed allowedMW Average No. Price Average No. PricePrice (e/MWh) Spikes Price (e) Spikes2000 49.76 27 49.70 84000 48.10 27 47.96 96000 45.21 21 45.11 8Figure 5.6: Change in exports across the interconnector when multi-mode operation ofCCGTs is enabledA direct consequence of the reduced prices is seen in Figure 5.6 which shows thechange in exports over the interconnector from the Irish to British systems that resultwhen multi-mode operation of CCGTs is allowed, for each of the wind scenarios examined.A substantial increase in exports is seen as a result of enabling multi-modeoperation of CCGTs, as the number of time periods when the electricity price on theIrish system is less than the British system increases. Imports are largely unaffected.The operation of the interconnector in the scenarios when multi-mode operation ofCCGTs was not allowed is shown in Table 5.10.The increase in exports, resulting from multi-mode operation of CCGTs being en-

Chapter 5. Multi-mode Operation of Combined-Cycle Gas Turbines 70Table 5.10: Operation of interconnector when multi-mode is not allowedInstalled Wind 2000 MW 4000 MW 6000 MWImport (MWh) 658,561 1,776,893 3,339,921Export (MWh) 3,859,473 2,418,475 1,598,117abled, supports a reduction in the level of wind curtailment, as more power is exportedto the British system during periods of high wind generation, thus avoiding generatorminimum load issues. The reduction in curtailment was significant, approximately 57%and 82% on the 2000 MW wind system with the interconnector traded intra-day andday-ahead respectively, but the actual percentage of the annual wind energy this representedwas small (

Chapter 5. Multi-mode Operation of Combined-Cycle Gas Turbines 71multi-mode operation is allowed. This implies a reduction in part-load operation, whichis particularly beneficial for CCGT plant, given HRSG components are susceptible todifferential thermal expansion resulting from flow instability, as well as water chemistryissues, when operated at part-load (Wambeke, 2006). As the time spent online increaseswhen multi-mode operation is allowed, the average duration of the offline period willbe reduced. If the duration of time spent offline decreases the plant is more likely to bein a warmer state when it starts up, thus alleviating the level of creep-fatigue damageassociated with start-ups (Lefton et al., 1997).Table 5.11: Impact of Multi-mode on CCGT 3, 4 & 5 with 2000 MW installed windpowerCCGT 3 CCGT 4 CCGT 5Start-upsNo multi-mode 257 157 29multi-mode 247 141 21Utilization FactorNo multi-mode 0.94 0.86 0.67Multi-mode 0.94 0.87 0.72Average Duration of No multi-mode 19 46 284Offline Period (Hours) Multi-mode 20 41 465.4.3 Sensitivity StudiesUsage of the multi-mode function is dependent on many factors, particularly the amountof flexibility already present in the system. A sensitivity study was conducted to examineusage of the multi-mode function when the system was less flexible to meetingdemand. This involved running the model with 2000 MW wind power (as this level ofwind generation led to the greatest usage of CCGTs in open-cycle mode) and powerexchange across the interconnector fixed day-ahead as opposed to intra-day. Examiningthe usage of the multi-mode function when the interconnector is scheduled day-aheadversus intra-day illustrates how a less flexible system will more frequently utilize theflexibility present in multi-mode CCGT operation. Figure 5.7 shows the average productionfrom the multi-mode CCGTs in open-cycle mode and the average number ofinstances the CCGTs utilized open-cycle operation, with the interconnector scheduled

Chapter 5. Multi-mode Operation of Combined-Cycle Gas Turbines 72Figure 5.7: Average production from a CCGT in open-cycle mode (line) and averagenumber of instances generators utilized open-cycle operation (grey column), withinterconnector scheduled day-ahead and intra-day on the 2000 MW wind systemday-ahead and intra-day on the 2000 MW wind power system. The average productionfrom CCGTs in open-cycle mode with day-ahead scheduling of the interconnector isseen to be more than three times greater than the system with intra-day scheduling ofthe interconnector. By fixing the power exchange between the Irish and British systemsday-ahead, when there is greater uncertainty in the expected wind generation and demand,the system is forced to dispatch generators such as the multi-mode CCGT units,as opposed to rescheduling imports/exports, to compensate for wind and load forecasterrors. Likewise, systems with seasonal hydro restrictions may see greater usage ofmulti-mode CCGT operation during those periods when the operating flexibility of thesystem is reduced.In addition, the quality of wind and load forecasts employed by a system will also determinethe usage of the multi-mode function. Additional simulations were completedrunning the model in stochastic and perfect foresight modes. These represent differentmeans of including load and wind forecasts in the scheduling process; whereby stochasticoptimization can be considered to represent a system employing ensemble forecasts,deterministic optimization is representative of a system utilizing a single forecast andthe perfect forecast scenario is a hypothetical case where no forecast error exists. The

Chapter 5. Multi-mode Operation of Combined-Cycle Gas Turbines 73Figure 5.8: Average production from CCGT in open-cycle mode (GWh), shown fordifferent methods of optimization with 2000 MW wind powerrobust solutions obtained by stochastic optimization showed less deployment of themulti-mode function compared with the deterministic results. The stochastic solution,optimized over several wind and load scenarios, typically has more units online to coverall scenarios and therefore is more prepared to deal with unforseen shortfalls in windgeneration or increases in demand, without the need for starting peaking plant. Thecapacity factors of the CCGT units are also higher for the stochastic case comparedto the deterministic case indicating that there was also less opportunity for these unitsto run in open-cycle mode when the system is optimized stochastically. Running theWilmar model with perfect foresight of the system demand and wind profile also revealseven less open-cycle operation from CCGTs, as in this case, with no forecast errors onthe system (except forced outages of generators), fast starting units are in less demandrelative to the deterministically optimized solution. Figure 5.8 compares the averageopen-cycle operation from the multi-mode CCGTs, on the system with 2000 MW windpower, when optimized with perfect foresight, stochastically and deterministically. Theaverage open-cycle production from a CCGT unit is seen to be 11% less on the stochasticallyoptimized system and 35% less on the system with perfect forecast compared tothe deterministic case.A sensitivity analysis was also conducted using a higher level of demand on the

Chapter 5. Multi-mode Operation of Combined-Cycle Gas Turbines 74system. In this case the original demand profile from AIGS (2008) with a 9.6 GWpeak, discussed in Chapter 3, was run for each wind scenario. The average productionfrom the multi-mode CCGTs in open-cycle mode over the year is shown in Figure 5.9to be six to eight times greater on the 9.6 GW peak demand system, where peakingcapacity is in greater demand, compared to the 7.55 GW peak demand system, ateach of the wind power penetrations examined. In addition to the increased demandresulting in increased open-cycle production from the multi-mode CCGTs (as wellas combined-cycle production), the other main difference between the scenarios is thepredominant direction of power transfer on the interconnector. With 2000 MW installedwind capacity the Irish system is a net importer of power from Britain, at both levelsof demand examined. However, as more wind power is installed on the 7.55 GW peakdemand system the marginal electricity price is reduced sufficiently with respect to theBritish system such that Ireland becomes a net exporter of power. Although increasingwind power penetration on the 9.6 GW peak demand system also reduces the marginalprice it is still a net importer with 6000 MW installed wind power. Thus, on occasionswhen forecast wind is overestimated and the system is in need of fast-starting plant,the 7.55 GW peak demand system, being a net exporter, can more frequently chooseto curtail exports or start up a unit to compensate. In contrast, the 9.6 GW peakdemand system, being a net importer, more often only has the option to turn on faststartingplant. Hence, this implies that a system which tends to be a net exporter isinherently more flexible, and has more options for dealing with variable wind powerthan a system that is a net importer of power. In this scenario with higher demand,each of the multi-mode CCGT units experienced increased total production (combinedcycleplus open-cycle) when multi-mode operation was allowed, suggesting that offeringmulti-mode capability may prove more profitable on a system with a smaller capacitymargin.Given the low deployment of the multi-mode functionality on the 7.55 GW peakdemand system and the high capacity factor in combined-cycle mode for CCGT 1 and2, as seen in Figure 5.4, it would appear that there is insufficient incentive for allCCGTs capable of multi-mode operation to offer this flexible capability. Thus, given

Chapter 5. Multi-mode Operation of Combined-Cycle Gas Turbines 75Figure 5.9: Average production from a CCGT in open-cycle mode on the 7.55 GWpeak demand system (light grey) and the 9.6 GW peak demand system (dark grey),shown for various levels of installed wind powerthat CCGTs 3, 4 and 5 have low capacity factors in combined-cycle mode, additionalsimulations were conducted to investigate the resulting benefits if these units alone,and if CCGT 5 alone, offered multi-mode capability. Table 5.12 shows the total systemcost (for Ireland and Britain) and the magnitude of the replacement reserve shortfallover the year for these configurations (in addition to other configurations examined inthe paper). Examining the shortfall in the replacement reserve target for the differentconfigurations reveals that the majority (≈ 80%) of the reduction in replacement reserveshortfall due to multi-mode capability is attributable to CCGT 5, while CCGTs 1 and 2are seen to have no impact on the replacement reserve shortfall. Thus, CCGTs capableof open-cycle operation, which have very low output in combined-cycle mode, havevalue in providing replacement reserve.

Chapter 5. Multi-mode Operation of Combined-Cycle Gas Turbines 76Table 5.12: Total system cost, replacement reserve shortfall and top-up payment, shown for various multi-mode configurationsConfiguration Total System Replacement Avg. Top-upCost Reserve Payment/ Saving Shortfall (no. units)All cases with 2000 MW wind power Me MWh Me7.55 GW Peak, No Multi-mode 13372.03 / - 1688.7 -7.55 GW Peak, 5 Multi-mode CCGTs 13370.48 / 1.55 861.4 1.36 (2)7.55 GW Peak, 3 Multi-mode CCGTs (3, 4 & 5) 13368.99 / 3.04 861.4 0.49 (3)7.55 GW Peak, 1 Multi-mode CCGT (5) 13371.73 / 0.3 1032.4 07.55 GW Peak, No Multi-mode, day-ahead interconnector trading 13384.64 / - 2197.9 -7.55 GW Peak, 5 Multi-mode CCGTs, day-ahead interconnector trading 13382.98 / 1.66 798 1.66 (2)7.55 GW Peak, No Multi-mode, stochastic 13371.23 / - 966.5 -7.55 GW Peak, 5 Multi-mode CCGTs, stochastic 13371.27 /-0.04 394 0.91 (2)7.55 GW Peak, No Multi-mode, perfect foresight 13370.87 / - 0 -7.55 GW Peak, 5 Multi-mode CCGTs, perfect foresight 13369.38 / 1.49 0 0.45 (1)9.6 GW Peak, Multi-mode not allowed 13997.24 / - 68345.9 -9.6 GW Peak, 5 Multi-mode CCGTs 13996.16 / 1.08 63265.3 0

Chapter 5. Multi-mode Operation of Combined-Cycle Gas Turbines 77As seen in Table 5.6, the multi-mode CCGTs may experience a reduction in totalproduction as a result of offering multi-mode capability to the market. This was alsoobserved to be the case for CCGTs 3 and 4, when only three units offered multi-modeoperation. This indicates that a system seeking to increase its flexibility via multimodeoperation of CCGTs, possibly to facilitate integration of variable renewables,may need to reward these units either through ancillary service payments or anothermarket mechanism to restore their revenue to original levels (i.e. when multi-modeoperation was not allowed). The subsidy or “top-up payment” required to restorethe revenue of these units to their original level is estimated here as the loss in totalproduction multiplied by the average electricity price. The average “top-up payment”required is shown in Table 5.12 with the number of units requiring this payment shownin parenthesis. However, it should be noted that this represents the worst-case figuregiven that the multi-mode CCGT unit offered this capability in all time periods, ratherthan when it was profitable for them to do so, as would likely be the case in reality.5.5 SummaryAmending the scheduling model used by TSOs to include the bids and technical characteristicsof a CCGT unit in open-cycle mode, in addition to the conventional CCGTunit, is a simple task. The CCGT unit in open-cycle mode can be defined as a new unit,with a constraint added to ensure that the CCGT unit and the CCGT unit in open-cyclemode cannot be scheduled to run at the same time. This chapter examined if allowingCCGT units to operate in open-cycle mode, when this is technically feasible and costoptimal, could deliver benefits to a system with a high wind penetration or to the generatorsthemselves. It was shown that the additional fast-starting capacity available frommulti-mode operation of CCGTs reduced the replacement reserve shortfall, indicatingan opportunity for increasing system reliability. Low-merit CCGTs were seen to utilizethe multi-mode function more than high-merit CCGTs, as they are frequently offlineand available for dispatch, whilst the increased competition among generators, typicalat higher levels of wind generation, resulted in multi-mode operation of CCGTs being

Chapter 5. Multi-mode Operation of Combined-Cycle Gas Turbines 78utilized less frequently. Peaking production from CCGTs in open-cycle mode displacedpeaking production from OCGTs, potentially reducing the need for such units to bebuilt. Sensitivity studies revealed that usage of the multi-mode function is dependenton the level of flexibility inherent in the system. Optimizing the system stochasticallyor allowing intra-day trading on interconnectors reduced the need for flexibility to beextracted from generators and consequently resulted in less frequent deployment of themulti-mode function.The analysis in this chapter assumed that the CCGT units capable of multi-modeoperation offered this flexibility in all time periods, whereas in reality generators wouldstrategise when to offer open-cycle operation such that plant production levels are notnegatively impacted, as was seen for some units under certain scenarios in this chapter.Nonetheless, it was shown that the payment required to restore generator revenue tolevels when they did not offer multi-mode operation, in those cases where generatorproduction was reduced, was typically less than the system cost saving, indicating a netbenefit to society. A cost saving is also associated with the reduction in replacementreserve shortfall which has not been considered here.

CHAPTER 6Power System Flexibility and the Impact on Plant Cycling6.1 IntroductionPOWER system flexibility is defined in IEA (2008) as the ability to respond rapidlyto large fluctuations in supply or demand. A flexible power system, therefore, isinherently capable of supporting a larger penetration of variable renewables. As windgeneration continues to grow, the operating flexibility of conventional plant may proveinsufficient to meet an increasingly variable net demand. In addition, increased cyclingof these plants can lead to extensive damage to the plant’s components, particularlyfor base-load plant, as described in Chapter 2. Thus, considerable interest surroundsthe idea of incorporating sources of flexibility into power systems to support a higherpenetration of wind power.Energy storage facilities, interconnection to neighbouring power systems and demandside management schemes (DSM) are well cited sources of flexibility within a79

Chapter 6. Options for Increasing Power System Flexibility 80power system (IEA, 2008; Van Hulle and Gardner, 2008). Each of these flexibilityoptions can assist in balancing variations in the net load. The flexibility of interconnectionis present in the ability to import electricity from, or export electricity to, aneighbouring power system, thus reducing the burden of managing net load variabilitydomestically. Energy storage will charge when the electricity price is low and generatewhen prices are high, which will tend to flatten the net load curve (somewhat). Lowprices are associated with high wind penetration, and if storage units charge duringthese periods it will raise the system demand, requiring increased production fromconventional plant and possibly keeping them online when they may otherwise havebeen forced off-line. Demand side management schemes, depending on their nature,can allow demand to be shed completely or shifted in time to better suit the net loadprofile. In the context of a system with a large wind penetration the ability to shedor reschedule demand is useful if forecast wind fails to materialise or wind generationbegins to reduce rapidly and production from conventional plant cannot be rampedup quickly enough to compensate or alternatively when high wind generation coincideswith low demand, potentially forcing generators to be shut-down.It was found in Brown et al. (2008) that pumped storage on isolated systems canallow a greater penetration of renewables and improve the dynamic security of thesystem, however Tuohy and O’Malley (2009) also shows that, although pumped storagecan reduce wind curtailment, the increased use of base-load units can actually lead toincreased emissions. Hamidi and Robinson (2008) found that responsive demand ona system with a high wind penetration makes greater use of the wind resource andreduces emissions, whilst Keane et al. (2011) finds DSM substitutes production frompeaking units and can provide a valuable source of reserve. It was also noted in Malik(2001) that the avoided cycling cost of thermal units is a major benefit of DSM. Thenet benefits of wind generation can be increased significantly by increasing the level ofinterconnection on the power system, as shown in Denny and O’Malley (2007), whilstGöransson (2008) also shows that investment in transmission to a region sufficiently faraway to make wind speeds uncorrelated (supergrid), or to a region with excess flexiblecapacity, can decrease the total system costs of a system with a high wind penetration.

Chapter 6. Options for Increasing Power System Flexibility 81In addition, the next generation of fossil-fired generation is set to be more flexibleas plant manufacturers, in response to the changing needs of power companies, are nowlaunching high efficiency power plants which are suited to cycling operation (GE, 2011;Siemens, 2008a,b). As discussed in Chapter 4, the high minimum loads of CCGT unitsresulting from emissions limitations often lead to them being forced off-line during highpenetrations of wind generation. However, plant manufacturers have now developedsolutions (such as bypassing compressed air around the combustor into the turbine toincrease the fuel-to-air ratio inside the combustor) to achieve higher firing temperaturesat lower loads, thus reducing the part-load emissions. This could facilitate new CCGTunits to remain online during periods of high wind generation (Siemens, 2008a). Thenew Siemens H class CCGT, for example, can operate stably at 100 MW, less than20% of its rated output (Probert, 2011). As shown in Chapter 5, it is also plausiblethat existing CCGT units may in the future be operated in open-cycle mode as wellas combined-cycle mode (assuming simple market changes are made), releasing anadditional source of flexibility to the system.This chapter examines how these various forms of flexibility will alter the operationof base-load plant and investigates which is most beneficial to scheduling a system witha large supply of variable wind power to reduce cycling of these inflexible plants. Theeffect on wind curtailment and CO 2 emissions are also examined. In addition, otherforms of flexibility are discussed, namely battery electric vehicles, maintenance scheduling(with consideration of system flexibility), the ability to control wind generation andfaster markets.6.2 MethodologyThe approach employed here was to incorporate equal capacities of the various sourcesof flexibility in turn into the test system. By comparing each scenario against thebase case, the impacts of each flexibility option on system operation, and in particularthe operation of base-load plant, could be determined. The test system used is theIrish 2020 test system with a 7.55 GW peak and 6000 MW installed wind capacity, as

Chapter 6. Options for Increasing Power System Flexibility 82described in Chapter 3. As per Chapter 4, the results have been normalized to give theresult for a typically sized CCGT or coal unit.Five scenarios were developed altogether, each incorporating 500 MW of a flexibleresource into the base case test system. These scenarios included 500 MW interconnection,pumped storage, DSM, additional turndown capability for CCGTs (i.e. reducedminimum operating levels) and open-cycle capacity from multi-mode operationof CCGTs, as summarised in Table 6.1.Table 6.1: Scenarios ExaminedScenario 1Scenario 2Scenario 3Scenario 4Scenario 5500 MW Interconnection500 MW Pumped Storage500 MW DSM500 MW Turndown500 MW Multi-modeIn scenario 1 which included 500 MW interconnection, the transfer of electricalenergy between the interconnected systems can be rescheduled in every planning period.In scenario 2, the 500 MW pumped storage was split into 4 identical 125 MW storageunits, with characteristics as shown in Table 6.2. The storage units in scenario 2all pumped to, and generated from, the same reservoir. (Thus if one unit has notpumped any water it can still generate, provided the other units have pumped waterinto the reservoir.) Pumping at maximum output required 8.5 hours to fill the reservoir.Running at minimum output, the storage units, as they can run independently, couldgenerate for 408 hours.Table 6.2: Characteristics of new pumped storage unitsMax generation (MW) 125Min generation (MW) 10Max storage content (total) (MWh) 5000Min storage content (total) (MWh) 920Max charging (MW) 120Min charging (MW) 120Max contribution to TR1 (MW) 50Round trip efficiency (%) 78

Chapter 6. Options for Increasing Power System Flexibility 83In scenario 3 which contained 500 MW DSM, the DSM was modelled as two 250MW units, one a peak shifting unit and the other a peak clipping unit (the 50:50ratio between shifting and clipping capacity was also used in KEMA (2005)). Thepeak shifting unit corresponded to load which could be shifted in time during theday without reducing the overall energy demand, for example refrigeration load. Assuch any reduction in demand must be replaced within the day (i.e. the total energyexchange is equal to zero). It was modelled as a storage unit with 100% efficiency, asdescribed in Chapter 3. When the storage unit generates it corresponds to a demandreduction by DSM, and when the storage unit charges it corresponds to the demandbeing replaced. The DSM peak shifting unit could contribute up to 42 MW of spinningreserve when it was actively reducing demand (i.e. when the representative storage unitwas generating) and had a variable operating cost of e40/MWh. The peak clipping unitcorresponded to peak load which could be reduced at times of high electricity pricesand does not increase demand at another time, for example lighting demand. The peakclipping unit had a variable operating cost of e80/MWh and could also deliver up to42 MW of spinning reserve when it was actively reducing demand. The values for thevariable operating costs and spinning reserve capabilities of the peak shifting and peakclipping units were taken from Keane et al. (2011).In scenario 4, the minimum operating level for five CCGT units on the test systemwas reduced by 100 MW each (from an average minimum operating level of 220 MW).For scenario 5, two CCGT units on the system (CCGT 4 & 5 from Chapter 5) wereassumed to be capable of multi-mode operation, thus releasing 500 MW additionalopen-cycle capacity to the system when the units were not running in combined-cyclemode. (The open-cycle capacity of these CCGTs is altered here compared with Chapter5 in order to provide 500 MW flexible capacity in total.)

Chapter 6. Options for Increasing Power System Flexibility 846.3 Results6.3.1 Impact on the Operation of Base-load UnitsThe cycling activity of the CCGT and coal units on the base case test system wasdescribed in Chapter 4. CCGTs were seen to undergo a large number of annual startupsrelative to the coal units. Given their high minimum operating levels they areforced off-line during periods of high wind penetration. The coal units on the otherhand avoid start-stop cycling as they provide the cheapest fossil-fired generation to thesystem and also their low minimum operating levels allow them to stay online duringperiods of high wind generation. However, the high part-load efficiency of the coal unitsmeans they are the main providers of spinning reserve on the system and so operateat part-load levels frequently. Coal units are also subject to severe ramping duringperiods of very high wind generation as they are some of the few thermal units onlineto provide power balancing. Severe ramping is defined here as a change in outputgreater than half the difference between a unit’s maximum and minimum output overone hour (excluding hours when the unit is starting up or shutting down).Figure 6.1: Change in start-ups and production for a typical CCGT unit in each scenariorelative to the base case

Chapter 6. Options for Increasing Power System Flexibility 85Scenario 1 - InterconnectionFigure 6.1 shows the change in the average number of annual startups and productionfor a typical CCGT unit, for each of the scenarios investigated. Of all the flexibilityoptions examined the addition of 500 MW interconnection on the test system resultedin the greatest reduction in start-stop cycling (17 less starts per year) for a typicalbase-load CCGT unit. The reduction in cycling for CCGTs was also seen in Figure6.1 to be correlated with increased production (an additional 80 GWh, an increase ofapproximately 3.4%). With 6000 MW installed wind power capacity on the system,prices on the Irish system frequently undercut those in Britain to the extent that theIrish system is a net exporter of electricity, as discussed in Chapter 4. The increasein exports allows for increased production from base-load plant and also with the opportunityto export during periods of high wind power penetration CCGT units canavoid being shut-down. Although not shown here, production from lower merit CCGTshowever, which are effectively in mid-merit operation, is displaced by the increased interconnectioncapacity, as import levels also tend to increase at times when these unitsare the marginal units on the system.Figure 6.2 shows the change in the average number of annual startups and productionfor a typical coal unit, for each of the scenarios investigated. The coal units inthe base case were at their minimum number of annual start-ups so no reduction incoal plant start-ups was possible. However, the coal units did benefit from a largereduction in ramping operation, as seen in Figure 6.3, which shows the average numberof hours severe ramping was required from CCGT and coal units over the year. Thereduction of 118 hours (38%) of severe coal ramping relative to the base scenario wasthe largest reduction in ramping of all scenarios examined. Likewise the reduction inCCGT ramping of 47 hours (42%) relative to the base case was the largest observedover all the scenarios investigated given that the increased export capacity will allowmore opportunities to balance net load variability through exchanges with the Britishsystem. Thus, the additional flexibility that interconnection provides is particularlybeneficial to a system that is a net exporter, however, as seen in Chapter 4, it can

Chapter 6. Options for Increasing Power System Flexibility 86exacerbate cycling on a system that is a net importer.Figure 6.2: Change in start-ups and production for a typical coal unit in each scenariorelative to the base caseScenario 2 - Pumped StorageThe addition of 500 MW of pumped storage increased the annual start-ups for a typicalCCGT unit by 9 relative to the base case, as seen in Figure 6.1. This increase in cyclingfor a typical CCGT is correlated with slightly increased CCGT production (+1%) andonline hours (not shown), implying that although the CCGT units are being cycled morethey are gaining new opportunities for generation due to the introduction of the newstorage units. The increase in start-ups for a typical CCGT, with the additional pumpedstorage on the system, was seen to arise as the addition of storage led to increased levelsof exports, requiring CCGT units to be started up to meet the additional demand.Table 6.3: Operation of new storage unitsUnit 1 Unit 2 Unit 3 Unit 4Utilization factor for generation (%) 40.9 44.5 43.4 43.1Utilization factor for spinning reserve (%) 38.5 40.4 42.9 45.3However, the production for a typical coal unit on the system, shown in Figure 6.2,was seen to decrease by 3.6%, while annual start-ups were seen to increase (+5), due tothe additional pumped storage capacity. Examining the operation of these new storage

Chapter 6. Options for Increasing Power System Flexibility 87units revealed that they were used as much to provide spinning reserve to the systemas generation to the system. Table 6.3, which provides the utilization factor for each ofthese new storage units on the system (total generation divided by maximum generationpossible during online hours) as well as the spinning reserve utilization factor (definedhere as total spinning reserve provided divided by maximum spinning reserve possibleduring online hours), illustrates this trend. Consequently, the demand for spinningreserve from coal units, which are the main thermal providers of primary reserve on thesystem, is reduced and as such, as was seen in Chapter 4 also, these units can now becycled off-line on occasion as the requirement for them to be online providing spinningreserve is reduced. Therefore, the amount of spinning reserve provided from coal unitsdrops 12% with the introduction of 500 MW pumped storage. Increased instances ofsevere coal ramping were also observed in Figure 6.3.Figure 6.3: Change in the number of hours severe ramping was required by a typicalCCGT or coal unit in each scenario relative to base caseScenario 3 - Demand Side ManagementThe schedule for the DSM peak shifting and peak clipping units are set day-aheadand cannot be revised intra-day. This limits the flexibility they can provide to thesystem and can lead to sub-optimal decisions due to forecast uncertainty. As shown in

Chapter 6. Options for Increasing Power System Flexibility 88Chapter 3, day-ahead wind generation is more often over-forecast than under-forecast,thus reducing the net load predicted for the following day. This will thereby tend toreduce the amount by which DSM will be utilised, particularly the expensive peakclipping DSM unit. As such, the peak clipping unit has a capacity factor of 1.3% overthe year, while the peak shifting unit has a capacity factor of 5.7%. The clipping unitis never dispatched at its maximum output, but provides its maximum contribution tospinning reserve (42 MW) in every hour that it is utilised. Similarly the peak shiftingunit provides its maximum contribution to spinning reserve (42 MW also) in 89% ofthe time that it is online. Thus the main functionality of the DSM units is in providingreserve rather than reducing the demand. This was seen to have a detrimental effecton the cycling of the base-load generation, despite its limited utilization.The addition of 500 MW of DSM increased start-ups for a typical CCGT by 18,relative to the base case, as seen in Figure 6.1, while starts for a typical coal unitincreased by 7, as seen in Figure 6.2. These were the largest increases observed acrossall scenarios. Figure 6.3 also showed a large increase in the instances of severe rampingfor a typical coal unit (+294). As was seen previously with pumped storage, when DSMunits contribute to the spinning reserve target there is less requirement for thermal unitsto be online providing spinning reserve. Thus, the system with DSM will tend to commitless generation day-ahead. When forecast wind generation then fails to materialize thefollowing day, conventional generation needs to be started at short notice, giving riseto the large increase in start-ups. (Production from peaking units also increased byalmost 400%). Ramping is also increased, particularly for coal units as seen in Figure6.3.To examine if DSM could bring about a reduction in base-load cycling, assumingthat it did not contribute to spinning reserve, sensitivities were run in which (i) the peakclipping and peak shifting units did not provide spinning reserve, (ii) the peak clippingand peak shifting units did not provide spinning reserve and their dispatches could berescheduled intra-day, and (iii) same as (ii), but with the variable operating cost forthe clipping unit reduced from e80/MWh to e60/MWh and the variable operatingcost for the shifting unit reduced from e40/MWh to e20/MWh. Table 6.4 shows the

Chapter 6. Options for Increasing Power System Flexibility 89Figure 6.4: Change in start-ups for a typical CCGT and coal unit for each DSM scenariocapacity factor of the DSM peak clipping and shifting units for each of these scenarios.Removing the ability to provide spinning reserve reduced the utilization of these units,while allowing their dispatch to be changed intra-day increased utilization of the peakclipping unit, but reduced utilization of the peak shifting unit. Reducing the variablecost of DSM increased its utilization relative to the original scenario. These relativelysmall reductions in the utilization of the DSM units had large impacts on cycling ofbase-load plant as seen in Figure 6.4 and Figure 6.5 which show the change in startsand production from the base case for each of the DSM scenarios examined.Table 6.4: Capacity factor of DSM units for various scenariosPeak clipping Peak shiftingDSM 1.3% 5.7%DSM, no reserve 0.16% 4.9%DSM, no reserve, with rescheduling 0.35% 4.8%DSM no reserve, with rescheduling, reduced cost 3.68% 7.5%For the new sensitivities the level of plant cycling is comparable with the base case.Only a minor reduction in CCGT start-ups was achieved (-2) and this was seen tocorrespond to reduced production from those units. Given the need for DSM shiftingunits to have zero impact on net energy over the day, its utilization is limited and assuch, as shown in these results, it does not hold benefits for cycling of base-load plant.

Chapter 6. Options for Increasing Power System Flexibility 90Figure 6.5: Change in production for a typical CCGT and coal unit for each DSMscenarioScenario 4 - TurndownIn this scenario the CCGTs whose minimum operating level was reduced were seen toutilise the increased turndown over an average of 330 hours throughout the year. Byreducing the minimum operating level of five CCGTs on the system, those CCGTs wereseen to benefit from reduced annual start-ups (annual start-ups for a typical CCGTwere down by 15) and subsequently increased levels of production (+95 GWh), asseen in Figure 6.1. However, these units did experience increased instances of severeramping, as they are now kept online during periods of high wind generation whenpreviously they were shut-down. As such, Figure 6.3 shows an increase of 103 hourswhen severe ramping was required from a typical CCGT unit. As might be expectedwith CCGTs gaining increased production, production for a coal unit was consequentlyreduced (-30 GWh), as seen in Figure 6.2. Increased production from the five CCGTsalso reduced production from peaking capacity (-27%), thus resulting in reduced CO 2emissions as seen in Table 6.6.Scenario 5 - Multi-mode operation of CCGTsThe total production over the year from the multi-mode CCGTs in open-cycle modewas 27.5 GWh, almost 4 times as much as the highest merit OCGT peaking unit in thebase case. Overall, the impact of including 500 MW of additional open-cycle capacity,

Chapter 6. Options for Increasing Power System Flexibility 91via multi-mode operation of 2 CCGT units, had a small impact on the system dispatch.The multi-mode CCGTs, when dispatched, were typically online around evening peakhours and tended to impact production from low-merit CCGT units and peaking units.As seen in Figure 6.1, the number of annual start-ups for a typical CCGT unit and thenumber of instances that a typical CCGT or coal unit was required to perform severeramping, as seen in Figure 6.3, decreased indicating avoided cycling damage.6.3.2 Impact on Wind Curtailment and CO 2 EmissionsThe available wind power on the test system in the test year was 18.4 TWh. Table 6.5shows the amount of available wind that was curtailed in each of the scenarios. It isclear that pumped energy storage, having the most flexible energy storage potential ofthe options examined, was most effective at minimising wind curtailment events on thesystem.Table 6.5: Curtailment of wind in each scenarioScenario Wind Curtailed % Change from(GWh) Base CaseBase Case 148.4 -Interconnection 61.5 -58.5Pumped storage 56.0 -62.3DSM 117.3 -20.9Turndown 108.9 -26.6Multi-mode 153.6 3.49The total Irish and British CO 2 emissions for each scenario can be seen in Table6.6. Each scenario is seen to reduce CO 2 relative to the base case, however, the overallchanges are small. The largest CO 2 reduction occurred in scenario 1 with increasedinterconnector capacity. Emissions increased on the Irish system due to the increasedproduction to meet increased export levels, however the production that was displacedon the British system was more CO 2 intensive, thus yielding a net reduction.

Chapter 6. Options for Increasing Power System Flexibility 92Table 6.6: CO 2 emissions in each scenarioScenario CO 2 emissions Change from(Mtonnes) base case (Mtonnes)Base Case 199.4 -Interconnection 198.9 -0.5Pumped storage 199.1 -0.3DSM 199.1 -0.3Turndown 199.2 -0.2Multi-mode 199.4 06.4 Summary of ResultsThis chapter so far has investigated how commonly cited sources of power systemflexibility will interact with base-load generation on a power system with a high windenergy penetration and alleviate or aggravate plant cycling. The results have beensomewhat counter-intuitive as several of the flexibility options examined, includingstorage, were shown to contribute to plant cycling. The limited utilization of DSM,had little impact on cycling (although a large increase in cycling if it is assumed toprovide reserve). Interconnection resulted in avoided cycling for both CCGT and coalplant while, increased turndown for CCGTs was also seen to benefit CCGT operation.The decision to invest in any of these options will be based on capital costs andexpected revenues. Benefits to the power system such as a reduction in productioncosts, emissions or wind energy curtailment are often considered also. This chapter hasshown that plant cycling is another important factor to be weighed up, regardless ofwhether the effects are positive or negative, particularly considering the high cyclingcosts that have been found, as discussed in Chapter 2.

Chapter 6. Options for Increasing Power System Flexibility 936.5 Other Flexibility Options6.5.1 Battery Electric VehiclesPlug-in hybrid electric vehicles (PHEVs) and fully electric vehicles (EVs) provide anopportunity to reduce emissions and decrease the dependence of the transport sector onpetroleum products. Consequently, many countries have announced national targets forelectric vehicles, for example, the Department of Energy (D.O.E) in the US is seeking1,000,000 vehicles on the road by 2015, while in Ireland the target is for 10% of thevehicle fleet (≈250,000 vehicles) to be electrified by 2020.Plug-in hybrid electric vehicles (PHEVs) and fully electric vehicles (EVs) can alsodeliver flexibility to power systems via the energy storage capacity present in the batteries.By employing a ‘smart charging’ strategy, whereby the system operator managesthe charging of electric vehicles, the net load profile can be flattened somewhat bycharging vehicles during the valleys, as depicted in Figure 6.6. This is particularlybeneficial on a windy night when base-load units may be forced off-line to accommodatehigh wind power penetration. Thus, EVs should facilitate more base-load and lesspart-load operation from generators alleviating cycling issues and reducing emissions,as was found to be the case in Göransson et al. (2010).Figure 6.6: Illustration of load valley filling by EV charging

Chapter 6. Options for Increasing Power System Flexibility 94However, as shown in Hadley and Tsvetkova (2009) and Göransson et al. (2010), ifa significant number of these vehicles are introduced without any control over the timeof charging, i.e. a typical owner charges the vehicle on arriving home from work and thebattery is charged until full, the evening peak demand will be exaggerated, requiringmore production from peaking units and thus resulting in higher CO 2 emissions.Assuming a smart charging scheme is in place, the system operator also has theability to stop vehicle charging temporarily if, for example, wind generation on thesystem unexpectedly dropped off. Likewise, if wind generation unexpectedly pickedup the system operator (or a demand aggregator) can begin charging vehicles withdepleted or partially charged batteries. Thus, EVs can effectively deliver both positiveand negative spinning reserve (not actually ‘spinning’ but with an equivalent activationtime) to a power system (Kiviluoma and Meibom, 2011). However, it has been shownthat the marginal benefits of EVs will saturate at a point as there is a limit to theamount of reserve that is required and the amount by which the net load profile canbe flattened (Kiviluoma and Meibom, 2009).Vehicle-to-Grid (V2G) schemes have also been investigated, whereby it is possiblefor electrical energy present in the battery to be delivered to the grid. In this caseEVs can deliver positive spinning reserve by not only reducing charging, but by actuallyproviding electrical energy to the power system. However, repeatedly reversingthe flow of electricity between the battery and the grid will result in some level ofdegradation to the battery which must be taken into consideration. When this, andthe cost of the bidirectional power electronics required, were taken into considerationin Dallinger et al. (2011), it was found that it was not economical to provide positivespinning reserve from EVs by discharging the battery.6.5.2 Maintenance SchedulingAnother area where improvements in plant cycling could be gained (without the needfor costly additions to the power system) is maintenance scheduling. One of the dutiesof a system operator is to agree an outage schedule with the power producing companies

Chapter 6. Options for Increasing Power System Flexibility 95which allows each generating unit to fulfill its maintenance requirements without compromisingthe reliability of the power system. This process typically involves generatorssubmitting their outage requests for the year ahead to the system operator, who thendetermines the impact of the aggregate outage requests on system reliability, based onsome reliability index (Feng and Wang, 2010; Shahidehpour and Marwali, 2000). Theloss of load expectation (LOLE), expected duration of unmet demand (EDUD), expectedunsupplied energy or expected lack of available reserve are typical indices usedto determine the impact on system reliability (Mukerji et al., 1991). If the requestedmaintenance schedules do not cause the system reliability to fall below some definedstandard (for example, the Irish system operator uses an LOLE of 8 hours per year),they will be approved. Otherwise, if maintenance requests are causing periods of reliabilityconcern, the generator(s) involved must revise their outage request(s) in orderto preserve system reliability. Typically generators seek to schedule their outages suchthat their overall revenue is maximized, or in other words they request outages forperiods with the lowest electricity prices and hence the lowest electricity demand.Traditionally, system operators have focussed on ensuring that there is sufficientcapacity to meet demand at all times during the year. However, a system with a largewind penetration will also need to maintain a certain level of operational flexibility, inaddition to plant capacity, in order to maintain a reliable system. For example, if alarge quantity of fast-starting or fast-ramping plant is unavailable due to maintenance,a system may still have sufficient capacity available to serve the load, however, shoulda sudden drop in wind power output occur, there may not be sufficient fast responsegeneration available to compensate, or inflexible generators may be forced to operateoutside their normal operational limits. This type of operation, particularly whenrequired frequently of base-load generators, is associated with equipment deterioration,increased maintenance costs and a reduction in reliability, as discussed in Chapter 2. Byensuring that there is sufficient operating flexibility available within the generation fleetto meet net load variations at all times during the year, excessive cycling of conventionalplant may be reduced/avoided.One approach to evaluating the level of flexibility present in power systems was

Chapter 6. Options for Increasing Power System Flexibility 96discussed in Lannoye et al. (2010), in a which a new metric, the insufficient rampingresource expectation (IRRE), based on the loss of load expectation (LOLE) metric forgeneration adequacy was presented. Utilizing such a metric in conjunction with multiplenet load projections (perhaps based on historical demand and wind power data fromseveral years) would go some way to ensuring that a power system maintained sufficientflexibility throughout the year.6.5.3 Control of Wind Power OutputBy controlling the pitch angle of wind turbine blades it is possible to curtail wind poweroutput or limit its upward ramp rate. Curtailment of wind power is often viewed as anegative outcome of a system having too little flexibility. However, there are occasionswhen wind curtailment is the most economic solution to meeting demand. For example,consider a system which has forecast a surge in wind power output, followed a shorttime later by a drop-off in wind power output. If accommodating this ‘short-lived’ highpenetration of wind means switching off thermal plant, that will need to be restartedshortly afterwards when the wind penetration begins to decline, the resulting startupfuel costs, cycling costs and carbon costs may instead make it more favourableto curtail the wind power output for the short period. Ideally this is achieved bythe system operator sending a dispatch instruction to the wind generators. PresentlyBonneville Power Administration and Alberta Electric Service Operator are utilisingramp controls on wind generation under certain reliability criteria. However, somesystems do not have the ability to control wind farm output (for example in Ireland alarge proportion of the wind generation is connected to the distribution system, whichcannot be controlled by the TSO), which can result in uneconomic system operationand plant cycling.6.5.4 Market OptionsThe power output from a wind farm is variable as the energy source itself, i.e. thewind, is variable.However, the correlation in wind speeds between any two given

Chapter 6. Options for Increasing Power System Flexibility 97sites decreases as the distance between those sites increases. Thus, when the poweroutput from various wind farms dispersed over a large area is aggregated, the overallvariability is less than the variability of the individual wind farms. This indicatesthat a system which is interconnected to a neighbouring system can benefit from theprinciple of statistical independence and thus reduce the burden on its thermal plantto manage net load variability. The US is divided into 130 balancing areas, eachresponsible for matching generation to demand in that area. Many studies have shownthat consolidating some of these balancing areas can benefit the integration of variablerenewables (NREL, 2011).With a high wind penetration ‘faster’ markets are also advantageous. Currentlymany markets are settled on an hourly basis, which can restrict access to flexible resourceson the system. For example, in an hourly market a fast starting generatorcannot be started up within the hour to meet an increase in net load. Instead it wouldhave to wait until the beginning of the next hour to be dispatched, while online unitswould have to ramp their output to meet the increased net load instead. Milligan et al.(2010) finds the benefits of faster markets include greater access to flexibility and reduceda ramping requirement from conventional units.

CHAPTER 7Unit Commitment with Dynamic Cycling Costs7.1 IntroductionTHE increased levels of cycling that base-load plant will be forced to undergo dueto increasing penetrations of wind generation have been shown in Chapter 4and have also been seen in Göransson and Johnsson (2009). As discussed in Chapter2, this can lead to high levels of damage accumulating within the plant’s componentsultimately resulting in increased maintenance requirements and forced outage rates.Cycling related costs will arise via increased maintenance costs for generators, loss ofrevenue resulting from longer and more frequent outages, increased fuel costs due toreduced plant efficiency, as well as capital costs due to component replacement. Studiesindicate that the magnitude of these cycling related costs are high but, as discussedin Chapter 2, accurately quantifying them is a challenging task given the range ofcomponents affected, the unit specific nature of the analysis and the lengthy time lagthat is typically seen before cycling damage becomes apparent through component98

Chapter 7. Unit Commitment with Dynamic Cycling Costs 99failure (Lefton, 2004).Not considering these costs however will result in the uneconomic dispatch of plants,yet still markets currently do not include specific cycling cost components in theirbidding mechanisms, or at best cycling costs are bundled into a generator’s start-upor ramping costs. Depending on the operating regime of a plant, these cycling relatedcosts can accumulate rapidly and are therefore dissimilar to plant characteristics suchas heat rate, which typically vary over a much longer time-scale. Therefore, to examinethe impact of these costs accurately, they should be modelled in a dynamic mannersuch that they accumulate within the optimization process based on how the unit isbeing operated and can thereby influence dispatch decisions.This chapter presents a novel formulation that allows these cycling costs to be modelleddynamically, which can be integrated into a MIP (mixed integer programming)unit commitment and economic dispatch model. This facilitates more accurate modellingof these costs and examination of how they accumulate in line with the operatingregime of a plant. The formulation sets up a cycling cost which increments with eachadditional plant start-up or ramp, with the resulting cost function being linear, piecewiselinear or step-shaped. This new approach to modeling cycling costs is particularlysuitable for long-term planning studies where it can be used to reflect the ageing effecton a plant over time. It may also have applications for real-world dispatch modelswhere it can discourage the same unit from being repeatedly dispatched to cycle, asthis will incur an incremental cost to reflect the wear-and-tear to that unit and canconsequently alter its position in the merit order. A case study is included to determinehow implementing dynamic cycling costs over a period of one year will affect theresulting dispatch relative to a scenario where cycling costs are not considered.7.2 Formulation of Dynamic Cycling CostsA detailed formulation for implementing dynamic cycling costs which increase in linewith unit operation is presented here.Cycling costs are subdivided into costs for

Chapter 7. Unit Commitment with Dynamic Cycling Costs 100(A) start-ups and (B) ramps. The formulation utilizes three main steps: (i) a binaryvariable is set to indicate that damaging operation has occurred at time step t, (ii) acounter tracks how much of that type of operation has occurred up to that point, and(iii) an incrementing cycling cost is incurred at that time step. Linear, piece-wise linearand step-shaped cost functions for both starts and ramps are detailed here.7.2.1 Cycling Costs Related to Start-upsLinearConstraints 7.1 - 7.3 allow a dynamic, linearly incrementing cost for wear-and-tearrelated to start-ups to be modelled. Based on the online binary variable, v g (t), constraint7.1 sets the start-up, s g (t), and shut-down, z g (t), binary variables equal to 1appropriately, when a unit ‘g’ is started or shut down at time t. Constraint 7.2 incrementsa counter, Ng S (t), to track how many start-ups have been performed by thatunit. Constraint 7.3 determines the start-up related cycling cost, Cg S (t), with the finalterm ensuring that a cost is only incurred when the decision is made to start the unitat time ‘t’ (i.e. s g (t) = 1). Figure 7.1 provides an example of this linearly increasingcost function, where the incremental cost, cost S g , is set equal to 100.s g (t) − z g (t) = v g (t) − v g (t − 1), ∀ t ∈ T, ∀ g ∈ G (7.1)N S g (t) ≥ N S g (t − 1) + s g (t), ∀ t ∈ T, ∀ g ∈ G (7.2)C S g (t) ≥ N S g (t).cost S g − M. ( 1 − s g (t) ) , ∀ t ∈ T, ∀ g ∈ G (7.3)

Chapter 7. Unit Commitment with Dynamic Cycling Costs 101Figure 7.1: Linearly increasing start-up related cycling costPiecewise LinearBy defining i thresholds, T h S g (i), each corresponding to a cumulative number of plantstart-ups, at which point the start-up related cycling cost, Cg S (t), will increase byincremental cost cost S g (i) for each additional start, a piecewise linear incremental costfunction can be modelled. Constraint 7.4 is a modified form of constraint 7.2 whichcounts the cumulative number of start-ups. For i > 1, the start-up counter, Ng S (t, i),will not have a positive value until Ng S (t, 1) has reached T h S g (i). T h S g (1) must equal 1.Constraint 7.5 determines the total cycling cost. Figure 7.2 provides an example of apiecewise linearly increasing cost function, where cost S g (1) is set equal to 100, cost S g (2)is set equal to 150 and T h S g (2) equals 4.N S g (t, i) ≥()Ng S (t − 1, 1) + s g (t) + 1− T h S g (i),∀ t ∈ T, ∀ g ∈ G, ∀ i ≤ I g(7.4)C S g (t) ≥I g∑i(N S g (t, i). ( cost S g (i) − cost S g (i − 1) ))− ( 1 − s g (t) ) .M, ∀ t ∈ T, ∀ g ∈ G(7.5)

Chapter 7. Unit Commitment with Dynamic Cycling Costs 102Figure 7.2: piecewise linearly increasing start-up related cycling costStep FunctionAlternatively, if less information is known regarding the shape of the cost functionan appropriate simplification may be to define a step function, where Cg S (t) does notincrement until T h S g (i) is reached. Again, it is required that T h S g (1) is equal to 1.Ng S (t, i) is determined by constraint 7.6 and in this case can be greater than or lessthan 0 (it was previously defined as a positive variable only). Constraint 7.7 sets thebinary variable step g (t, i) equal to 1 when Ng S (t, i) has exceeded T h S g (i), and constraint7.8 determines the cycling cost. Figure 7.3 provides an example of this incrementing,step-shaped cost function, where cost S g (t, 1) is set equal to 100, cost S g (t, 2) is set equalto 150 and T h S g (2) equals 4.N S g (t, i) =()Ng S (t − 1, 1) + s g (t) + 1− T h S g (i),∀ t ∈ T, ∀ g ∈ G, ∀ i ≤ I g(7.6)N S g (t, i) − step g (t, i).M≤ 0, ∀ t ∈ T, ∀ g ∈ G, ∀ i ≤ I g (7.7)

Chapter 7. Unit Commitment with Dynamic Cycling Costs 103C S (t) ≥ cost S g (i).step g (t, i) − ( 1 − s g (t) ) .M,∀ t ∈ T, ∀ g ∈ G, ∀ i ≤ I g(7.8)Figure 7.3: Step increasing start-up related cycling costHot and Cold StartsEither the linear, piecewise linear or step formulations can be extended to differentiatebetween hot and cold start-ups for units. Constraint 7.9 will set the binary variables coldg (t) equal to 1 only if a unit is started at time t, having been offline for t cold plus itsminimum downtime, DT g . In constraints 7.2, 7.4 and 7.6 ‘+ s g (t)’ is replaced with ‘+s g (t) + α.s coldg (t)’. A scaling factor, α, is chosen based on the ratio of cycling damagecaused by a hot start relative to a cold start, and thus normalizes N S g (t, i) to count interms of hot starts.s coldg (t) ≥ v g (t) −Tgcold∑+DT gn=1v g (t − n), ∀ t ∈ T, ∀ g ∈ G (7.9)

Chapter 7. Unit Commitment with Dynamic Cycling Costs 1047.2.2 Cycling Costs Related to Ramping7.2.2.1 Define one ramp levelThe simplest form of incurring cycling costs related to ramping duty is to define achange in output, R g , between consecutive time periods, greater than which, damagingtransients will occur within the unit. Constraints 7.10 and 7.11 ensure that the binaryvariable r(t) is set to 1 when a change in output exceeding R g occurs. To avoid doublecounting cycling costs when large ramps are experienced in the start-up or shut-downprocess, the final term ensures that the constraints are non-binding when the unit is inthe start-up or shut-down process. If the ramp-related cycling costs are likely to exceedthe start-up or shut-down cost, constraint 7.12 is needed to prevent the model settings(t) and z(t) both equal to 1 in constraint 7.1, in order to make constraints 7.10 and7.11 non-binding.(pg (t) − p g (t − 1) ) − M.r g (t) ≤ R g + s g (t).M, ∀ t ∈ T, ∀ g ∈ G (7.10)(pg (t − 1) − p g (t) ) − M.r g (t) ≤ R g + z g (t).M, ∀ t ∈ T, ∀ g ∈ G (7.11)s g (t) + z g (t) ≤ 1, ∀ t ∈ T, ∀ g ∈ G (7.12)Utilizing the binary variable, r g (t), a counter is defined, as before, to incur anincrementing, ramp-related cycling cost, Cg R (t). Using the formulation from Section7.2.1, the ramp-related cycling cost function may be linear, piecewise linear or stepshaped.Constraints 7.2 and 7.3 are replaced with the analogous ramp terms shown inTable 7.1 to implement a linearly incrementing cost. Constraints 7.4 and 7.5, or 7.6to 7.8, are replaced with the analogous ramp terms as shown in Table 7.1 to define apiecewise linear, or a step shaped, incrementing ramp related cycling cost respectively.

Chapter 7. Unit Commitment with Dynamic Cycling Costs 105Table 7.1: Analogous TermsStarts Ramps Bi-directional Rampss g (t) r g (t) x g (t)Linear cost S g cost R g cost X gN S g (t) N R g (t) N X g (t)C S g (t) C R g (t) C X g (t)s g (t) r g (t) x g (t)Piecewise cost S g (i) cost R g (i) cost X g (i)Linear & N S g (t,i) N R g (t,i) N X g (t,i)Step Th S g (i) Th R g (i) Th X g (i)C S g (t) C R g (t) C X g (t)step S g (t) step R g (t) step X g (t)7.2.2.2 Define multiple ramp levelsThe previous formulation, where one level R g is set to define a ramp, can be expandedto incur a dynamic ramp-related cycling cost, for j ramps of different magnitudes,R g (j). Constraint 7.13 ensures that for a ramp less than R g (1), r g (t, j) will equal zerofor all j. A ramp greater than R g (1), but less than R g (2), will set r g (t, 1) equal to one,and so forth. The final term ensures that the constraint is non-binding when the unitis starting up. A corresponding constraint is needed for down ramps, where ( p g (t)-p g (t − 1) ) in constraint 7.13 is replaced with ( p g (t − 1)-p g (t) ) and M.s(t) is replacedwith M.z(t). Constraint 7.14 ensures that the binary variable, r g (t, j), which indicatesthat a ramp ≥ R g (j) has occurred, can only have a value of 1 for one ramp level j, atany given time. As before, constraint 7.12 is required to prevent s g (t) and z g (t) bothbeing set to 1 to make constraint 7.13 and its corresponding constraint non-binding.

Chapter 7. Unit Commitment with Dynamic Cycling Costs 106(pg (t) − p g (t − 1) ) < R g (1). ( 1 −j∑r g (t, j) ) + R g (2).r g (t, 1)j=1+... + R g (j).r g (t, j − 1) + ¯P(7.13)g .r g (t, j) + M.s g (t),where R g (1) < R g (2) < R g (j)... < ¯P g , ∀ t ∈ T, ∀ g ∈ G, ∀ j ≤ ¯j gj∑r g (t, j) ≤ 1, ∀ t ∈ T, ∀ g ∈ G, ∀ j ≤ ¯j g (7.14)j=1As with hot and cold starts, scaling factors are used to normalize Ng R (t) to count interms of one ramp level, as shown in constraint 7.15, where r(t, j) is expressed in termsof r(t, 1). Constraint 7.16 determines the total ramp-related cycling cost, shown herewith a constant cost increment, cost R g , with the final term ensuring that a cost is onlyincurred in a time period when a ramp (> R g (1)) occurs.N R g (t) = N R g (t − 1) + r g (t, 1) + β.r g (t, 2) + .... + γ.r g (t, j),∀ t ∈ T, ∀ g ∈ G, ∀ j ≤ ¯j g(7.15)Cg R (t) ≥ Ng R (t).cost R g − ( j∑1 − r g (t, j) ) .Mj=1(7.16)∀ t ∈ T, ∀ g ∈ G, ∀ j ≤ ¯j gTo combine this formulation of j ramp levels with i cost thresholds (i.e piecewiselinear) constraints 7.15 and 7.16 are replaced by constraints 7.17 and 7.18, such thatonce N R g (t, i) reaches T h R g (i), C R g (t, i) will begin incrementing by cost R g (i).

Chapter 7. Unit Commitment with Dynamic Cycling Costs 107N R g (t, i) = ( N R g (t − 1, 1) + r g (t, 1) + β.r g (t, 2)+.... + γ.r g (t, j) + 1 ) − T h R g (i)(7.17)∀ t ∈ T, ∀ g ∈ G, ∀ j ≤ ¯j g , ∀ i ≤ I gC R g (t) ≥−I g∑i(N R g (t, i). ( cost R g (i) − cost R g (i − 1) ))(7.18)j∑r g (t, j).M, ∀ t ∈ T, ∀ g ∈ G, ∀ j ≤ ¯j gj=1To include a step-shaped ramp related cycling cost function, constraints 7.6-7.8 arereplaced with the analogous terms for ramping from Table 1.7.2.2.3 Bi-directional rampsBi-directional ramping, typically experienced by a load-following unit, is thought tobe significantly more severe than ramps in one direction. A more detailed analysisof cycling costs can include costs for bi-directional ramping as follows. Constraints7.19 and 7.20 set the binary variables up g (t) and down g (t) to indicate the direction oframping. Only ramps of magnitude greater than R g are considered as there will besome level of ramping capability a generator can undertake relatively free of wear-andtear.Constraints 7.21 and 7.22 determine when a unit experiences large load changesin opposite directions between two consecutive time periods.p g (t) − p g (t − 1) ≤ R g + M.up g (t) + M.s g (t), ∀ t ∈ T, ∀ g ∈ G (7.19)

Chapter 7. Unit Commitment with Dynamic Cycling Costs 108p g (t − 1) − p g (t) ≤ R g + M.down g (t) + M.z g (t), ∀ t ∈ T, ∀ g ∈ G (7.20)up g (t) + down g (t − 1) − M.x g (t) ≤ 1, ∀ t ∈ T, ∀ g ∈ G (7.21)up g (t − 1) + down g (t) − M.x g (t) ≤ 1, ∀ t ∈ T, ∀ g ∈ G (7.22)The binary variable x g (t) can now be used to increment a counter which in turncan incur a dynamic bi-directional ramp-related cycling cost. To implement a linearlyincrementing cost function constraints 7.2 and 7.3 are replaced with the analogous rampterms shown in Table 1. Again, constraints 7.4 and 7.5, or constraints 7.6 to 7.8, arereplaced with the analogous ramp terms, as shown in Table 1, to implement a piecewiselinearly incrementing cost or a step-shaped incrementing cost respectively.If dynamic cycling costs for ramping and bi-directional ramping are implementedtogether it is necessary to avoid double counting ramping costs. This is achievedby subtracting [r(t, j).cost R (j) + r(t − 1, j).cost R (j)] from the total cycling cost forbi-directional ramping, Cg X (t), when the reverse directional ramp is detected (whenx g (t)=1).7.3 Model and Test SystemTo examine how cycling costs, modelled dynamically, will impact plant dispatch thenew formulation was implemented in a conventional MIP unit commitment model basedon Carrión and Arroyo (2006) and Arroyo and Conejo (2000). The unit commitmentproblem was formulated as

Chapter 7. Unit Commitment with Dynamic Cycling Costs 109Minimize ∑ t∈T∑g∈Gc p g(t) + c s g(t) + C S g (t) + C R g (t) (7.23)subject to∑p g (t) = D(t), ∀ t ∈ T (7.24)g∈Gp g (t) ≤ ¯P g .v g (t), ∀ t ∈ T (7.25)p g (t) ≥ P g .v g (t), ∀ t ∈ T (7.26)As per Carrión and Arroyo (2006) and illustrated by Figure 7.4, a piecewise linearapproximation of a quadratic production cost function for each unit was adopted asrepresented by:c p g(t) = A g v g (t) +NL g∑l=1F lg δ l g(t), ∀ t ∈ T, ∀ g ∈ G (7.27)p g (t) =NL g∑l=1δ l g(t) + P g v g (t), ∀ t ∈ T, ∀ g ∈ G (7.28)δ 1 (g, t) ≤ T 1g − P g , ∀ t ∈ T, ∀ g ∈ G (7.29)δ l (g, t) ≤ T lg − T l−1g , ∀ t ∈ T, ∀ g ∈ G ∀ l = 2...NL g − 1 (7.30)δ NLg (g, t) ≤ ¯P g − T NLg −1 − T l−1g , ∀ t ∈ T, ∀ g ∈ G (7.31)

Chapter 7. Unit Commitment with Dynamic Cycling Costs 110δ l (g, t) ≥ 0, ∀ t ∈ T, ∀ g ∈ G ∀ l = 1...NL g (7.32)where A g = a g + b g P g + c g P 2 g.Figure 7.4: Piecewise linear production cost (Carrión and Arroyo, 2006)Start-up costs which were dependent on the period of time the unit had been offlinewere modelled as follows:c s g(t) ≥ ( v g (t) − v g (t − 1) ) .hc g ∀ t ∈ T, ∀ g ∈ G (7.33)c s g(t) ≥ ( v g (t) −Tgcold +DT ∑ gn=1v g (t − n) ) .cc g , ∀ t ∈ T, ∀ g ∈ G (7.34)Minimum up time constraints were formulated by constraints 7.35, 7.36 and 7.37.Equation 7.35 is only included if the number of hours a unit must remain online tosatisfy its minimum uptime, B g , is greater than or equal to 1.t≤B∑ gt(1 − vg (t) ) = 0, ∀ g ∈ G (7.35)

Chapter 7. Unit Commitment with Dynamic Cycling Costs 111t+UT g −1∑n=tv g (n) ≥ UT g .s g (t), ∀ g ∈ G, ∀ t = B g + 1... ¯T − UT g + 1 (7.36)¯T∑n=t(vg (n) − s g (t) ) ≥ 0, ∀ g ∈ G, ∀ t = ¯T − UT + 2... ¯T (7.37)where B g = max ( 0, v g (T)UT g -h upg +v g (T) ) .Minimum down time constraints were formulated using constraints 7.38, 7.39 and7.40. Equation 7.35 is only included if L g ≥ 1.t≤L g∑t(vg (t) ) = 0, ∀ g ∈ G. (7.38)t+DT g−1∑n=tv g (n) ≥ DT g .z g (t), ∀ g ∈ G, ∀ t = L g + 1... ¯T − DT g + 1 (7.39)¯T∑n=t(1 − vg (n) − z g (t) ) ≥ 0, ∀ g ∈ G, ∀ t = ¯T − DT + 2... ¯T (7.40)where L g = max ( 0, (1 − v g (T)).DT g -h downg +(1 − v g (T)) ) .The formulation was applied to the 10 unit test system used in Carrión and Arroyo(2006); Kazarlis et al. (1996); Damousis et al. (2004), which was duplicated to givea 20 unit system, thus facilitating a larger case study. The technical and economiccharacteristics of these units are given in Table 7.2 and Table 7.3. (The initial state

Chapter 7. Unit Commitment with Dynamic Cycling Costs 112is the number of hours a unit is assumed to have been online for at the start of theoptimization.) The fuel cost curves for the test units are given in Appendix D. Thepeak demand (1500 MW) was doubled (3000 MW) and a historical year-long hourlydemand profile for the Irish system was scaled to produce a demand profile with a 3000MW peak. The model was run for the test year, optimizing each day at an hourlyresolution.Table 7.2: Generator DataUnits ¯Pg P g UT g DT g Initial State(MW) (MW) (h) (h) (h)1-4 455 150 8 8 85-8 130 20 5 5 -59-10 162 25 6 6 -611-12 80 20 3 3 -313-14 85 25 3 3 -315-20 55 10 1 1 -1Table 7.3: Generator production cost dataUnits a g b g c g hc g cc g t coldg($/h) ($/MWh) ($/MW 2 h) ($/h) ($/h) (h)1-2 1000 16.19 0.00048 4500 9000 53-4 970 17.26 0.00031 5000 10000 55-6 700 16.60 0.00200 550 1100 47-8 680 16.50 0.00211 560 1120 49-10 450 19.70 0.00398 900 1800 411-12 370 22.26 0.00712 170 340 213-14 480 27.74 0.00079 260 520 215-16 660 25.92 0.00413 30 60 017-18 665 27.27 0.00222 30 60 019-20 670 27.79 0.00173 30 60 0Generator cycling costs are difficult to determine and largely uncertain as discussedin Section I. The figures used here, shown in Table 7.4, to implement dynamic cyclingcosts for the test system, are a conservative assumption based on those shownin Lefton et al. (2006) and are intended to illustrate how dynamic cycling costs could

Chapter 7. Unit Commitment with Dynamic Cycling Costs 113impact system operation, rather than provide an accurate estimate of such costs.Piecewise linear costs for starts and ramps were implemented with the incrementalcost (cost S g (i) or cost R g (i)) increasing by 10% and 20% when the start counter (Ng S (t, 1)),or ramp counter (Ng R (t, 1)), exceeded 100 (T h S g (2) or T h R g (2)) and 200 (T h S g (3) orT h R g (3)) respectively. The scaling factor, α, was chosen to be 2, i.e. each cold startincremented Ng S (t, 1) by 2 (while a hot start incremented Ng S (t, 1) by 1). Two ramplevels, R g (1) and R g (2) corresponding to 20% and 40% of the difference between maximumand minimum output for a unit, were modelled. Scaling factors were chosen suchthat ramps greater than R g (1) or R g (2) incremented Ng R (t, 1) by 1 or 2 respectively.Table 7.4: Incremental cycling costs $, (i=1)Units cost S g (i) cost R g (i)Base-load (Units 1-4) 300 15Mid-merit (Units 5-10) 60 3Peaking (Units 11-20) 30 1.57.4 ResultsThis section examines how plant dispatches are affected when (i) a cycling cost relatedto start-ups is implemented, (ii) a cycling cost related to ramping is implemented and(iii) cycling costs related to start-ups and ramping are implemented simultaneously.7.4.1 Start-up Related Cycling Costs ResultsImplementing a dynamic cycling cost for plant start-ups, as shown in Table 7.4, wasseen to result in an overall reduction in plant start-ups. This is seen in Table 7.5,which reveals reducing starts for base-load and mid-merit units. For base-load units,the reduction in starts was correlated with increased production as, having the largestincremental cycling costs, these units avoided shut-downs and gained more online hours.This is seen via the average capacity factor shown in Table 7.6. Mid-merit units how-

Chapter 7. Unit Commitment with Dynamic Cycling Costs 114ever, who also had reduced starts, saw reduced production indicating that they wereutilised less often. As these units were started up and shut down, and subsequentlyincurred cycling costs, it became more economical after some point to dispatch peakingunits. Thus, starts and production increased for peaking units when a dynamic cyclingcost for start-ups was modelled. Figure 7.5 illustrates the cumulative start-ups for themid-merit and peaking units over the year when (i) cycling costs were modelled and (ii)when cycling costs were not modelled. Starts are seen to accumulate rapidly between 0and 2000 hours and from hours greater than 7000, as these are the winter months andthus have higher demand, requiring more plant start-ups. Up to 1000 hours, the level ofcycling costs incurred by the mid-merit and peaking units is seen to have no impact onthe number of start-ups. However, beyond 1000 hours the cycling costs which are accumulatedby mid-merit begins to have an effect on their position in the merit order andconsequently peaking plant are seen to be dispatched more frequently. Modelling dynamiccycling costs related to plant start-ups was also found to have the knock on effectof increasing generator ramping. Over the year a 22% increase in ramping (Ng R (t, 1))was observed relative to the case when no cycling costs were modelled as generatorswere more frequently ramped down to minimum output, rather than shut-down, in aneffort to avoid the increasing cycling costs.Table 7.5: Impact of dynamic cycling costs for start-ups on total annual startsNo cycling Cycling cost forUnits costs modeled starts modeledBase-load (Units 1-4) 34 12Mid-merit (Units 5-10) 1372 1005Peaking (Units 11-20) 577 838Total 1983 1855Units within the same class, i.e. base-load, mid-merit or peaking, were also seen toconverge to a similar number of annual start-ups, as indicated by the reduced standarddeviation of annual start-ups seen in Table 7.7. This indicates that once a unit hasbeen cycled and its cycling cost is incremented, the next time a unit needs to be cycledthe costs will have now changed such that a different unit (most likely the next in the

Chapter 7. Unit Commitment with Dynamic Cycling Costs 115Table 7.6: Impact of dynamic cycling costs for start-ups on average plant capacityfactors (%)No cyclingCycling cost forUnits costs modeled starts modeledBase-load (Units 1-4) 92.59 92.73Mid-merit (Units 5-10) 27.82 25.42Peaking (Units 11-20) 0.85 2.23Figure 7.5: Cumulative plant start-ups over the year, shown when dynamic cyclingcosts for starts were (i) modelled and (ii) not modelledmerit order) may be scheduled. This leads to the burden of cycling operation beingmore evenly distributed across the units. Over a long horizon, i.e. several years, thiseffect can lead to a shift in the merit order, a trend which is somewhat emerging inFigure 7.5.To facilitate a sensitivity analysis, multiples of the initial incremental cycling costs,cost S g (1), that were shown in Table 7.4, were also examined. As the incremental costwas increased the reduction in start stop cycling that is achieved by modelling dynamiccycling costs quickly saturated as seen in Figure 7.6, thus indicating that the majorityof plant cycling is unavoidable. Table 7.8 shows a breakdown of the total number of

Chapter 7. Unit Commitment with Dynamic Cycling Costs 116Table 7.7: Impact of dynamic cycling costs on plant start-ups by unit typeNo cycling Cycling cost forcost modelled starts modelledUnits Avg Std. Dev Avg Std. DevBase-load (Units 1-4) 8.5 9.9 3 3.6Mid-merit (Units 5-10) 228.7 75.7 167.5 26.1Peaking (Units 11-20) 57.7 73.1 83.8 27.5starts by unit group, which again reveals that increasing starts for peaking units arecorrelated with increasing incremental cycling cost, as it becomes more favourable todispatch these units due to the relatively larger cycling costs associated with the midmeritunits. (The ripples in the curve shown in Figure 7.6 result from the increasingstarts for peaking units, as seen in Table 7.8.)Figure 7.6: Impact of dynamic cycling cost on total start-ups, shown for various multiplesof cost S g (i)A scenario where cycling costs were only modelled for a subset of the total fleetwas also examined. The 6 largest units on the system (units 1, 2, 3, 4, 9, 10) werechosen based on the assumption that these units would be most impacted by cyclingoperation and thus most likely to bid a wear-and-tear cost into the market to reflectthis. The results showed that although the number of annual start-ups was reduced forthese units, the start-ups for other units increased by an amount much greater than

Chapter 7. Unit Commitment with Dynamic Cycling Costs 117Table 7.8: Impact of dynamic cycling costs for starts on total plant start-ups, shownfor various multiples of cost S g (i)Base-load Mid-merit PeakingUnits 1-4 Units 5-10 Units 11-20No cycling cost 34 1372 577cost S g (i)*0.5 13 1104 781cost S g (i)*1 12 1005 838cost S g (i)*2 13 941 896cost S g (i)*3 13 907 948cost S g (i)*10 13 869 992the reduction achieved for the units which bid a cycling cost, as seen in Table 7.9. Thiswould indicate the need for a uniform policy relating to the bidding of cycling costs tobe implemented in markets, such that all units reflect their cycling costs, or do not, toavoid the situation where only some generators are bidding cycling costs which leadsto inefficient operation and excessive costs.Table 7.9: Change in starts when a subset of units bid cycling costs for start-ups∆ StartsUnits 1, 2, 3, 4, 9, 10 -86All other units +2567.4.2 Ramping Related Cycling Costs ResultsImplementing a dynamic cycling cost for plant ramping (shown in Table 7.4) resultedin a 90% reduction in ramping overall as seen in Table 7.10. As described previously,assuming a ramp greater than 20% or 40% of the difference between a unit’s maximumand minimum output increments the ramp counter, Ng R (t), by a value of 1 or2 respectively. The total value of Ng R (t) at the end of the test year, summed for allunits, is shown in Table 7.10. Base-load units which carried out the greatest amount oframping when cycling costs were not modelled, saw the greatest reduction in ramping

Chapter 7. Unit Commitment with Dynamic Cycling Costs 118operation when cycling costs for ramps were implemented. The drastic reduction inramping that was achieved by implementing dynamic ramping costs, however, led toincreased start-stop cycling as might be expected, although only by 3.3% over the year.The most notable change to the overall dispatch that resulted from the introductionof dynamic ramping costs was a slight reduction in production from base-load plantallowing for increased production from mid-merit and peaking units as seen in Table7.11, thereby spreading the ramping requirement over more units. Thus, including theramping cost was also seen to result in a slightly greater number of units online (5.94per hour on average when dynamic ramping costs were modelled, versus 5.92 when nocycling costs were modelled).Table 7.10: Impact of dynamic cycling costs for ramping on total annual ramping(N R g (t, 1))No cycling Cycling cost forUnits costs modeled ramps modeledBase-load (Units 1-4) 3717 120Mid-merit (Units 5-10) 2214 1224Peaking (Units 11-20) 795 623Total ramping 6726 1967Table 7.11: Impact of dynamic cycling costs for ramping on average plant capacityfactors (%)No cyclingCycling cost forUnits costs modeled ramps modeledBase-load (Units 1-4) 92.59 92.21Mid merit (Units 5-10) 27.82 28.61Peaking (Units 11-20) 0.85 1.027.4.3 Start-up and Ramping Cycling Costs ResultsImplementing dynamic cycling costs (as shown in Table 7.4) for starts and rampingsimultaneously, reduced both types of cycling operation relative to the case when no

Chapter 7. Unit Commitment with Dynamic Cycling Costs 119cycling costs were modelled, as shown in Table 7.12. Base-load units, having the largestcycling costs, see the greatest reductions in cycling operation.Nonetheless, neithertotal starts nor total ramps were reduced in this scenario as much as starts aloneor ramps alone were reduced when cycling costs for starts or ramps were modelledindividually. However, when cycling costs for start-ups only were modelled, rampingoperation increased and likewise when cycling costs for ramping only were modelled,starts increased, thus when the cycling costs that would have been incurred, assumingthe costs given in Table 7.4 increment as described in Section 7.3, the case in whichcycling costs for start-ups and ramping were modelled simultaneously had the lowestoverall cycling costs, as shown in Figure 7.7. This would indicate that modelling cyclingcosts for starts and ramping simultaneously most cost effectively reduces cycling andas such one should not be considered without the other.Table 7.12: Impact on total annual starts and ramps when dynamic cycling costs forboth start-ups and ramping were modelledNo cycling costsCycling cost for startsUnits modeled and ramps modeledStarts Ramps Starts RampsBase-load (Units 1-4) 34 3717 12 144Mid merit (Units 5-10) 1372 2214 1003 2069Peaking (Units 11-20) 577 795 855 1456Total 1983 6726 1870 3669Finally, when total system costs are examined for the scenario including cycling costsand compared to the total system cost for the scenario in which cycling costs were notmodeled, but were calculated and added afterwards, it can be seen that modelingcycling costs leads to lower system costs overall. This is shown in Figure 7.8. In thisexample, the cost saving seen is considerable i.e. 14%.

Chapter 7. Unit Commitment with Dynamic Cycling Costs 120Figure 7.7: Cycling costs (that would have been incurred) shown for various scenarios7.5 SummaryFigure 7.8: Total system costs shown for various scenariosInterest concerning cycling costs is growing and this paper sets out a formulation thatcan utilize knowledge of incremental wear-and-tear costs related to plant start-ups orramping, to implement a dynamic incrementing cycling cost. The formulation coverslinear, piecewise linear and step-shaped cycling cost functions, the appropriate choicefor a user being determined by the level of knowledge of the generator’s cycling costs.The formulation for piecewise linear incremental cycling costs related to plant start-

Chapter 7. Unit Commitment with Dynamic Cycling Costs 121ups and ramps was implemented for a test system. Although the incremental costschosen are approximations, the results reveal certain trends that are likely for powersystems where generators undergo regular cycling and reflect the resulting wear-andtearcosts in their bids. For example, dynamically modeling cycling costs for generatorstarts was seen to reduce the number of starts, but caused ramping operation to beincreased (and vice-versa), whilst modeling cycling costs for only a subset of the generationfleet was seen to induce much higher levels of cycling in the remaining generation.It was also seen that as cycling costs accumulated over time changes in the merit orderoccurred, and that modeling cycling costs led to an overall saving for the system ascycling operation was subsequently reduced.

CHAPTER 8ConclusionsTHIS thesis presented research related to the cycling of base-load generation withincreasing penetrations of wind energy on a power system. In Chapter 1 theevolution of power systems to incorporate higher levels of wind generation against abackground of deregulation and increased competition is discussed.The likelihoodof increased generator cycling resulting has been found in many studies, such as GE(2010); NREL (2010); NYISO (2010), and is beginning to become apparent in real worldsystems (MMU, 2010). The physical consequences for increased cycling are explored inChapter 2 and thus provides the motivation for this research.Chapter 4 outlined how the operation of CCGT and coal units will be impacted byincreasing levels of wind generation on a power system. Base-load CCGT units wereseen to undergo a large increase in start-stop cycling as wind penetration increased,while coal units, being the most base-load generation, tended to remain online butwere subject to increased ramping and part-load operation. Thus, both CCGT andcoal units would be expected to experience increasing costs and forced outage rates122

Chapter 8. Conclusions 123over time due to wear and degradation of components from cycling operation.Sensitivity analyses were conducted to examine the level of cycling occurring whenstorage and interconnection were removed (individually) from the system. The resultsshowed reduced cycling for base-load plant in both cases. Without storage on the system,there is an increased requirement on base-load units to be online providing reserveto the system, resulting in reduced start-stop cycling, while without interconnection theentire system demand must be met domestically yielding increased production from andreduced cycling of base-load units.Having observed the decreasing production and online hours for CCGT units, Chapter5 examined a new mode of operation for these units. Many CCGT units are fittedwith a bypass stack which allows the steam cycle to be bypassed and the gas turbine tobe run in open-cycle mode; a highly flexible, although less efficient, mode of operation.The benefits of allowing CCGTs to operate in this manner, when technically possibleand economically optimal, included increased availability of replacement reserve. Productionfrom peaking plant was also seen to be displaced when multi-mode operation ofCCGTs was introduced, indicating a reduced need for these units to be built and consequentlya saving to society. The results also showed that low-merit CCGTs utilizedthe multi-mode function more than high-merit CCGTs, as they are frequently offlineand available for dispatch, whilst the increased competition among generators, typicalat higher levels of wind generation, resulted in multi-mode operation of CCGTs beingutilized less frequently.Chapter 6 examined how incorporating various sources of flexibility onto a powersystem would impact cycling of base-load units and interestingly some were found tohave negative impacts on plant cycling. Pumped storage and DSM (assuming it providedreserve) increased coal cycling as the requirement on these units to remain onlinefor reserve provision was reduced. Interconnection and lower minimum operating levelsfor CCGT units were found to reduce CCGT cycling and yield increased productionfor these units.Chapter 7 presented a novel formulation to allow cycling costs to be represented in

Chapter 8. Conclusions 124a dynamic manner. Implementation of this formulation in a unit commitment modelallowed a case study to be conducted. The results showed that modelling dynamiccycling costs will result in a reduction in cycling operation, however, if cycling costs aremodelled for a subset of generation only (the 6 largest units on the test system in thiscase), the resulting level of cycling is significantly higher than the case when no cyclingcosts were modelled. This indicates the importance of a uniform approach to biddingcycling costs in electricity markets. It was also found that as cycling costs accumulatedover time, changes in the merit order became apparent. Specifically, as mid-merit unitswere started up and shut down, and subsequently accumulated cycling costs, it wasfound that after some point it became more economical to dispatch peaking units,which had lower incremental cycling costs. This highlights the importance of investingin flexible generation and retrofitting existing plant to be more capable of frequentcycling.8.1 Future WorkThe analysis completed in Chapter 4, which examined the impact of increasing windpenetrations on the operation of base-load plant, was conducted with an hourly timeresolution model, ie. the Wilmar Planning Tool. Each of the generators modelled onthe test system was capable of ramping from its minimum to maximum output (or viceversa) in under one hour, so ramp rate constraints were non-binding. However, at atime resolution under one hour modelling generator ramp rates would almost certainlyhave an impact on the resulting dispatch, particularly as wind energy penetrationincreases and the magnitude of net load ramps also increase. Consequently, a newversion of the Wilmar model, which operates with a 15 minute time step, has beenin development at the Electricity Research Centre in conjunction with this work. Achange to the structure of the model requires a change to the structure of the scenariotrees which are inputted into the model. Thus, an updated Scenario Tree Tool is alsobeing developed which will allow greater flexibility in making alterations to the modelstructure, such as the time step, frequency of rolling planning, length of optimization

Chapter 8. Conclusions 125Figure 8.1: CO 2 emissions increase linearly with productionhorizon or the number of branches in the scenario tree. Future work should utilise thisnew version of the model to analyse if generator cycling is currently underestimatedusing an hourly time step.The analysis completed in Chapters 4 to 6 included estimates of the CO 2 emissionsfrom generators based on the fuel consumption of the generators. Each fuel type wasassigned a carbon content (tonnes/GJ) and this was used to determine the CO 2 emissionsof the fuel consumed (GJ) by each generator in each hour. Thus, CO 2 emissionsincreased linearly (or piece-wise linearly if multiple heat rate slopes were modelled) asproduction from a generator (and therefore fuel consumption) increased, as shown inFigure 8.1 for a CCGT unit from the test system described in Chapter 3.However, in reality generator fuel consumption, and thus emissions, are nonlinearand Figure 8.1 represents a common modelling simplification for linear models. Motivatedby the need to understand the link between emissions and generator cycling,recent work conducted at NREL (as part of the Western Wind and Solar Phase 2 Study)has utilised CEMs data to analyse the emissions from generators at various levels ofproduction. The results of this work determines the increase in emissions or ‘emissionspenalty’ that is incurred, relative to one hour of full-load operation, when a unit is(i) operated at part-load (defined as 50% of max generation), (ii) ramped (defined as

Chapter 8. Conclusions 1265% capacity change in one hour) and (iii) started-up (Brinkman, 2011). The findingsare detailed in Table 8.1. Future work could perform a similar analysis on emissionspenalties using data for generators on the Irish system. Reproducing Table 8.1 for theIrish system would allow for more accurate analysis of the impact of generation cyclingon system emissions. Also as CO 2 costs can represent almost a quarter of total systemcosts, more detailed analysis of CO 2 costs is warranted.Table 8.1: CO 2 emissions penalties for cycling operation (Brinkman, 2011)Unit type Part-load Ramping Start-uppenalty penalty penaltyCoal 5.1% 0.4% 110%CCGT 15.6% 0.3% 32%OCGT 12.4% 0.3% 32%A technical approach has been taken in this thesis to examine the issue of baseloadcycling with increasing wind penetration. However, many interesting policy andmarket design issues have been indirectly raised, for example the fact that variablewind generation may perversely support inflexible generation or that generators mayseek to bid cycling costs into the market and by doing so can avoid cycling operation.If the goal for power systems is to achieve a high penetration of renewable generation,in order to improve security of supply and reduce emissions without compromisingsystem reliability, the portfolio of conventional generation will need to become moreflexible, which may require incentives. The evolution of existing portfolios into moreflexible portfolios, in light of these concerns, is an interesting research area that warrantsinvestigation.

ReferencesAIGS, 2008. All Island Renewable Grid Study, Workstream 2B, [Online] Available:http://www.dcenr.gov.ie.Al-Sunaidy, A., Green, R., 2006. Electricity deregulation in OECD (Organization forEconomic Cooperation and Development) countries. Energy 31 (6-7), 769–787.Albright, D., Albright, D., Albright, J., 1999. Generator field winding shorted turndetection technology. Generatortech, Inc.Anderson, R., van Ballegooyen, H., 2003. Steam turbine bypass systems.Combined Cycle Journal, Fourth Quarter, [Online] Available:http://www.psimedia.info/CCJ.htm.APPrO, 2006. The Association of Power Producers of Ontario, Adopting a RampCharge to Improve Performance of the Ontario Market, [Online] Available:http://www.ieso.ca/imoweb/pubs/consult/mep/MP WG-20060707-ramp-cost.pdf.Arroyo, J., Conejo, A., 2000. Optimal response of a thermal unit to an electricity spotmarket. IEEE Transactions on Power Systems 15 (3), 1098–1104.Arroyo, J., Conejo, A., 2004. Modeling of start-up and shut-down power trajectories ofthermal units. IEEE Transactions on Power Systems 19 (3), 1562–1568.AWEA, 2011a. American Wind Energy Association, U.S. wind industrycontinues growth, Press Release, [Online] Available:http://www.awea.org/newsroom/pressreleases/release 07APR11 .cfm.AWEA, 2011b. American Wind Energy Association, U.S. Wind Industry First Quarter2011 Market Report, [Online] Available: http://www.awea.org.Axford, M., 2009. Recession reduces demand for electricity. Power, [Online] Available:http://www.powermag.com.127

References 128Balling, L., Hofmann, D., 2007. Fast cycling towards bigger profits. Modern powersystems, [Online] Available: http://www.modernpowersystems.com.Bird, L., Bolinger, M., Gagliano, T., Wiser, R., Brown, M., Parsons, B., 2005. Policiesand market factors driving wind power development in the United States. EnergyPolicy 33 (11), 1397–1407.Bixby, R., Fenelon, M., Gu, Z., Rothberg, E., Wunderling, R., 2000. MIP: Theoryand Practice Closing the Gap. In: Proc. of 19th IFIP TC7 Conference on systemmodelling and optimization. Vol. 174. pp. 19–49.Black & Veatch, 2011. 2011 Strategic Directions Survey Results, [Online] Available:http://www.bv.com/Downloads/Resources/Brochures/.Blevins, B., 2007. Combined-cycle unit modeling in the nodal design, [Online] Available:http://www.ercot.com.Bloomberg, 2011. Bloomberg New Energy Finance, Wind turbine pricesfall to their lowest in recent years, Press Release, [Online] Available:http://www.bnef.com/PressReleases/view/139.Braun, M., 2004. Environmental external costs from power generation by renewableenergies. Thesis, Stuttgart University, Stuttgart, Germany.Brinkman, G., 2011. NREL, Wind impact on emissions?, [Online] Available:http://www.uwig.org/kcworkshop/Brinkman-KCWork.pdf.Brown, P., Lopes, J., Matos, M., May 2008. Optimization of pumped storage capacityin an isolated power system with large renewable penetration. IEEE Transactions onPower Systems 23 (2), 523–531.Brown, T. B., 1994. Assessing the effect of thermal transients on the life of boiler plant.In: Proc. of International Conference on Life Management of Power Plants. pp.137–143.CAISO, 2010a. California ISO, Integration of renewable resources - operational requirementsand generation fleet capability at 20 percent RPS, [Online] Available:http://www.caiso.com/2804/2804d036401f0ex.html.CAISO, 2010b. California ISO, Multi-stage generating (MSG) unit modeling, [Online]Available: http://www.caiso.com/2078/2078908392d0.html.Carrino, A., Jones, R., 2011. Coal plants challenged as gas plants surge. Power, [Online]Available: http://www.powermag.com/.Carrión, M., Arroyo, J., 2006. A computationally efficient mixed-integer linear formulationfor the thermal unit commitment problem. IEEE Transactions on PowerSystems 21 (3), 1371–1378.Castano, I., 2011. China installing wind-power capacity as fast as it can [Online] Available:http://www.renewableenergyworld.com.

References 129CER, 2010. Commission for Energy Regulation, Redpoint Validated ForecastModel and PLEXOS Validation Report 2010, [Online] Available:http://www.allislandproject.org.Charles River Associates, 2010. Spp witf wind integration study. prepared for SouthwestPower Pool, [Online] Available: http://www.uwig.org.Colpier, U., Cornland, D., 2002. The economics of the combined cycle gas turbine- anexperience curve analysis. Energy Policy 30 (4), 309–316.Dallinger, D., Krampe, D., Wietschel, M., 2011. Vehicle-to-grid regulation reservesbased on a dynamic simulation of mobility behavior. IEEE Transactions on SmartGrid (available online).Damousis, I., Bakirtzis, A., Dokopoulos, P., 2004. A solution to the unit-commitmentproblem using integer-coded genetic algorithm. IEEE Transactions on Power Systems19 (2), 1165–1172.Danneman, E., Beuning, S., 2011. Wind integration[Online] Available: http://www.energy-tech.com.System and generation issues,Dany, G., 2001. Power reserve in interconnected systems with high wind power production.In: Proc. of IEEE PowerTech Conference. Vol. 4.Denny, E., O’Malley, M., 2007. Quantifying the total net benefits of grid integratedwind. IEEE Transactions on Power Systems 22 (2), 605–615.DOE, 2009. United States Department of Energy, Stateswith Renewable Portfolio Standards, [Online] Available:http://apps1.eere.energy.gov/states/maps/renewable portfolio states.cfm.Doherty, R., O’Malley, M., 2005. A new approach to quantify reserve demand in systemswith significant installed wind capacity. IEEE Transactions on Power Systems 20 (2),587–595.Dupacova, J., Growe-Kuska, N., Romisch, W., 2003. Scenario reduction in stochasticprogramming: An approach using probability metrics. Mathematical Programming95 (3), 493–511.EirGrid, 2009. Generation Adequacy Report 2010 - 2016, [Online] Available:http://www.eirgrid.com.EirGrid, SONI, 2010a. All Island TSO Facilitation of Renewbles Studies, [Online] Available:http://www.eirgrid.com/renewables/facilitationofrenewables/.EirGrid, SONI, 2010b. Solver Choice in the SEM: A Comparative Study ofLagrangian Relaxation vs. Mixed Integer Programming, [Online] Available:http://www.sem-o.com.Energy-Tech, 2004. Cycling of Combined-Cycle Plants, [Online] Available:http://www.energy-tech.com.

References 130EPRI, 2001a. Correlating cycle duty with cost at fossil fuel power plants. EPRI, PaloAlto, CA. 1004010.EPRI, 2001b. Damage to power plants due to cycling. EPRI, Palo Alto, CA. 1001507.EPRI, 2002. Determining the cost of cycling and varied load operations: Methodology.EPRI, Palo Alto, CA. 1004412.Eskom, 2007. Eskom Holdings Ltd., Ankerlig power station conversion and transmissionintegration project, [Online] Available: http://www.eskom.co.za/.EU, 2008. European Union, Climate change: Commission welcomes final adoptionof Europe’s climate and energy package, Press Release, [Online] Available:http://europa.eu/rapid/pressReleasesAction.do?reference=IP/08/1998.EWEA, 2011a. European Wind Energy Association, Large Scale Integration of WindEnergy in the European Power Supply: Analysis, Issues and Recommendations, [Online]Available: Availablehttp://www.ewea.org/index.php?id=178.EWEA, 2011b. European Wind Energy Association, Wind Energy and the Grid, [Online]Available: http://www.ewea.org/index.php?id=196.EWEA, 2011c. European Wind Energy Association, Wind in power, 2010 Europeanstatistics, [Online] Available: http://ewea.org/index.php?id=1665.Feng, C., Wang, X., 2010. A competitive mechanism of unit maintenance scheduling ina deregulated environment. IEEE Transactions on Power Systems 25 (1), 351–359.Flynn, M., Walsh, M., O’Malley, M., 2000. Efficient use of generator resources in emergingelectricity markets. IEEE Transactions on Power Systems 15 (1), 241–249.French, D. N., 1993. Metallurgical failures in fossil fired boilers. John Wiley & SonsInc.GE, 2005. GE Energy and Consulting, The effects of integrating wind poweron transmission system planning, reliability, and operations, prepared for NewYork State Energy Research and Development Authority, [Online] Available:http://www.nyserda.org/publications.GE, 2010. GE Energy Applications & Systems Engineering and EnerNexCorporation and AWS Truepower, New England Wind IntegrationStudy, prepared for ISO New England, [Online] Available:http://www.uwig.org/CRA SPP WITF Wind Integration Study Final Report.pdf.GE, 2011. General Electric, FlexEfficiency Power Plant Technology, [Online] Available:http://www.ge-flexibility.com/index.jsp.Göransson, L., 2008. Wind power in thermal power systems - large-scale integration.Licentiate thesis, Dept. of Energy and Environment, Chalmers University of Technology,Goteburg, Sweeden.

References 131Göransson, L., Johnsson, F., 2009. Dispatch modeling of a regional power generationsystem-integrating wind power. Renewable Energy 34 (4), 1040–1049.Göransson, L., Karlsson, S., Johnsson, F., 2010. Integration of plug-in hybrid electricvehicles in a regional wind-thermal power system. Energy Policy 38 (10), 5482–5492.Hadley, S., Tsvetkova, A., 2009. Potential impacts of plug-in hybrid electric vehicles onregional power generation. The Electricity Journal 22 (10), 56–68.Hamidi, V., Robinson, F., 2008. Responsive demand in networks with high penetrationof wind power. In: Proc. of IEEE/PES Transmission and Distribution Conferenceand Exposition.Hatch, 2008. Nova Scotia Wind Integration Study prepared for Nova Scotia Departmentof Energy, [Online] Available: http://www.gov.ns.ca.Holttinen, H., 2005. Impact of hourly wind power variations on the system operationin the Nordic countries. Wind Energy 8 (2), 197–218.Holttinen, H., Milligan, M., Kirby, B., Acker, T., Neimane, V., Molinski, T., 2008.Using standard deviation as a measure of increased operational reserve requirementfor wind power. Wind Engineering 32 (4), 355–377.Holttinen, H., Pedersen, J., 2003. The effect of large-scale wind power on a thermalsystem operation. In: Proc. of 4th International Workshop on Large-scale Integrationof Wind Power for Offshore Wind Farms.IEA, 2008. International Energy Agency, Empowering variable renewables, options forflexible electricity systems, [Online] Available: http://www.iea.org.IEA, 2010. International Energy Agency, IEA WindEnergy, Annual Report 2009, [Online] Available:http://www.ieawind.org/AnnualReports PDF/2009/2009AR 92210.pdf.Kazarlis, S., Bakirtzis, A., Petridis, V., 1996. A genetic algorithm solution to the unitcommitment problem. IEEE Transactions on Power Systems 11 (1), 83–92.Keane, A., Tuohy, A., Meibom, P., Denny, E., Flynn, D., Mullane, A., O’Malley, M.,2011. Demand side resource operation on the Irish power system with high windpower penetration. Energy Policy 39 (5), 2925 – 2934.Kehlhofer, R., Rukes, B., Hannemann, F., Stirnimann, F., 2009. Combined-cycle gas& steam turbine power plants. Pennwell Books.KEMA, 2005. A Scoping Study: Demand Side Measures for Small Business and ResidentialCustomers on Ireland’s Electrical System prepared for Sustainable EnergyIreland, [Online] Available: http://www.seai.ie.King, J., 1996. Recent experience in condition assessments of boiler header componentsand supports. In: Proc. of ASME Pressure Vessels and Piping Conference.

References 132Kirby, B., Milligan, M., 2008. Facilitating wind development: The importance of electricindustry structure. The Electricity Journal 21 (3), 40–54.Kitto Jr, J., Bryk, S., Piepho, J., 1996. Upgrades and enhancements for competitivecoal-fired boiler systems. In: Proc. of the 1996 International Joint Power GenerationConference.Kiviluoma, J., Meibom, P., 2009. Coping with wind power variability: How plug-inelectric vehicles could help. In: Proc. of the 8th International Workshop on Large-Scale Integration of Wind Power into Power Systems as well as on TransmissionNetworks for Offshore Wind Farms. pp. 336–340.Kiviluoma, J., Meibom, P., 2011. Methodology for modelling plug-in electric vehicles inthe power system and cost estimates for a system with either smart or dumb electricvehicles. Energy 36 (3), 1758–1767.Lannoye, E., Milligan, M., Adams, J., Tuohy, A., Chandler, H., Flynn, D., O’Malley,M., 2010. Integration of variable generation: Capacity value and evaluation of flexibility.In: Proc. of IEEE Power and Energy Society General Meeting.Lefton, S., 2004. Profitable operation requires knowing how much itcosts to cycle your unit. Combined Cycle Journal, [Online] Available:http://www.combinedcyclejournal.com/.Lefton, S., 2011. Power plant asset management Presented atUWIG Spring Workshop, Kansas City, [Online] Available:http://www.uwig.org/kcworkshop/Lefton-KCWork.pdf.Lefton, S., Besuner, P., 2006. The cost of cycling coal fired power plants. Coal PowerMagazine, pp. 16–20.Lefton, S., Besuner, P., Agan, D., 2006. The real cost of on/off cycling. Modern powersystems, [Online] Available: http://www.modernpowersystems.com.Lefton, S., Besuner, P., Grimsrud, G., 1995. Managing utility power plant assets toeconomically optimize power plant cycling costs, life, and reliability. In: Proc. of 4thIEEE Conference on Control Applications. pp. 195–208.Lefton, S., Besuner, P., Grimsrud, G., 1997. Cycling fossil fired units proves costlybusiness. Electric Light and Power 75 (7).Lefton, S., Besuner, P., Grimsrud, P., Bissel, A., Norman, G., 1998. Optimizing powerplant cycling operations while reducing generating plant damage and costs at the IrishElectricity Supply Board, Aptech Engineering Service Technical Report (TP123),[Online] Available: http://forgoodpower.com/technical papers.html.Lu, B., Shahidehpour, M., 2004. Short-term scheduling of combined cycle units. IEEETransactions on Power Systems 19 (3), 1616–1625.Malik, A., 2001. Modelling and economic analysis of DSM programs in generationplanning. International Journal of Electrical Power & Energy Systems 23 (5), 413–419.

References 133Meibom, P., 2006. WILMAR - Wind Power Integration in Liberalised Electricity Markets,[Online] Available: http://www.wilmar.risoe.dk/Results.htm.Meibom, P., Barth, R., Brand, H., O’Malley, M., 2011. Stochastic optimization modelto study the operational impacts of high wind penetrations in ireland. IEEE Transactionson Power Systems (available online).Meibom, P., Weber, C., Barth, R., Brand, H., 2009. Operational costs induced by fluctuatingwind power production in germany and scandinavia. IET Renewable PowerGeneration 3 (1), 75–83.Milligan, M., Kirby, B., Beuning, S., 2010. Combining Balancing Areas’ Variability:Impacts on Wind Integration in the Western Interconnection. National RenewableEnergy Laboratory.MMU, 2010. Market Monitoring Unit, Power Plant Cyling in SEM, [Online] Available:http://www.allislandproject.org/en/mmu decision documents.aspx.Möhrlen, C., Jørgensen, J., Pinson, P., Madsen, H., Runge Kristoffersen, J., 2007. HighResolution Ensemble for Horns Rev: A project overview. In: Proc. European offshorewind energy conference.Monteiro, C., Bessa, R., Miranda, V., Botterud, A., Wang, J., Conzelmann, G., 2009.Wind power forecasting: State-of-the-art 2009 Argonne National Laboratory (ANL),[Online] Available: http://www.dis.anl.gov/pubs/65613.pdf.Moore, W., 2006. Include generators and exciters in your outage inspections. Power,[Online] Available: http://www.powermag.com.Mukerji, R., Merrill, H., Erickson, B., Parker, J., Friedman, R., 1991. Power plantmaintenance scheduling: optimizing economics and reliability. IEEE Transactionson Power Systems 6 (2), 476–483.Narula, R., Massy, M., Singh, J., 2002. Design Consideration for Combined CyclePlants for the Deregulated Market - An EPC Contractors Perspective. Proceedingsof International Gas Turbine Institute ASME TURBO EXPO.Norgaard, P., Holttinen, H., 2004. A multi-turbine power curve approach. In: NordicWind Power Conference.NREL, 2010. National Renewable Energy Laboratory, Westernwind and solar integration study, [Online] Available:http://www.nrel.gov/wind/systemsintegration/wwsis.html.NREL, 2011. National Renewable Energy Laboratory, The Role of LargeBalancing Areas in Integrating Solar Generation, [Online] Available:http://www.nrel.gov/docs/fy11osti/50059.pdf.NYISO, 2010. Growing Wind - Final Report of the NYISO Wind Integration Study,[Online] Available: http://www.nyiso.com.

References 134OSIsoft, 2007. Cutting the Cost of Flexible Operation in a Competitive Power Market,[Online] Available: http://www.osisoft.com/.Oswald, J., Raine, M., Ashraf-Ball, H., 2008. Will British weather provide reliableelectricity? Energy Policy 36 (8), 3212–3225.Padhy, N., 2004. Unit commitment - a bibliographical survey. IEEE Transactions onPower Systems 19 (2), 1196–1204.Probert, T., 2011. Fast starts and flexibility: Let the gas turbine battlecommence. Power Engineering International, [Online] Available:http: // www. powerengineeringint. com/ 19 (6).Risø National Laboratory, 1989. European wind resources at 50 metres a.g.l., [Online]Available: http://www.windatlas.dk/Europe/landmap.html.Salam, S., 2007. Unit commitment solution methods. In: Proc. of World Academy ofScience, Engineering and Technology. Vol. 26. pp. 320–325.SEAI, 2010. Sustainable Energy Authority Ireland, History of Wind Energy, [Online]Available: http://www.seai.ie/Renewables/Wind Energy/History of Wind Energy/.Sen, S., Kothari, D., 1998. Optimal thermal generating unit commitment: a review.International Journal of Electrical Power & Energy Systems 20 (7), 443–451.Shahidehpour, M., Marwali, M., 2000. Maintenance scheduling in restructured powersystems. Springer.Shibli, A., Starr, F., 2007. Some aspects of plant and research experience in the use ofnew high strength martensitic steel p91. International Journal of Pressure Vesselsand Piping 84 (1-2), 114–122.Siebert, N., 2008. Development of methods for regional wind power forecasting. Ph.D.thesis, Mines ParisTech, Paris, France.Siemens, 2008a. Low Load Operational Flexibility for Siemens G-class Gas Turbines.In: Proc. of Power-Gen International.Siemens, 2008b. Operational flexibility enhancements of combined-cycle power plants.In: Proc. of Power-Gen Asia.Söder, L., 2004. Simulation of wind speed forecast errors for operation planning of multiareapower systems. In: IEEE International Conference on Probabilistic MethodsApplied to Power Systems. pp. 723–728.Starr, F., 2003. Background to the design of HRSG systems and implications for CCGTplant cycling. Operation Maintenance and Materials Issues 2 (1).Strbac, G., Shakoor, A., Black, M., Pudjianto, D., Bopp, T., 2007. Impact of windgeneration on the operation and development of the UK electricity systems. ElectricPower Systems Research 77 (9), 1214–1227.

References 135Streiffert, D., Philbrick, R., Ott, A., 2005. A mixed integer programming solution formarket clearing and reliability analysis. In: Proc. of IEEE Power Engineering SocietyGeneral Meeting. pp. 2724–2731.Troy, N., Denny, E., O’Malley, M., 2010. Base-load cycling on a system with significantwind penetration. IEEE Transactions on Power Systems 25 (2), 1088–1097.Tuohy, A., Meibom, P., Denny, E., O’Malley, M., 2009. Unit commitment for systemswith significant wind penetration. IEEE Transactions on Power Systems 24 (2),592–601.Tuohy, A., O’Malley, M., 2009. Impact of pumped storage on power systems withincreasing wind penetration. In: Proceedings of 2009 IEEE PES General Meeting.Calgary, Alberta, Canada.Ummels, B., Gibescu, M., Pelgrum, E., Kling, W., Brand, A., 2007. Impacts of windpower on thermal generation unit commitment and dispatch. IEEE Transactions onEnergy Conversion 22 (1), 44–51.Van Hulle, F., Gardner, P., 2008. Wind Energy - The Facts, Part 2 Grid Integration,[Online]. Available: http://www.wind-energy-the-facts.org/.Wambeke, S., 2006. Risks to HRSGs in low-load operation. Combined Cycle JournalSecond Quarter, [Online] Available: http://www.combinedcyclejournal.com/.Watson, W., 1996. The success of the combined cycle gas turbine. In: Proc. of theIEEE Conference on Opportunities and Advances in International Electric PowerGeneration. pp. 87–92.Xcel Energy, 2010. Integrating Wind Cost of Cycling Analysisfor Harrington Station Unit 3 [Online] Available:http://www.blankslatecommunications.com/Images/Aptech-HarringtonStation.pdf.

Appendix A. Probability distribution of net load rampsFigure 8.2: Probability distribution of hourly net load ramps on the 7.55 GW peaksystem136

Appendix A. Probability distribution of net load ramps 137Figure 8.3: Probability distribution of hourly net load ramps on the 9.6 GW peaksystem

Appendix B. Cycling data for CCGT and coal unitsFigure 8.4: Start-ups and capacity factor for a typical low-merit CCGT unit on the7.55 and 9.6 GW peak demand systems, with increasing wind penetration138

Appendix B. Cycling data for CCGT and coal units 139Table 8.2: Start-up data for CCGT and coal units on the 7.55 GW peak demand systemStatistic CCGT CoalWind energy penetration 15% 29% 43% 15% 29% 43%Max. value 161 186 197 65 54 43Min. value 18 41 70 8 9 6Average value 72.4 90.6 115.4 28.4 26.4 20.6Std. Deviation 63.8 63.6 52.3 26.4 22.6 15.4Table 8.3: Capacity factor data for CCGT and coal units on the 7.55 GW peak demandsystemStatistic CCGT CoalWind energy penetration 15% 29% 43% 15% 29% 43%Max. value 0.81 0.75 0.64 0.77 0.71 0.69Min. value 0.73 0.59 0.44 0.72 0.67 0.61Average value 0.79 0.71 0.59 0.75 0.70 0.66Std. Deviation 0.034 0.066 0.086 0.024 0.175 0.029Table 8.4: Start-up data for CCGT and coal units on the 9.6 GW peak demand systemStatistic CCGT CoalWind energy penetration 11% 23% 34% 11% 23% 34%Max. value 116 170 198 56 81 67Min. value 7 23 44 8 9 5Average value 32.4 63.6 93 27.2 36.4 32.4Std. Deviation 47.1 63.4 65.3 25.4 36.3 30.9

Appendix B. Cycling data for CCGT and coal units 140Table 8.5: Capacity factor data for CCGT and coal units on the 9.6 GW peak demandsystemStatistic CCGT CoalWind energy penetration 11% 23% 34% 11% 23% 34%Max. value 0.90 0.85 0.77 0.83 0.79 0.76Min. value 0.85 0.76 0.65 0.77 0.75 0.72Average value 0.87 0.82 0.74 0.80 0.76 0.74Std. Deviation 0.02 0.03 0.05 0.02 0.02 0.02

Appendix C. Base-load cycling with/without storage/interconnectionFigure 8.5: Number of hours online for an average CCGT and coal unit with/withoutstorage and an increasing wind penetration on the 9.6 GW peak demand system141

Appendix C. Base-load cycling with/without interconnection 142Figure 8.6: Number of start-ups for an average CCGT and coal unit with/withoutstorage and an increasing wind penetration on the 9.6 GW peak demand systemFigure 8.7: Number of hours online for an average CCGT and coal unit with/withoutinterconnection and an increasing wind penetration on the 9.6 GW peak demand system

Appendix D. Fuel Cost CurvesFigure 8.8: Fuel cost curves for test units143

Appendix D. Fuel Cost Curves 144Figure 8.9: Fuel cost curves for test units

Appendix E. Publications1. Troy, N., Denny, E. and O’Malley, M. “Base-load cycling on a system with significantwind penetration”, IEEE Transactions on Power Systems, vol. 25, issue2, pp. 1088 - 1097, 2010.2. Troy, N., Flynn, D. and O’Malley, M. “Multi-mode Operation of Combined-CycleGas Turbines with Increasing Wind Penetration”, IEEE Transactions on PowerSystems, In Press3. Troy, N., Flynn, D., Milligan M. and O’Malley, M. “Unit Commitment withDynamic Cycling Costs”, IEEE Transactions on Power Systems, in review.145

1088 IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 25, NO. 2, MAY 2010Base-Load Cycling on a SystemWith Significant Wind PenetrationNiamh Troy, Graduate Student Member, IEEE, Eleanor Denny, Member, IEEE, and Mark O’Malley, Fellow, IEEEAbstract—Certain developments in the electricity sector may resultin suboptimal operation of base-load generating units in countriesworldwide. Despite the fact they were not designed to operatein a flexible manner, increasing penetration of variable powersources coupled with the deregulation of the electricity sector couldlead to these base-load units being shut down or operated at partloadlevels more often. This cycling operation would have onerouseffects on the components of these units and potentially lead toincreased outages and significant costs. This paper shows the seriousimpact increasing levels of wind power will have on the operationof base-load units. Those base-load units which are notlarge contributors of primary reserve to the system and have relativelyshorter start-up times were found to be the most impacted aswind penetration increases. A sensitivity analysis shows the presenceof storage or interconnection on a power system actually exacerbatesbase-load cycling until very high levels of wind powerare reached. Finally, it is shown that if the total cycling costs ofthe individual base-load units are taken into consideration in thescheduling model, subsequent cycling operation can be reduced.Index Terms—Costs, interconnected power systems, powersystem modeling, pumped storage power generation, thermalpower generation, wind power generation.I. INTRODUCTIONAS higher penetrations of wind power are achieved, systemoperation becomes increasingly complex, as variationsin the net load (load minus wind) curve increase [1]. Wind isa variable energy source and fluctuations in output must beoffset to maintain the supply/demand balance, thus resulting ina greater demand for operational flexibility from the thermalunits on the system [2]. These units must also carry additionalreserves to maintain system reliability should an unexpecteddrop in wind occur, as the power output from wind farms isalso relatively difficult to predict [3]. However, even whenstate-of-the-art methods of forecasting are employed, the nextday hourly predicted wind output can vary by 10%–15% ofManuscript received May 25, 2009; revised September 24, 2009. First publishedJanuary 08, 2010; current version published April 21, 2010. This workwas conducted in the Electricity Research Centre, University College Dublin,Ireland, which is supported by Airtricity, Bord Gais, Bord na Mona, Cylon Controls,the Commission for Energy Regulation, Eirgrid, Electricity Supply Board(ESB) International, ESB Networks, ESB Power Generation, Siemens, SWSGroup, and Viridian. This work was supported by a Charles Parsons Energy ResearchAward from the Department of Communications, Energy and Natural Resourcesadministered by Science Foundation Ireland. Paper no. TPWRS-00377-2009.N. Troy and M. O’Malley are with the School of Electrical, Electronic, andMechanical Engineering, University College Dublin, Dublin, Ireland (e-mail:niamh.troy@ucd.ie; mark.omalley@ucd.ie).E. Denny is with the Department of Economics, Trinity College Dublin,Dublin, Ireland (e-mail: dennye@tcd.ie).Digital Object Identifier 10.1109/TPWRS.2009.2037326the total wind capacity as reported in [4], which can resultin thermal units being over- and under-committed [2]. Furthermore,in certain systems wind is allowed to self-dispatch,so forecast output is not included in the day-ahead schedule.This can lead to increased transmission constraints whichwill further intensify plant cycling and has been shown todisplace energy from combined cycle gas turbines (CCGTs) inparticular [5]. The culmination of adding more variability andunpredictability to a power system is that thermal units willundergo increased start-ups, ramping and periods of operationat low load levels collectively termed “cycling”[6]–[9].In addition to wind, the competitive markets in which theseunits operate are also a significant driver of plant cycling;increased levels of competition brought about by widespreadderegulation results in all types of generators being forcedinto more market-orientated, flexible operation to increaseprofits [10]. The severity of plant cycling, will be dependenton the generation mix and the physical characteristics of thepower system. It is widely reported that the availability ofinterconnection and storage can assist the integration of windon a power system [11], [12]. Interconnection can allow imbalancesfrom predicted wind power output to be compensatedvia imports/exports whereas some form of energy storage canenable excess wind to be moderated in time to correlate withdemand. This should relieve cycling duty on thermal units asthe onus on them to balance fluctuations is relieved.Although all conventional units will be impacted to some degreeby wind integration, it is cycling of base-load units that isparticularly concerning for system operators and plant ownersalike. As these units are designed with minimal operationalflexibility, cycling these units will result in accelerated deteriorationof the units’ components through various degenerationmechanisms such as fatigue, erosion, corrosion, etc, leading tomore frequent forced outages and loss of income. The start/stopoperation and varying load levels result in thermal transientsbeing set up in thick-walled components placing them understress and causing them to crack. The interruptions to operationcaused by cycling disrupts the plant chemistry and resultsin higher amounts of oxygen and other ionic species beingpresent, leading to corrosion and fouling issues. A multitudeof other cycling related issues have been documented in theliterature [13]–[19]. Excessive cycling of base-load units couldpotentially leave them permanently out of operation prior totheir expected lifetimes.Hence cycling of base-load units will impose additional costson the unit, the most apparent being increased operations andmaintenance (O&M) and capital costs resulting from deteriorationof the components. However, fuel costs will also increasewith cycling operation as the unit will be starting up more frequently,and also because the overall efficiency of the unit will0885-8950/$26.00 © 2010 IEEE

TROY et al.: BASE-LOAD CYCLING ON A SYSTEM WITH SIGNIFICANT WIND PENETRATION 1089deteriorate. Environmental penalties will arise as a result of increasedfuel usage, while income losses arise as the unit will undergolonger and more frequent outages [17], [19], [20]. Quantifyingthese costs is particularly difficult given the vast arrayof components affected. Also, cycling related damage may notbe immediately apparent. Studies have suggested it can take upto seven years for an increase in the failure rate to become apparentafter switching from base-load to cycling [21]. The uncertaintysurrounding cycling costs can lead to these costs beingunder-valued by generators, which in turn can lead to increasedcycling.This paper examines the effect that increasing penetration ofwind power will have on the operation of base-load units. Therole that interconnection and storage play in alleviating or aggravatingthe cycling of base-load units is investigated acrossdifferent wind penetration scenarios. Finally, the effect of increasingstart-up costs (to represent increasing depreciation) onthe operation of base-load units is examined. Section II detailsthe methodology used in the study. Section III reports theresults and discusses the impact of modeling assumptions onthese results. Section IV provides some discussion surroundinghow wind and plant cycling is treated in electricity markets.Section V concludes the paper.II. METHODOLOGYA. Modeling ToolSimulations were carried out using a scheduling model calledthe Wilmar Planning Tool, which is described extensively in[22] and [23]. The Wilmar Planning Tool was originally developedto model the Nordic electricity system and was lateradapted to the Irish system as part of the All Island Grid Study[23]. It is currently employed in the European Wind IntegrationStudy [24]. The Wilmar Planning Tool was the tool of choicefor this study as it combined the benefits of mixed integer optimizationwith stochastic modeling. The main functionality ofthe Wilmar Planning Tool is embedded in the Scenario Tree Tooland the Scheduling Model.The Scenario Tree Tool generates scenario trees containingthree inputs to the scheduling model: wind, load and demandfor replacement reserve. Realistic possible wind forecast errorsare generated using an auto regressive moving average (ARMA)approach which considers the historical statistical behavior ofwind at individual sites. Historical wind speed series taken fromthe various sites are then added to the wind speed forecast errorscenarios to generate wind speed forecast scenarios. These arethen transformed to wind power forecast scenarios. Load forecastscenarios are generated in a similar manner. A multi dimensionalARMA model, as in [25], is used to simulate the windcorrelation between sites. A scenario reduction technique similarto that in [26] is employed to reduce the large number ofpossible scenarios generated.In the modeling tool reserve is categorized as primary or replacement.Primary reserve, which is needed in short time scales(less than five minutes), is supplied only by synchronized units.The system should have enough primary reserve to cover anoutage of the largest online unit occurring at the same time asa fast decrease in wind power production. Positive primary reserveis provided by increased production from online units orpumped storage, whilst negative primary reserve is provided bydecreased production from online units or by pumped storagewhen in pumping mode. The demand for replacement reserve,which is reserve with an activation time greater than 5 min, isdetermined by the total forecast error which is defined accordingto the hourly distribution of wind power and load forecast errorsand the possibilities of forced outages. A forced outage time seriesfor each unit is also generated by the scenario tree tool usinga semi-Markov process based on given data of forced outagerates, mean time to repair and scheduled outages is produced.Any unit that is offline and can come online in under one hourcan provide replacement reserve.The Scheduling Model minimizes the expected cost of thesystem over the optimization period covering all scenarios generatedby the scenario tree tool and subject to the generatingunits’ operational constraints, such as minimum down times (theminimum time a unit must remain offline following shut-down),synchronization times (time taken to come online), minimumoperating times (minimum time a unit must spend online oncesynchronized) and ramp rates. In order to maintain adequatesystem inertia and dynamic reactive support at times of highwind, a minimum number of large base-load units must be onlineat all times. Details of the objective function which containsfuel, carbon and start-up costs are given in Appendix A and furtherdetails are included in [22]. The Generic Algebraic ModelingSystem (GAMS) was used to solve the unit commitmentproblem using the mixed integer feature of the Cplex solver. Forall the simulations in this study the model was run with a dualitygap of 0.01%.Rolling planning is used to re-optimize the system as newwind and load information becomes available. Starting at noonthe system is scheduled over 36 hours until the end of the nextday. The model steps forward with a three hour time step withnew forecasts used in each step. In each planning period a threestage stochastic optimization model is solved having a deterministicfirst stage, a stochastic second stage with three scenarioscovering three hours and a stochastic third stage with sixscenarios covering a variable number of hours according to theplanning period in question. The state of the units at the start ofany time step must be the same as the state of the units at theend of the previous time step.B. Test SystemThe 2020 Irish system was chosen as a test case for this studybecause its unique features make it suitable for investigatingbase-load cycling. It is a small island system, with limited interconnectionto Great Britain, a large portion of base-load plantand significant wind penetration. Thus, potential issues with cyclingof base-load units may arise on this system at a lower windpenetration.Various portfolios were developed in the Wilmar PlanningTool for the All Island Grid Study [27] to investigate the effectsof different penetrations of renewables on the Irish system forthe year 2020. Portfolios 1, 2, and 5 from [27] were used inthis study and are outlined in Table I as the “moderate wind”,“high wind”, and “very high wind” cases. A “no wind” case hasalso been added. As seen in Table I, the test system is a thermalsystem, with a small portion of inflexible hydro capacity and thebase-load is composed of coal and combined cycle gas turbine(CCGT) generation. The three wind cases examined have 2000MW, 4000 MW, and 6000 MW wind installed on the system,

1090 IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 25, NO. 2, MAY 2010TABLE IINSTALLED CAPACITY (MW) BY FUEL TYPETABLE IIICHARACTERISTICS OF A TYPICAL CCGTAND COAL UNIT ON THE TEST SYSTEMTABLE IIFUEL PRICES (C/GJ) BY FUEL TYPEwhich supply 11%, 23%, and 34% of the total energy demandand represent 19%, 32%, and 42% of the total installed capacityon the system, respectively.The 2020 winter peak forecast is 9.6 GW and the summernight valley is 3.5 GW. Losses on the transmission system areincluded in the load. The test system includes four 73 MWpumped storage units with a round-trip efficiency of 75% anda maximum pumping capacity of 70 MW each and two 83 MWCHP units with “must-run” status as they provide heat for industrialpurposes. The 2020 fuel prices used are shown in Table IIand a carbon price of 30/ton was assumed. The gas pricesshown in Table II are the averages over the year and the otherfuel prices remain constant throughout. As this study is primarilyconcerned with the operation of base-load units, the characteristicsof those units are shown in Table III.A simplified model of the British power system is includedin which units are aggregated by fuel type. Wind and load is assumedto be perfectly forecast on the British system. The modelincludes 1000 MW of HVDC interconnection between Irelandand Great Britain and it is scheduled on an intra-day basis, i.e.,it is rescheduled in every rolling planning period. Flows on theinterconnector to Britain are optimized such that the total costsof both systems are minimized. A maximum of 873 MW can beimported as 100 MW is used as primary reserve at all times andthere are 3% losses on the remainder.C. Scenarios ExaminedDifferent wind cases, as described in the previous section,were used in this study to allow various penetrations of windpower to be examined. The model was run stochastically, for oneyear, for the “no wind” case and each of the three wind cases toexamine the effect that increasing wind power penetration willhave on the operation of base-load units, as these are the unitswith the most limited operational flexibility and as such, willsuffer the greatest deterioration from increased cycling.To conduct a sensitivity analysis investigating the role thatstorage and interconnection play in altering the impact of increasingwind penetration on base-load operation, the modelwas run stochastically, for one year, for the “no wind” case andeach of the three wind cases, first, without any pumped storageon the system and second, without any interconnection on thesystem. In order to fairly compare systems without storage/interconnectionto the systems with storage/interconnection, thesystems must maintain the same reliability. Thus it was necessaryto replace the pumped storage units and interconnectorwith conventional plant. The 292 MW of pumped storage wasreplaced with three 97.5-MW open cycle gas turbine (OCGT)units and the 1000 MW of interconnection was replaced withnine 100-MW OCGT units (as 100 MW is always used as primaryreserve, the maximum import capacity is 900 MW). Thecharacteristics of these units were set such that they could deliverthe same capacity over the same time period as the interconnection/storageunits they replaced. Thus, in terms of flexibilitythe systems with storage/interconnection were no moreor less flexible than the systems without storage/interconnection.The OCGT units which replaced the storage units werecapable of delivering the same amount to primary reserve (132MW in total). The OCGT units that replaced the interconnectiondid not contribute to primary reserve but instead 100 MW wassubtracted from the demand for primary reserve in each hour.This is the assumption used when the interconnector is in place.The cost of running these units is generally greater than thecost of imports or production from a storage unit thus productionfrom storage/interconnection is not shifted directly to theseunits. This is advantageous in this type of study, as the operationof other units on the system without storage/interconnection canbe observed whilst the system adequacy is not undermined byreduced capacity, thus facilitating sensitivity analysis. For example,had a CCGT unit been used to replace the interconnector,it would likely provide the energy that had been previously deliveredby the interconnector but this would not allow examinationof how the existing units on the system would be affected

TROY et al.: BASE-LOAD CYCLING ON A SYSTEM WITH SIGNIFICANT WIND PENETRATION 1091TABLE IVFLUCTUATIONS IN WIND POWER OUTPUT WITH INCREASING WINDTABLE VNUMBER OF THERMAL UNITS ONLINE WITH INCREASING WIND PENETRATION(AVERAGED AT EACH HOUR SHOWN OVER A TWO-WEEK PERIOD IN APRIL)Fig. 1. Annual number of start-ups and capacity factor for an average CCGTand coal unit with increasing wind penetration.in the absence of interconnection. The results from the systemswithout storage and interconnection were compared to the basecase (i.e., with storage and interconnection).The final part of the study examined the effect that increasingthe start-up costs of the base-load units will have on their operation.It was assumed the cost of starting these units would increase,as they experienced more wear and tear, from increasedcycling. Given the uncertainty surrounding what this increasein costs might be [17], [19], the operation of the base-load unitswas examined over a range of start-up costs. The start-up costof each of the base-load units on the system was increased bya multiple of its original value and the model was run for oneyear. The process was repeated with the start-up costs incrementedby a greater multiple of the original amount each time.This was carried out for the “moderate” (19% installed windcapacity) and “very high” (42% installed wind capacity) windcases.To examine the results, the base-load units were categorizedas coal or CCGT. As the total capacity of the coal and CCGTunits varied across the portfolios, the results for the individualunits in each group were normalized by their capacity to obtainthe result per MW for each unit. The average result per MWwas then obtained and this was multiplied by the capacity of atypical coal or CCGT unit (chosen to be 260 MW and 400 MW,respectively) to give the result for a typical coal or CCGT unitas shown as follows:where is the result for the th unit, is the capacity of the thunit and is the number of unitsIII. RESULTSA. Effect of Increasing Wind Penetration on the Operation ofBase-Load UnitsAs the wind penetration on a power system is increased, largefluctuations in the wind power output will become more frequent,as seen in Table IV. In addition, generation from thermalunits is increasingly displaced, thus the number of units onlinewill decrease. This is shown in Table V.(1)Therefore the onus on thermal units to compensate fluctuationsin the wind power output becomes more demanding withincreasing wind penetration. Fig. 1 shows the annual numberof start-ups and capacity factor for an average sized CCGT andcoal unit of 400 MW and 260 MW, respectively, as wind penetrationincreases. The capacity factor is the ratio of actual generationto maximum possible generation in a given time period.As the wind penetration grows and the variability and unpredictabilityinvolved in system operation is increased, the operationof a base-load CCGT unit is severely impacted. Movingfrom 0% to 42% installed wind capacity the annual start-ups fora typical CCGT unit rise from 22 to 98, an increase of 340%.This increase in CCGT start-ups corresponds to a plummetingcapacity factor as seen in Fig. 1. Thus increasing levels of windeffectively displaces CCGT units into mid-merit operation.Similar to a CCGT unit, start-ups for a coal unit increase withwind penetration up to 32% installed wind capacity, albeit notas drastically as a CCGT unit. However, at penetrations greaterthan 32% installed wind capacity, this correlation diverges andthe start-ups for a coal unit begin to decrease, as seen in Fig. 1.As wind penetration grows, demand for primary reserve willgrow. Due to high part-load efficiencies, as indicated by the minimumload heat rates seen in Table III, coal units are the mainthermal providers of primary reserve on this system. In additionto this they have low minimum outputs so at times of high windmore coal units can remain online to meet the minimum unitsonline constraint thus minimizing wind curtailment. Coal unitsare also highly inflexible; once taken offline it is a minimumof ten hours (minimum down time plus synchronization time asseen in Table III) before the unit can be online and generatingagain. The combination of these characteristics, increases theneed for these units to be kept online to provide primary reserveto the system as high levels of wind are reached. Thus, despitethe fact that the cost of starting a CCGT unit on this system isgreater than the cost of starting a coal unit as seen in Table III,the CCGT unit has the greatest increase in start-stop cyclingwith increasing wind as it does not supply a large amount of reserveto the system, has a large minimum output and can comeonline in a shorter time compared to a coal unit.As CCGT units are taken offline more frequently with increasingwind penetration, the requirement on coal units to providereserve to the system is driven even higher. Thus, althoughthe capacity factor of a coal unit decreases as wind increases,

1092 IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 25, NO. 2, MAY 2010Fig. 2. Utilization factor and annual number of hours where severe ramping isperformed for an average CCGT and coal unit with increasing wind penetration.Fig. 3. Number of hours online for an average CCGT and coal unit with/without storage and an increasing wind penetration.the rate of decrease is much less than for a CCGT as seen inFig. 1. Therefore, as wind penetration exceeds approximately32% installed capacity a crossover point occurs and the inflexiblecoal units now become the most base-loaded units on thesystem whilst the relatively more flexible CCGT are forced intotwo-shifting, as seen by the capacity factors in Fig. 1. Thus, ifcapacity factor is indicative of the revenue earned by these units,the units with the most limited operational flexibility are themost rewarded at high levels of wind. This would suggest thatsome form of incentive may be needed to secure investment inflexible plants (for example OCGTs), which are commonly reportedas beneficial to system operation with large amounts ofwind [28], [29].Fig. 2 shows the utilization factor for an average base-loadcoal and CCGT unit and the number of hours they performsevere ramping as wind penetration increases. The utilizationfactor is the ratio of actual generation to maximum possiblegeneration during hours of operation in a given period. Severeramping is defined in this paper as a change in output greaterthan half the difference between a unit’s maximum and minimumoutput over one hour. Hours when the unit was staringup or shutting down were not included. Although coal units willavoid heavy start-stop cycling as wind levels grow by being themain thermal providers of primary reserve and highly inflexible,they do experience increased part-load operation. This isindicated by a drop in utilization factor from 0.94 to 0.88 aswind levels increase from 0% to 42% installed wind capacity,as seen in Fig. 2. The utilization factor for a CCGT unit alsodecreases with increasing levels of wind as seen in Fig. 2, however,it remains high in comparison with a coal unit, indicatingthe small contribution of reserve it provides to the system andcorrespondingly the infrequent periods of part-load operation.As seen in Fig. 2, both types of unit experience a dramatic increasein hours where severe ramping is required, as wind penetrationexceeds 32% installed capacity. As wind penetrationmoves from 32% to 42% installed wind capacity a coal unitexperiences the greatest increase in severe ramping operationgoing from 4 to 78 h, compared to an increase from 4 to 32h for a CCGT unit, as these units are now offline more often.The sharp increase in ramping corresponds to the substantial increasein wind fluctuations seen in Table IV between 32% and42% installed wind capacity, which must be compensated by asmaller number of online units. Such an increase in part-loadoperation and ramping can lead to fatigue damage, boiler corrosion,cracking of headers and component depreciation througha variety of damage mechanisms. This is of major concern toplant managers.The results reported are for “average” CCGT and coal units.In order to show how these results correspond to the actual resultsfor the real units modeled, the maximum value, minimumvalue, average value and standard deviation of the number ofstart-ups and capacity factor for the modeled CCGT and coalunits are given in Appendix B.B. Sensitivity AnalysisSection III-A showed the serious impact increasing levels ofwind will have on the operation of base-load units. The extent ofthis impact will be determined by the generation portfolio andthe characteristics of the system. This section provides a sensitivityanalysis of the effect of the portfolio on the results, byexamining the operation of the base-load units with increasinglevels of wind power when storage and interconnection are removedfrom the system.1) No Storage Case: Fig. 3 shows the number of hours onlinefor an average CCGT and coal unit on systems with andwithout pumped storage and an increasing wind penetration. Onthe system without pumped storage the base-load units spendmore hours online compared to the system with storage, untila very high wind penetration (greater than 32% installed capacityfor a CCGT and greater than 42% installed capacity fora coal unit) is reached. The presence of pumped storage on asystem will displace the primary reserve contribution requiredfrom conventional units and thus reduce the need for them to beonline. Correspondingly, an average base-load unit spends morehours online on the system without pumped storage as there ismore requirement on the unit to be online providing primaryreserve to the system. As coal units, in this case, are the mainthermal provider of primary reserve to the system they are themost affected by the addition of a storage unit, as seen for atypical coal unit in Fig. 3. The difference in hours online for atypical CCGT unit on the system with storage compared to thesystem without storage is small as they are not large contributorsto primary reserve.However, at very high wind penetrations a crossover point occurswhen large fluctuations in wind power output occur morefrequently, as seen in Table IV, and now the system with pumped

TROY et al.: BASE-LOAD CYCLING ON A SYSTEM WITH SIGNIFICANT WIND PENETRATION 1093Fig. 4. Number of start-ups for an average CCGT and coal unit with/withoutstorage and an increasing wind penetration.Fig. 5. Number of hours online for an average CCGT and coal unit with/without interconnection and an increasing wind penetration.storage is more equipped to balance these fluctuations. As thedemand for reserve is sufficiently large at very high wind penetrations,such that reserve from many thermal units is neededin addition to the reserve from the storage units, storage will nolonger be a factor in base-load units going offline. Thus, at veryhigh levels of wind, base-load units now spend more hours onlineon the system with storage compared to the system withoutstorage.Fig. 4 shows the number of start-ups for an average base-loadCCGT and coal unit on a system with and without pumpedstorage as wind penetration increases. Almost no difference inthe number of start-ups for a typical CCGT unit is seen on thesystems with and without storage until installed wind reachesgreater than 32%. However, the number of start-ups for a typicalcoal unit is seen to be much greater on the system withstorage compared to the system without storage, again indicatingthat storage will most adversely affect the units that providethe largest portion of primary reserve to the system. Againa crossover point is reached at some very high wind penetrationafter which start-ups rise rapidly on the system without storagedue to large and frequent fluctuations in wind power output. Thisoccurs at 32% installed wind for a CCGT and greater than 42%installed wind capacity for a coal unit. Thus, until very highwind penetrations are reached the existence of a pumped storageunit is shown to actually exacerbate cycling of base-load units.2) No Interconnection Case: Fig. 5 compares the numberof hours spent online by a typical CCGT and coal unit on systemswith and without interconnection, as wind is increased.The base-load units are seen to spend significantly more hoursonline on the system without interconnection compared to thesystem with interconnection until a very high wind penetrationis reached.Due to a large portion of base-load nuclear plant and cheapergas prices compared with Ireland, the market price for electricitytends to be cheaper in Great Britain. As a consequence Irelandtends to be a net importer of electricity from Great Britain andas such will import electricity before turning on domestic units.Thus interconnection to Great Britain displaces conventionalgeneration on the Irish system, forcing units down the meritorder and exacerbating plant cycling. Without the option to importelectricity, as in the “no interconnection case”, all demandmust be met by domestic units requiring more units to be onlinegenerating more often. Thus a typical CCGT and coal unit areFig. 6. Number of start-ups for an average CCGT and coal unit with/withoutinterconnection and an increasing wind penetration.seen in Figs. 5 and 6 to spend more hours online and have lessstart-ups on the system without interconnection.However, as seen in Fig. 5 at some wind penetration between32% and 42% installed wind capacity for a CCGT unit andgreater than 42% installed capacity for a coal unit, a crossoverpoint will occur when the units spend more hours online on thesystem with interconnection. As very high wind penetrationsare reached, the electricity price in Ireland undercuts Britishprices more often making exports economically viable. Thus atvery high penetrations of wind, the system with interconnectioncan deal with large fluctuations in the wind power outputvia imports/exports more favorably and avoid plant shut-downs.Thus interconnection is shown not to benefit the operation ofbase-load units on a system that is a net importer until windpenetration increases to such point that exports are economicallyviable.C. Effect of Increasing Start-Up CostsHaving shown in Sections III-A and B the severe impact increasingwind penetration will have on the operation of the baseloadunits, this section now examines how the increasing costsimposed on these units by cycling operation, will subsequentlyaffect their operation. A component of a unit’s start-up costshould be the cost of wear and tear inflicted on the unit duringthe start-up process [16]. However, given the uncertainty in determiningsuch a cost, this aspect is often neglected, leading tothe units being scheduled to start more frequently, yielding more

1094 IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 25, NO. 2, MAY 2010Fig. 7. Number of base-load start-ups for increasing start-up costs.Fig. 8.Number of hours of severe ramping duty for increasing start-up costs.cycling related damage. This section examines how the operationof the base-load units changes as the start-up costs are incrementallyincreased to represent the increasing depreciationof the unit.1) Start-Ups: The number of start-ups for an average CCGTand coal unit is shown in Fig. 7, as start-up costs are increased,with 19% and 42% installed wind capacity, respectively. Increasingthe start-up costs of a CCGT unit results in a substantialreduction in start-stop cycling, particularly at the higher windpenetration. This indicates a feedback effect, whereby increasedcycling will lead to increased costs, but when these costs areincluded in the cost function, cycling will subsequently be reduced.With 42% installed wind capacity, increasing the start-upcosts by a factor of 6 sees the start-ups for a CCGT drop from98 to 27, a decrease of 72%. Doubling the start-up costs of acoal unit in the low wind case reduced start-ups by 19, a 68%reduction. No further reduction in coal start-ups was possibleas these units were then at their minimum number of annualstart-ups (governed by scheduled and forced outages).A greater reduction in cycling is achieved by increasingstart-up costs on the system with 42% installed wind capacitycompared to the system with 19% installed wind capacity, asthis system can export more due to lower electricity prices.Increasing the start-up costs of the base-load units in Irelandby a factor of 6, results in a 29% increase in exports on thesystem with 42% installed wind capacity as it becomes moreeconomical to allow the base-load units in Ireland to stay onlineand avoid shut-downs by increasing exports to Britain.2) Ramping and Part-Load Operation: Fig. 8 shows thenumber of hours that severe ramping is required by an averageCCGT and coal unit, as start-up costs are increased with 19%and 42% installed wind capacity. Fig. 9 shows the utilizationfactor for an average CCGT and coal unit, with 19% and 42%installed wind capacity as their start-up costs are increased. Thetrade-off for the reduction in start-stop cycling of base-loadunits, achieved by increasing the start-up costs, is an increasein ramping activity as seen in Fig. 8 and part-load operationas seen in Fig. 9, which will also leads to plant deteriorationalthough it is reported to be less costly compared with start-ups[30].By increasing the start-up costs of the base-load units,start-ups are reduced and these units are kept online more, butat the expense of more flexible units which are taken offline.As a result the number of hours when the base-load units areFig. 9. Utilization factor for increasing start-up costs.the only thermal units online increases with increasing start-upcosts. During such hours there will be a considerable rampingrequirement on these units to balance fluctuations in the windpower output. As there will be even less thermal units onlinein the 42% installed wind capacity case compared to the 19%installed capacity case the greatest increase in ramping isobserved for the 42% installed wind capacity case as start-upcosts are increased, as seen in Fig. 8. Some inconsistenciesin the trend can occur because “severe ramping” is defineddiscretely, as seen for a CCGT with 42% installed wind.As the base-load units are being kept online more often, astheir start-up costs are increased, they will experience increasedpart-load operation as indicated by the reduction in utilizationfactor in Fig. 9. As start-up costs are increased sufficiently itbecomes more economical to run these units at part-load, thanto take them offline and forgo expensive start-up costs at alater time. The greater increase in part-load operation occurson the system with 42% installed wind capacity compared tothe system with 19% installed wind capacity, correspondingto the large reduction in start-ups seen at 42% installed windcapacity. The difference in start-ups and ramping for a CCGTand coal unit between 19% installed wind and 42% installedwind is also seen in Figs. 1 and 2 for the original start-up costsand for brevity is not discussed again here.D. Effect of Modeling AssumptionsThe model used was limited to hourly time resolution. Thelack of intra-hourly data may have lead to the severity of the

TROY et al.: BASE-LOAD CYCLING ON A SYSTEM WITH SIGNIFICANT WIND PENETRATION 1095cycling being seriously underestimated, for example the severeramping events. The frequency of severe ramping events foundin the study may be underestimated as severe ramps may haveoccurred over shorter time frames than one hour. Also, sucha sizeable ramp occurring over a period shorter than one hourwould have a much more damaging effect on the unit.For all simulations, rolling planning with a three hour timestep was used. Had the system been re-optimized more regularly,the wind and load forecasts would have been updated moreoften. However, [22] shows this would have minimal impact onthe operation of the base-load units examined here so a threehour time step was deemed adequate.IV. DISCUSSIONHow electricity markets evolve to manage plant cycling is beyondthe scope of this paper, however, this section offers somediscussion as to how cycling costs could be represented andareas for future market development with a large wind penetration.In many electricity markets generators submit complexbids for energy in addition to the technical characteristics ofthe plant. If the current trend for wind development continues,plant cycling, as shown in this paper, will inevitably becomingan increasing concern and generators may subsequently altertheir bids or plant characteristics in order to minimize cyclingdamage. Section III-C examines how by taking the cost of cyclinginto consideration in a unit’s start-up cost, subsequent cyclingcan be reduced. Generators in SEM, the Irish electricitymarket, are directed to include cycling costs in their start-upcosts so this approach was taken in this paper.Cycling costs could also be included in no-load or energycosts, or even defined as a new market product such as rampingcosts [31]. However, increasing the energy cost will also increasethe marginal cost of the unit, which risks changing theposition of the unit in the merit order and inducing further cycling.Alternatively cycling costs could be incorporated in aunit’s shut-down costs. The Wilmar Planning Tool used in thisstudy does not model shut-down costs at present. Future workcould investigate the effect of incorporating shut-down costs inthe scheduling algorithm on a generators dispatch.As cycling costs are difficult to quantify, generators may usethe opportunity to exercise market power. For example a generatormay increase the start-up costs excessively in order to avoidshut-down, although this strategy may result in them being leftoffline following a trip or scheduled shut-down because of theirexcessive start-up cost. Thus some may instead favor setting amaximum number of start-ups a unit can carry out over a periodof time, however, this approach would unfairly reward inflexibleunits and provide no incentive to improve operational flexibility.In some electricity markets generators submit simple bids.This can result in increased start-ups for generators as no explicitconsideration of the cost of starting the unit is taken. Incorporatingwind in such a market would induce further cycling,indicating that a move to complex bidding could be beneficial.Longer scheduling horizons that take future wind forecasts intoconsideration may also reduce plant start-ups, however the forecasterror increases with the time horizon. Thus enabling a latergate closure in a market with a significant wind penetration,which would allow the most up-to-date wind forecasts to be employed,could be more effective at reducing unnecessary plantstart-ups [32].V. CONCLUSIONSIncreasing wind penetration on a power system will lead tochanges in the operation of the thermal units on that system, butmost worryingly to the base-load units. The base-load units areimpacted differently by increasing levels of wind, depending ontheir characteristics. CCGT units see rapid increases in startstopcycling and plummeting capacity factor and are essentiallydisplaced into mid-merit operation. On the test system examinedcoal units are the main thermal providers of primary reserveto the system and as a result see increased part-load operationand ramping. This increase in cycling operation will lead to increasedoutages and plant depreciation.Certain power system assets are widely reported to assistthe integration of wind power. This paper examined if storageand interconnection reduced cycling of base-load units bycomparing a system with storage and interconnection to asystem without storage and without interconnection, across arange of wind penetrations. It was found that until very highpenetrations of wind are reached storage will actually displacethe need for base-load units to be online providing reserve tothe system. This results in increased cycling of base-load unitscompared to the system without storage. Similarly, for a systemthat is a net importer, interconnection will actually displacegeneration from domestic units, also resulting in increasedcycling of base-load units compared to a system without interconnection.At very large penetrations of wind a crossoverpoint exists, where larger and more frequent fluctuations in thewind power output, can be dealt with more effectively on asystem with interconnection and storage and thus the systemwith storage and interconnection becomes the most favorableto the operation of base-load units.Having shown how the operation of the base-load units isdramatically affected by increasing levels of wind power andassuming this would lead to added costs in various guises, theeffect that increasing start-up costs for base-load units had ontheir subsequent operation was examined. This showed that asthe cost of starting a base-loaded CCGT unit increased, startstopcycling of the unit was subsequently reduced. However, areduction in start-ups is seen to be correlated with an increasein part-load operation and ramping.APPENDIX AWILMAR OBJECTIVE FUNCTIONThe objective function shown in (A1) consists of operatingfuel cost, start up fuel cost (if a unit starts in that hour), emissionscosts and penalties incurred for not meeting load or reservetargets. If a unit is online at the end of the day, its start-upcosts are subtracted from the objective function to ensure thatthere are still units online at the end of the optimization period.The decision variable is given in the first three lines, showingwhether a unit is online or offline. Further detail on the formulationof the unit commitment problem is given in [22].

1096 IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 25, NO. 2, MAY 2010Indices:FFuel.i,IUnit group.r,RRegion.s,SScenario.START Units with start-up fuel consumption.t,TTime.USEFUEL Unit using fuel.Parameters:TABLE VIVARIATION IN CCGT START-UPS WITH INCREASING WINDTABLE VIIVARIATION IN COAL START-UPS WITH INCREASING WINDEMISSIONENDkLLOADPRICEREPSPINTAXVariables:Rate of emission.Endtime of optimization period.Probability of scenario.Infeasibility penalty.Penalty for loss of load.Fuel price.Penalty for not meeting replacement reserve.Penalty for not meeting primary reserve.Emission tax.TABLE VIIIVARIATION IN CCGT CAPACITY FACTOR WITH INCREASING WINDCONS Fuel consumed.OBJObjective function.URelaxation variable.VDecision variable—on or off.ONLINE Integer on/off for unit.QDAY Day ahead demand not met.QINTRA Intra day demand not met.QREP Replacement reserve not met.QSPIN Primary reserve not met.+, - Up, down regulation.TABLE IXVARIATION IN COAL CAPACITY FACTOR WITH INCREASING WINDAPPENDIX BSUMMARY OF NON-NORMALIZED BASE CASE RESULTSTables VI–IX indicate the variation in start-ups and capacityfactor of the CCGT and coal units in the base case (i.e.,Tables VI–IX relate to Fig. 1), for each of the wind penetrations.The maximum value, minimum value, average and standarddeviation are shown. It can be seen that the CCGT units havea greater spread in start-ups compared to the coal units andthe standard deviation of start-ups is least at the highest windcase for both types of units. For capacity factor the spread inresults across the units increased as the wind increased, withthe CCGT units again having a greater variation compared tothe coal units, however, there are more CCGT units than coalunits in each of the wind cases.REFERENCES(A1)[1] H. Holttinen, “Impact of hourly wind power variations on the systemoperation in the Nordic countries,” Wind Energy, vol. 8, no. 2, pp.197–218, Apr./Jun. 2005.[2] B. C. Ummels, M. Gibescu, E. Pelgrum, W. Kling, and A. Brand, “Impactsof wind power on thermal generation unit commitment and dispatch,”IEEE Trans. Energy Convers., vol. 22, no. 1, pp. 44–51, Mar.2007.

TROY et al.: BASE-LOAD CYCLING ON A SYSTEM WITH SIGNIFICANT WIND PENETRATION 1097[3] G. Dany, “Power reserve in interconnected systems with high windpower production,” in Proc. IEEE Power Tech Conf., vol. 4, 6 pp, 2001.[4] Ahlstrom, L. Jones, R. Zavadil, and W. Grant, “The future of windforecasting and utility operations,” IEEE Power and Energy Mag., vol.3, no. 6, pp. 57–64, Nov.–Dec. 2005.[5] The Effect of Integrating Wind Power on Transmission System Planning,Reliability and Operations, Report prepared for New York StateEnergy Research and Development Agency, 2005. [Online]. Available:http://www.nyserda.org/publications/wind_integration_report.pdf.[6] P. Meibom, C. Weber, R. Barth, and H. Brand, “Operational costs inducedby fluctuating wind power production in Germany and Scandinavia,”Proc. IET Renew. Power Gen., vol. 3, no. 1, pp. 75–83, Jan.2009.[7] M. Braun, “Environmental external costs from power generation by renewableenergies,” Master’s thesis, Stuttgart Univ., Stuttgart, Germany,2004.[8] H. Holttinen, V. T. T. Finland, J. Pedersen, and E. Denmark, “The effectof large scale wind power on thermal system operation,” in Proc.4th Int. Workshop Large-Scale Integration of Wind Power and TransmissionNetworks for Offshore Wind Farms, Billund, Denmark, Oct.2003.[9] L. Goransson and F. Johnsson, “Dispatch modeling of a regional powergenerating system—Integrating wind power,” Renew. Energy, vol. 34,no. 4, pp. 1040–1049, Apr. 2009.[10] L. Balling and D. Hoffman, Fast Cycling Towards Bigger Profits,Modern Power Systems, 2007. [Online]. Available: http://www.modernpowersystems.com.[11] Variability of Wind Power and Other Renewables—Management Optionsand Strategies, International Energy Agency. [Online]. Available:http://www.iea.org/textbase/papers/2005/variability.pdf.[12] Large Scale Integration of Wind Energy in the European Power Supply:Analysis, Issues and Recommendations, European Wind Energy Association.[Online]. Available: http://www.ewea.org/index.php?id=178.[13] R. Viswanathan and J. Stringer, “Failure mechanisms of high temperaturecomponents in power plants,” Trans. ASME, vol. 122, pp. 246–255,Jul. 2000.[14] V. Viswanathan and D. Gray, Damage to Power Plants Due to Cycling,EPRI, Palo Alto, CA, 2001, Tech. Rep. 1001507.[15] J. Gostling, “Two shifting of power plant: Damage to power plants dueto cycling—A brief overview,” OMMI, vol. 1, no. 1, Apr. 2002. [Online].Available: http://www.ommi.co.uk/.[16] K. D. Le, R. R. Jackups, J. Feinstein, H. Thompson, H. M. Wolf, E. C.Stein, A. D. Gorski, and J. S. Griffith, “Operational aspects of generationcycling,” IEEE Trans. Power Syst., vol. 5, no. 4, pp. 1194–1203,Nov. 1990.[17] F. J. Berte and D. S. Moelling, “Assessing the true cost of cycling is achallenging assignment,” Combined Cycle J., pp. 23–25, 2003.[18] C. Johnston, “An approach to power station boiler and turbine lifemanagement,” in Proc. World Conf. NDT, Montreal, QC, Canada, Sep.2004.[19] S. A. Lefton, P. M. Besuner, and G. P. Grimsrud, “Managing utilitypower plant assets to economically optimize power plant cycling costs,life, and reliability,” in Proc. 4th IEEE Conf. Control Applications, Albany,NY, Sep. 1995.[20] E. Denny and M. O’Malley, “The impact of carbon prices on generationcycling costs,” Energy Pol., vol. 37, no. 4, pp. 1204–1212, Apr. 2009.[21] S. A. Lefton, P. M. Besuner, G. P. Grimsrud, A. Bissel, and G. L.Norman, Optimizing Power Plant Cycling Operations While ReducingGenerating Plant Damage and Costs at the Irish Electricity SupplyBoard. Sunnyvale, CA: Aptech Eng. Service, 1998.[22] A. Tuohy, P. Meibom, E. Denny, and M. O’Malley, “Unit commitmentfor systems with significant wind penetration,” IEEE Trans. PowerSyst., vol. 24, no. 2, pp. 592–601, May 2009.[23] Wind Variability Management Studies, All Island Renewable GridStudy—Workstream 2B, 2008. [Online]. Available: http://www.dcmnr.gov.ie.[24] European Wind Integration Study. [Online]. Available: http://www.wind-integration.eu/.[25] L. Soder, “Simulation of wind speed forecast errors for operations planningof multi-area power systems,” in Proc. 2004 IEEE Int. Conf. ProbabilisticMethods Applied to Power Systems, Ames, IA, Sep. 2004, pp.723–728.[26] J. Dupacova, N. Growe-Kuska, and W. Romisch, “Scenario reductionin stochastic programming: An approach using probability metrics,”Math. Program., vol. 95, no. 3, pp. 493–511, 2003.[27] High Level Assessment of Suitable Generation Portfolios for the All-Island System in 2020, All Island Renewable Grid Study—Workstream2A, 2008. [Online]. Available: http://www.dcmnr.gov.ie.[28] B. Kirby and M. Milligan, “Facilitating wind development: The importanceof electric industry structure,” Elect. J., vol. 21, no. 3, pp. 40–54,Apr. 2008.[29] G. Strbac, A. Shakoor, M. Black, D. Pudjianto, and T. Bopp, “Impact ofwind generation on the operation and development of the UK electricitysystems,” Elect. Power Syst. Res., vol. 77, no. 9, pp. 1214–1227, Jul.2007.[30] Editorial, “Profitable operation requires knowing how much it costs tocycle your unit,” Combined Cycle J., pp. 49–52, 2004.[31] M. Flynn, M. Walsh, and M. O’Malley, “Efficient use of generator resourcesin emerging electricity markets,” IEEE Trans. Power Syst., vol.15, no. 1, pp. 241–249, Feb. 2000.[32] C. Hiroux and M. Saguan, “Large-scale wind power in European electricitymarkets: Time for revisiting support schemes and market designs,”Energy Policy, to be published.Niamh Troy (GS’09) received the B.Sc. degree in appliedphysics from the University of Limerick, Limerick,Ireland. She is currently pursuing the Ph.D. degreeat the Electricity Research Centre in the UniversityCollege Dublin, Dublin, Ireland.Eleanor Denny (M’07) received the B.A. degreein economics and mathematics, the M.B.S. degreein quantitative finance, and the Ph.D. degree inwind generation integration from University CollegeDublin, Dublin, Ireland, in 2000, 2001, and 2007,respectively.She is currently a Lecturer in the Departmentof Economics at Trinity College Dublin and hasresearch interests in renewable generation andintegration, distributed energy resources, and systemoperation.Mark O’Malley (F’07) received the B.E. and Ph.D.degrees from University College Dublin, Dublin, Ireland,in 1983 and 1987, respectively.He is a Professor of electrical engineering atUniversity College Dublin and is director of theElectricity Research Centre with research interestsin power systems, control theory, and biomedicalengineering.

This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.IEEE TRANSACTIONS ON POWER SYSTEMS 1Multi-Mode Operation of Combined-Cycle GasTurbines With Increasing Wind PenetrationNiamh Troy, Member, IEEE, Damian Flynn, Senior Member, IEEE, and Mark OMalley, Fellow, IEEEAbstract—As power systems evolve to incorporate greater penetrationsof variable renewables, the demand for flexibility withinthe system is increased. Combined-cycle gas turbines are traditionallyconsidered as relatively inflexible units, but those whichincorporate a steam bypass stack are capable of open-cycle operation.Facilitating these units to also operate in open-cycle mode canbenefit the power system via improved system reliability, while reducingthe production needed from dedicated peaking units. Theutilization of the multi-mode functionality is shown to be dependenton the flexibility inherent in the system and the manner inwhich the system is operated.Index Terms—Power system modeling, thermal power generation,wind power generation.I. INTRODUCTIONCOMBINED-CYCLE gas turbines (CCGTs) are a type ofpower generating unit that achieve high efficiencies (upto 60%) by capturing the waste heat from a gas turbine in a heatrecovery steam generator (HRSG) and using it to produce superheatedsteam to drive a steam turbine [1]. The high efficienciesachieved, combined with their ease of installation, short-buildtimes, and relatively low gas prices, have made the CCGT a populartechnology choice [2], [3]. In the Republic of Ireland, forexample, 43% of the installed thermal capacity is CCGT technology,while in the markets of Texas (ERCOT) and New England(NEPOOL), CCGTs represent 37% of the total installedcapacity.The operational flexibility of a CCGT unit is limited by thesteam cycle, which contains many thick-walled components,necessary to withstand extreme temperatures and pressures [4],[5]. To avoid differential thermal expansion across these componentsand the subsequent risk of cracking, these componentsmust be brought up to temperature slowly, resulting in slowerstart-up times and ramp rates for the unit overall [6]. However,by incorporating a bypass stack upstream of the HRSG at thedesign stage, a CCGT unit has the option to bypass the steamcycle and run in open-cycle mode, whereby exhaust heat fromManuscript received March 02, 2011; revised June 24, 2011; accepted July22, 2011. This work was conducted in the Electricity Research Centre, UniversityCollege Dublin, Ireland, which is supported by the Commission for EnergyRegulation, Bord Gais Energy, Bord na Mona Energy, Cylon Controls, EirGrid,the Electric Power Research Institute (EPRI), ESB Energy International, ESBEnergy Solutions, ESB Networks, Gaelectric, Siemens, SSE Renewables, andViridian Power & Energy. This work was supported by Science Foundation Irelandunder Grant Number 06/CP/E005. Paper no. TPWRS-00128-2011.The authors are with the School of Electrical, Electronic, and MechanicalEngineering, University College Dublin, Dublin, Ireland (e-mail:niamh.troy@ucd.ie; damian.flynn@ucd.ie; mark.omalley@ucd.ie).Digital Object Identifier 10.1109/TPWRS.2011.2163649the gas turbine is ejected directly into the atmosphere via thebypass stack [6]. This reduces the power output and efficiencyof the plant but offers greater operational flexibility. Runningin open-cycle mode, the gas turbine has a short start-up time of15 to 30 min and is capable of changing load quickly. However,bypass stacks are not always incorporated because they can potentiallylead to leakage losses, thus reducing plant efficiency,while also introducing additional capital costs [1].As international energy policy drives ever greater penetrationsof renewable energy, wind power is set to represent alarger portion of the generation mix [7]. This is driving a greaterdemand for flexibility within power systems in order to dealwith high penetrations of variable and difficult to predict energysources [8], [9]. Storage, interconnection, and responsivedemand are commonly cited as flexible options for dealing withvariability issues [10]–[12]; however, these options have considerablecosts associated with them. Facilitating open-cycleoperation of CCGT units that have the technical capability torun in open-cycle mode (i.e., those with a bypass stack) canalso deliver much needed flexibility to a system with a highwind penetration. This resource is often technically available,but inaccessible due to market arrangements.In order to derive the greatest benefits from a CCGT unit thatcan run in open-cycle mode, it is necessary for the schedulingalgorithm to explicitly consider both modes of operation for theunit, i.e., open-cycle and combined-cycle [13]. These will havegreatly different technical and cost characteristics and so needto be declared individually. Currently most markets do not facilitateCCGT units to submit multiple bids representing differentmodes of operation; thus, presently open-cycle operationof a CCGT unit is typically limited to periods when the steamsection is undergoing maintenance. However, some U.S. systemshave begun addressing this issue to varying degrees, withERCOT and CAISO seeking to implement configuration-basedmodeling of CCGTs [14], [15].The option to run in open-cycle mode could also providebenefits for the generators. Renewable integration studies haveshown that CCGT units will experience significant decreasesin running hours and thus will receive less revenue from themarket as they are displaced by greater levels of wind generationwhich has an almost zero marginal cost [16]–[20]. Dueto their high minimum loads, CCGTs are shut down frequentlywith high wind penetrations as they cannot reduce output sufficientlyto accommodate the wind power output [16]. By facilitatingCCGT units to operate in open-cycle mode, these unitsmay have a new opportunity to capture revenue from increasedoperation during periods when they might otherwise be offline.For example, if a CCGT unit has been forced offline by high0885-8950/$26.00 © 2011 IEEE

This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.2 IEEE TRANSACTIONS ON POWER SYSTEMSwind generation on the system, it may have the opportunity torun as a peaking unit.This paper builds on preliminary work in [21] and includesimproved modeling of CCGTs from that in [21] to examineif a power system with a high wind penetration can benefitfrom the additional flexibility introduced when these units arefacilitated to operate in open-cycle mode, when technicallyfeasible and economically suitable. The all-island Irish 2020system [22] is considered here as it is expected to contain botha large share of wind power and CCGT units. In addition, asit is a small, island system that is weakly interconnected, thechallenges of maintaining the supply/demand balance with ahigh wind penetration are exacerbated, and so the solutionsfound can hold insights for other systems pursuing large-scalewind power. Section II describes the modeling tool used in thisstudy and also the changes that were made to model multi-modeoperation of CCGTs. Section III outlines the test system used.Section IV describes the results of the study and Section Vconcludes the paper.II. MODELING TOOLThe Wilmar Planning Tool is a stochastic, mixed integerunit commitment and economic dispatch model, originallydeveloped to model the Nordic electricity system and lateradapted to the Irish system as part of the All Island Grid Study[22]–[25]. The main functionality of the Wilmar Planning Toolis embedded in the Scenario Tree Tool and Scheduling Model.The Scenario Tree Tool utilizes historical wind power or windspeed data, load data, and wind and load forecasts for differenttime horizons to identify an auto regressive moving average(ARMA) series which can then simulate wind and load forecasterrors for various time horizons [26]. These simulated windand load forecasts errors are paired in a random way before ascenario reduction technique, following the approach of [27], isapplied. The wind and load forecast errors are combined withscaled up wind and load time series to produce wind power productionand load forecast scenarios. For each scenario, the demandfor replacement reserve (activation time min) is calculatedbased on a comparison of the hourly power balance consideringperfect forecasts and no forced outages with the powerbalance considering scenarios of wind and load forecast errorsas well as forced outages. A percentile of the deviation betweenthe compared power balances must be covered by replacementreserves; in this case, the 90th percentile is chosen based on currentpractice [23]. A forced outage time series for each unit isalso generated by the Scenario Tree Tool using a semi-Markovprocess based on historical plant data of forced outage rates,mean time to repair, and scheduled outages.The model can also be run in deterministic and perfect foresightmodes whereby only one wind generation and load scenariois planned for. In deterministic mode, this scenario is theexpected value of wind and load. The expected value of windis found by summing, for all (post-reduction) scenarios, theproduct of the wind power forecasts and their probability ofoccurring. The expected value of load and replacement reserveis found similarly [24]. Consequently, the scenario planned forwill differ from the realized scenario. This mode is typical ofthe scheduling process currently practiced by most system operators,i.e., only one scenario is planned for and it will containsome level of forecast error. Perfect foresight mode contains noforecast error for wind generation or load but forced outages stilloccur, as with all other modes.The Scheduling Model minimizes the expected costs for allscenarios, subject to system constraints for reserve and the minimumnumber of units online (6 units in the Republic of Irelandand 2 units in Northern Ireland). These costs include fuel,carbon, and start-up fuel costs (always assumed to be hot starts).In addition to replacement reserve, one category of spinningreserve, namely tertiary operating reserve (TR1), is modeled,which has a response time of 90 s to 5 min and is only suppliedby online units. Enough spinning reserve must be available tocover an outage of the largest online unit occurring concurrentlywith a fast decrease in wind power production over the TR1 timeframe, as described in [28].Generator constraints such as minimum down times, synchronizationtimes, minimum operating times, and ramp rates mustalso be obeyed. Rolling planning is employed to re-optimize thesystem as new wind generation and load information becomeavailable. Starting at noon each day, the system is scheduledover 36 h until the end of the next day. The model steps forwardwith a 3-h time step and reschedules the units based on informationfrom new forecasts. The model produces a year-long dispatchat an hourly time resolution for each individual generatingunit. Further detail on the model and formulation of the unitcommitment problem can be found in [23]. The Generic AlgebraicModeling System (GAMS) is used to solve the unit commitmentproblem using the mixed integer feature of the Cplexsolver (version 12). For all simulations in this study, the modelwas run with a duality gap of 0.5%. A year-long simulation takesh when run in deterministic mode or h in stochasticmode, on an Intel core quad 3-GHz processor with 4 GB ofRAM.A. Modeling Multi-Mode Operation of CCGTsIn order to examine the potential for multi-mode operationof CCGT units a set, “ ”, of all CCGT units capable of prolongedopen-cycle operation, i.e., those with bypass stacks, wasdefined. The set “ ” corresponds to these CCGT unitswhen run in open-cycle mode. CCGT units comprised of twoor more gas turbines will have multiple “ ” units, as indicatedby index “a”. The relation “multi-mode” is defined topair each member of “ ” with the corresponding member(s)of “ ”. To ensure the mutually exclusive operation ofthese “ ” units and the corresponding “ ” units, theconstraint shown in (1) was added to the model, whereis the state binary variable which describes the online status ofthe unit. This allows the model to dispatch, when economicallyoptimal, either the “ ” (combined-cycle mode) or any/all ofthe corresponding “ ” units (open cycle mode), for allscenarios “s” and time steps “t”, but not both simultaneously asthey are in reality the same unit:(1)

This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.TROY et al.: MULTI-MODE OPERATION OF COMBINED-CYCLE GAS TURBINES WITH INCREASING WIND PENETRATION 3Equation (2), taken from [29], sets the state binary variablesor equal to 1 for all units “i”, when a unit is startedup or shut down, respectively:(2)When modeling multi-mode operation of CCGT units, twonew circumstances arise when calculating the start-up fuelconsumption, , which must be explicitly represented.Firstly, when a “ ” unit transitions from conventional combined-cycleoperation into open-cycle operation no start-up fuelis consumed by the “ ” unit as represented by inequality(3), where is the start-up energy used by each unit(measured in MWh). When the “ ” unit starts from zeroproduction ( and ), the first termon the right-hand side of inequality (3) determines the fuel usedby the unit while the second term equals zero. Alternatively,when the unit switches from combined-cycle to open-cycleoperation ( and ), the second termcauses the right-hand side of (3) to equal zero. Settingas a positive variable and using an inequality condition ensuresthat when a “ ” unit is shutting down and the corresponding“ ” unit is not starting up, will be 0:The second circumstance relates to the unit transitioningfrom open-cycle to combined-cycle operation. In this case, thestart-up fuel consumed is less than the start-up fuel used inbringing the CCGT online from zero production, as some ofthis start-up fuel has already been used to bring the unit onlinein open-cycle mode and the gas section of the plant is in a hotstate. As an approximation, the start-up fuel used to bring theunit into combined-cycle operation from open-cycle operationis the difference between the start-up fuel for the “ ” and afraction, , of the start-up fuel for the “ ”, as seen in(4). Based on the operating experience of generators, waschosen to be 0.5 here. When the “ ” unit is started fromzero production ( and ), the firstterm on the right-hand side of (4) provides the start-up fuelconsumed, while the second term equals zero. When the unitswitches from open-cycle to combined-cycle operation, thesecond term is included, thus approximating the start-up fuelconsumed in this situation:In the Wilmar model, any unit can contribute to the targetfor replacement (non-spinning) reserve, provided that an offlineunit can come online in time to provide reserve for the hourin question and the reserve available from an online unit is notneeded to meet spinning reserve targets. In Wilmar, the contributionfrom online and offline units to the replacement reservetarget, (MW), are calculated individually. In this case the“ ” units cannot provide offline replacement reserve as theyhave long start-up times, but the corresponding “ ” unitscan, given their fast start-up times. The constraints shown in (5)(3)(4)Fig. 1. CCGT start-up from open-cycle mode.and (6), where is a unit’s minimum stable operatinglevel (MW) and is a unit’s maximum capacity (MW),ensure that if either the “ ” unit or the “ ” unit is online,then the “ ” unit cannot contribute to the portionof replacement reserve that is provided from offline units. Thisis necessary to avoid the situation where a “ ” unit is onlineand the model allows the corresponding “ ” unit to contributeto offline replacement reserve:Improved modeling of plant start-ups was also implementedfollowing the formulation given in [29]. This allows for thoseunits with start-up times greater than 1 h to be block-loaded overthe course of their start-up time. In earlier versions of the Wilmarmodel, units remained at zero production for the duration ofstart-up process. The addition of this feature significantly increasedthe computation time, so only the start-up process of theCCGT units was modeled in detail. Other units with a start-uptime greater than 1 h, namely the coal-fired units, typically havefewer starts over the year and lower minimum operating levelsrelative to the CCGTs and so modeling their start-up process indetail would have little impact on the results.When the bypass stack is utilized to switch from combined-cycleto open-cycle operation, the transition is automaticand occurs without shutting down the gas turbine or reducingits power output. However, the transition from open-cycle tocombined-cycle operation is dependent on the temperature stateof the boiler. Therefore, if the CCGT unit has been operatingfor a period of time in open-cycle mode and is then scheduled toswitch to combined-cycle mode, its output must adjust in orderto achieve the correct HRSG inlet temperature, as depicted inFig. 1. This was implemented by setting the allowable poweroutput ( from [29]) for each interval of the CCGT’sstart-up process, which begins at hour 0 in Fig. 1, such that theappropriate soak time is achieved.Scheduled outages for each unit, determined from historicalexperience [22], are inputted in time-series format to the Wilmarmodel. In this case, CCGT units with the capability to operatein open-cycle mode are considered to be available to run inopen-cycle mode for a portion of their scheduled outage. Given(5)(6)

This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.4 IEEE TRANSACTIONS ON POWER SYSTEMSTABLE IGENERATION MIX OF TEST SYSTEMTABLE IIICHARACTERISTICS OF CCGT UNITS (CAPABLE OF MULTI-MODEOPERATION) IN COMBINED- AND OPEN-CYCLE MODESTABLE IIFUEL PRICES BY FUEL TYPEthat gas turbine equipment is more accessible and compact incomparison with the steam turbine equipment, it was assumedthat one third of the maintenance period was sufficient for thegas turbine.III. TEST SYSTEMThe test system used is the Irish 2020 system, based onportfolio 5 from the All Island Grid Study [22], [30]. Four103.5-MW OCGT units were removed from the original gridstudy portfolio as recent generation adequacy reports wouldindicate they are unlikely to be built by 2020 [31]. Table I showsthe number of units, installed capacity, and average operatingcost (fuel) by generation type. (The multi-mode capable CCGTunits in open-cycle mode are shown on the last row.) Threedifferent levels of installed wind power were examined: 2000,4000, and 6000 MW, which supply 15%, 29%, and 44% ofthe total energy demand, respectively. Fuel prices are as givenin Table II. Base-load gas generators (i.e., CCGTs and CHP)are assumed to have long-term fuel contracts and thereforepay a cheaper fuel price compared to mid-merit gas generators(i.e., OCGTs, ADGTs, and legacy CCGTs). Differences in thefuel price for coal and gas oil in the Republic of Ireland andNorthern Ireland reflect varying delivery costs. The originaldemand profile from [22] with a 9.6-GW peak and 54-TWhtotal demand was scaled down to a profile with a 7.55-GW peakand 42-TWh total demand to reflect a reduction in predicteddemand, seen in recent long-term forecasts [31].The test system assumes that there is 1000 MW of HVDCinterconnection in place between Ireland and Great Britain andit is scheduled on an intra-day basis, i.e., it can be rescheduledin every 3-h rolling planning period. A simplified model of theBritish power system is included, with aggregated units, no integervariables for generators and where wind generation andload are assumed to be perfectly forecast. The total demand inBritain is assumed to be 370 TWh with a peak of 63 GW andthe installed wind capacity is assumed to be 14 GW. A carbonprice ofwas assumed.Five (of the ten) CCGT units on the Irish system include bypassstacks and therefore can run in open-cycle mode. Eachof these units is currently installed and operational. The characteristicsof these units in combined-cycle mode are given inTable III. Limited data was available for these units in opencyclemode so each was given characteristics similar to a typicalopen-cycle gas turbine (OCGT) unit, as shown in Table III.As CCGT 2 and CCGT 5 are comprised of two gas turbinesconnected to one steam turbine ( configuration), these unitswere modeled as having two identical open-cycle units availablefor dispatch when the CCGT is operated in open-cycle mode.CCGTs 2 and 3, located in Northern Ireland and CCGTs 1, 4,and 5, located in the Republic of Ireland, contribute to the minimumunits online constraint in their respective regions.IV. RESULTSA number of model runs were conducted to investigate thepotential for multi-mode operation of CCGT units. The Wilmarmodel was run in deterministic mode as this is more representativeof current scheduling practice. A year-long dispatch wasproduced for each of the three wind power penetrations outlinedin Section III, when 1) multi-mode operation of CCGT units isnot allowed and 2) when multi-mode operation of CCGT unitsis allowed.

This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.TROY et al.: MULTI-MODE OPERATION OF COMBINED-CYCLE GAS TURBINES WITH INCREASING WIND PENETRATION 5TABLE VOCGT PRODUCTION (GWh) WITH INCREASING WIND PENETRATIONTABLE VIDIFFERENCE IN OPEN-CYCLE PRODUCTION (GWh) FROM MULTI-MODEUNITS WITH NO REPLACEMENT RESERVE TARGET ENFORCEDFig. 2. Average production from a CCGT in open-cycle mode (line) andaverage number of instances generators utilized open-cycle operation (greycolumn), shown for various levels of installed wind capacity.TABLE VIIAVERAGE HOURLY SURPLUS SPINNING RESERVE (MW) AVAILABLEAND REPLACEMENT RESERVE TARGET (MW)TABLE IVAVERAGE UTILIZATION FACTORS WITH INCREASING WIND PENETRATIONA. Usage of the Multi-Mode FunctionThe average number of times a CCGT unit with multi-modecapability was run in open-cycle mode and the average productionfrom a CCGT in open-cycle mode over the year, at eachof the wind penetrations examined, is shown in Fig. 2. Despiteincreasing wind penetration being correlated with an increaseddemand for flexibility, be it fast starting or ramping, Fig. 2 showsthe multi-mode function is used less frequently as wind penetrationon the system increases.As more wind power, with an almost zero marginal cost, isadded to a system, the production from thermal plant is increasinglydisplaced and as such there is an increased likelihood ofgenerators operating at part-load. To illustrate, Table IV givesthe annual utilization factor (ratio of actual generation to maximumpossible generation during hours of operation) averagedfor the coal, CCGT, and peat units on the system with 2000-,4000-, and 6000-MW wind power. Therefore, as wind penetrationincreases, online part-loaded units are more often availableto ramp up their output to meet unexpected shortfalls in production,avoiding the need to switch on fast-starting units, such asthe CCGTs in open-cycle mode.The trend seen in Fig. 2 is consistent with the production frompeaking plants as wind penetration increases. Table V shows thedrop in production from the most utilized OCGT unit, with increasingwind penetration when multi-mode operation is and isnot allowed. Reduced production from peaking plants due to increasedwind penetration has also been observed in other windintegration studies such as [17]; however, it is also likely thatsystems with base-load units that have slower ramp rates thanthose examined in this study will rely on fast-starting units (suchas CCGTs in open-cycle mode) more often as wind penetrationincreases. (All units on the test system are assumed to be capableof ramping from minimum to maximum output in onehour or less.) The average production from the CCGT units inopen-cycle mode, as seen in Fig. 2, is comparable with averageproduction levels from dedicated OCGT peaking plants on thesystem when multi-mode operation of CCGTs is not enabled.As wind penetration increases, so too will the demand for replacementreserve, due to the increased forecast error. The replacementreserve target can be met by fast-starting offline unitsor from excess spinning reserve if available. If sufficient excessspinning reserve is not available to meet the replacement reservetarget, the model must ensure a number of fast-starting units areoffline and available for operation to maintain a secure system.Consequently, as a result of maintaining the replacement reservetarget, production from fast-start units (such as the multi-modeunits in open-cycle mode) is reduced. Additional simulationswere conducted for the various wind penetrations with no replacementreserve target, to investigate the extent that maintainingreplacement reserve suppressed the multi-mode unitsfrom running in open-cycle mode. For many systems, such asthe Irish system, this is more representative of current practice,where no replacement reserve target formally exists. Table VIshows the difference in the average open-cycle production frommulti-mode units that results when no replacement reserve targetsare enforced.As seen, in the absence of a target for replacement reserve,open-cycle production from the multi-mode units is utilizedsubstantially more for the 2000-MW and 4000-MW windpower scenarios. However, with 6000-MW wind power, due tomore frequent part-loading of units, there is more frequentlyan excess of spinning reserve on the system, as well as offlinefast-starting units (as per Table V) which can contribute to thereplacement reserve target. Thus, with 6000-MW wind power,the replacement reserve target has little effect on the open-cycleoperation of multi-mode units. Table VII shows the averagesurplus spinning reserve available and the average replacementreserve target per hour for each of the wind cases examined.Fig. 3 shows the capacity factor for each CCGT in combined-cyclemode and its production over the year in open-cyclemode for the 2000-MW wind power scenario. An inverse relationshipis evident between the open-cycle production from aCCGT and the capacity factor of the CCGT, which indicatesthat usage of the multi-mode function is related to the amount

This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.6 IEEE TRANSACTIONS ON POWER SYSTEMSFig. 3. Combined-cycle capacity factor (dashed line) and open-cycle production(solid line) for each CCGT with multi-mode capability for the 2000-MWwind power system.Fig. 4. Average production from OCGT peaking units in each wind power scenario,with multi-mode operation of CCGTs not allowed (light grey) and allowed(dark grey).TABLE VIIIPERCENTAGE CHANGE IN TOTAL PRODUCTION WHEN MULTI-MODE ISENABLED, SHOWN FOR EACH WIND PENETRATIONTABLE IXMAGNITUDE AND FREQUENCY OF REPLACEMENT RESERVE SHORTFALL,SHOWN FOR VARIOUS LEVELS OF INSTALLED WINDof time the CCGT is offline. The more often a CCGT is not inoperation but available for dispatch, the more opportunities ithas to run in open-cycle mode and this relationship would beexpected regardless of the plant portfolio.The percentage change in total production (combined-cycleplus open-cycle) that results when multi-mode operation ofCCGTs is enabled is shown in Table VIII, for each of thewind penetrations examined. Multi-mode operation increasedproduction for CCGT5, the lowest merit CCGT which wasseen to utilize the function most frequently, across all thewind penetrations examined. Total production from CCGT3and CCGT4, which are mid-merit CCGTs, is reduced in allcases but one. There is a risk (particularly for CCGTs that arefrequently the marginal unit on the system such as CCGT3and CCGT4), when offering open-cycle operation, of beingdispatched from combined-cycle to open-cycle operation attimes of low net demand (demand minus wind generation) toalleviate minimum load issues and then losing out to anothergenerator that can come online faster/cheaper, when the netdemand increases again. However, it is also likely that in amarket environment, generators would strategize when theywould offer this multi-mode capability to avoid losing out onproduction. CCGT1, the highest merit CCGT, benefits fromincreased production when multi-mode operation is enabledon the system with 2000-MW and 4000-MW installed windpower. This is due to increased exports and reduced productionfrom the other CCGTs, as opposed to increased production inopen-cycle mode.B. Benefits Arising From Multi-Mode OperationThe efficiencies of the OCGT peaking units on the system arecomparable with the CCGT units in open-cycle mode. However,the CCGT units running in open-cycle operation are assumedto have a lower gas price, to reflect the advantage of long-termcontracts. Their open-cycle capacity (as seen in Table III) isalso larger than the capacity of the OCGTs (103.5 MW each)and they benefit from avoided start-up costs when transitioningfrom combined-cycle mode. Thus, when multi-mode operationof CCGTs was enabled, production from OCGT peaking planttended to be substituted by production from the CCGTs in opencyclemode. Fig. 4, which shows the average production fromOCGTs for each wind penetration level when multi-mode operationof CCGTs is allowed and not allowed, illustrates thispoint. Assuming open-cycle production from CCGTs is moreeconomic than production from OCGTs, as is the case here, it ispossible that by enabling multi-mode operation of CCGTs sufficientflexibility could be extracted from a systems portfolio ofplant to avoid building additional peaking units, or equally thatOCGT units would no longer be able to cover their costs and sowould be forced to retire from service. Both situations may thenlead to increased production from CCGTs in open-cycle mode.Table IX shows the total shortfall in replacement reserve overthe year and the number of hours in which this occurred, for eachof the wind penetrations examined, when multi-mode operationof CCGTs is and is not allowed. The additional fast-startinggeneration available to the system when multi-mode operationof CCGT units is allowed significantly reduces the shortfall inreplacement reserve. This contributes to a more secure systemby preventing capacity shortfalls when wind forecasts prove tobe overly optimistic and also indicates that, depending on themarket structure, the generators may benefit from an additionalrevenue stream, via ancillary services payments for the replacementreserve provided.In addition to enhanced system security, the additional flexibilityavailable to the system when multi-mode operation ofCCGT units is allowed will also yield production costs savings.Table X shows the total system operating cost savings achievedby enabling multi-mode operation of CCGTs. The total system

This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.TROY et al.: MULTI-MODE OPERATION OF COMBINED-CYCLE GAS TURBINES WITH INCREASING WIND PENETRATION 7TABLE XTOTAL SYSTEM COST SAVING (MC) RESULTINGFROM MULTI-MODE OPERATION OF CCGTScost is made up of fuel, carbon, and start-up costs for the Irishand British system combined, as they are co-optimized. In thiscase, these savings were achieved at no additional cost as eachof the CCGTs is currently capable of multi-mode operation.A modest reduction in plant start-ups for multi-mode units(in combined-cycle mode) was also observed ( averagedover the three wind power scenarios), relative to the case whenmulti-mode operation is not allowed, which would indicate benefitsfor the steam equipment via avoided wear-and-tear.C. Sensitivity StudiesUsage of the multi-mode function is dependent on manyfactors, particularly the amount of flexibility already presentin the system. A sensitivity study was conducted to examinethe usage of the multi-mode function when the system was lessflexible to meeting demand. This involved running the modelwith 2000-MW wind power (as this level of wind generationgreatest usage of CCGTs in open-cycle mode) and powerexchange across the interconnector fixed day-ahead as opposedto intra-day. Examining the usage of the multi-mode functionwhen the interconnector is scheduled day-ahead versusintra-day illustrates how a less flexible system will utilize thisflexible resource more frequently. Fig. 5 shows the averageproduction from a CCGT in open-cycle mode and the averagenumber of instances CCGTs utilized open-cycle operation,with the interconnector scheduled day-ahead and intra-day onthe 2000-MW wind power system. The average productionfrom CCGTs in open-cycle mode on the system with day-aheadscheduling of the interconnector is seen to be more than threetimes greater than the system with intra-day scheduling ofthe interconnector. By fixing the power exchange betweenthe Irish and British systems day-ahead, when there is greateruncertainty in the expected wind generation and demand, thesystem is forced to dispatch generators such as the multi-modeCCGT units, as opposed to reschedule imports/exports, tocompensate for wind and load forecast errors. Likewise, systemswith seasonal hydro restrictions may see greater usageof multi-mode CCGT operation during these periods when theoperating flexibility of the system is reduced.In addition, the type of wind and load forecasts employed bya system will also determine the usage of the multi-mode function.Additional simulations were completed running the modelin stochastic and perfect foresight mode. These represent differentmeans of including load and wind forecasts in the schedulingprocess; whereby stochastic optimization can be consideredto represent a system employing ensemble forecasts, deterministicoptimization is representative of a system utilizing asingle forecast, and the perfect forecast scenario is a hypotheticalcase where no forecast error exists. The robust solutionsobtained by stochastic optimization showed less deployment ofthe multi-mode function compared with the deterministic results.The stochastic solution, optimized for several wind andFig. 5. Average production from a CCGT in open-cycle mode (line) andaverage number of instances generators utilized open-cycle operation (greycolumn), with interconnector scheduled day-ahead and intra-day on 2000-MWwind system.Fig. 6. Average production from CCGT in open-cycle mode (GWh), shown fordifferent methods of optimization with 2000-MW wind power.load scenarios, typically has more units online to cover all scenariosand therefore is more prepared to deal with unforseenshortfalls in wind generation or increases in demand withoutthe need for starting peaking plant. The capacity factors of theCCGT units are also higher for the stochastic case comparedto the deterministic case, indicating that there was also less opportunityfor these units to run in open-cycle mode when thesystem is optimized stochastically. Running the Wilmar modelwith perfect foresight of the system demand and wind profilealso reveals even less open-cycle operation from CCGTs as inthis case, with no forecast errors on the system (except forcedoutages of generators), fast starting units are in less demand relativeto the deterministically optimized solution. Fig. 6 comparesthe average open-cycle operation from the multi-mode CCGTs,on the system with 2000-MW wind power, when optimized withperfect foresight, stochastically and deterministically. The averageopen-cycle production from a CCGT unit is seen to be11% less on the stochastically optimized system and 35% lesson the system with perfect forecast compared to the deterministiccase.A sensitivity analysis was also conducted using a higher levelof demand on the system. In this case, the original demand profilefrom [22] with a 9.6-GW peak, discussed in Section III,was run for each wind scenario. The average production froma CCGT in open-cycle mode over the year is shown in Fig. 7to be six to eight times greater on the 9.6-GW peak demandsystem, where peaking capacity is in greater demand, comparedto the 7.55-GW peak demand system, at each of the wind powerpenetrations examined. In addition to the increased demand resultingin increased open-cycle production from the multi-mode

This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.8 IEEE TRANSACTIONS ON POWER SYSTEMSTABLE XITOTAL SYSTEM COST, REPLACEMENT RESERVE SHORTFALL AND TOP-UP PAYMENT, SHOWN FOR VARIOUS MULTI-MODE CONFIGURATIONSFig. 7. Average production from a CCGT in open-cycle mode on the 7.55-GWpeak demand system (light grey) and the 9.6-GW peak demand system (darkgrey), shown for various levels of installed wind power.CCGTs (as well as combined-cycle production), the other maindifference between the scenarios is the predominant direction ofpower transfer on the interconnector. With 2000-MW installedwind capacity, the Irish system is a net importer of power fromBritain, at both levels of demand examined. However, as morewind power is installed on the 7.55-GW peak demand system,the marginal electricity price is reduced sufficiently with respectto the British system such that Ireland becomes a net exporterof power. Although increasing wind power penetration on the9.6-GW peak demand system also reduces the marginal price,it is still a net importer with 6000-MW installed wind power.Thus, on occasions when forecast wind is overestimated and thesystem is in need of fast-starting plant, the 7.55-GW peak demandsystem, being a net exporter, can more frequently chooseto curtail exports or start up a unit to compensate. In contrast, the9.6-GW peak demand system, being a net importer, more oftenonly has the option to turn on fast-starting plant. Hence, this impliesthat a system which tends to be a net exporter is inherentlymore flexible, and has more options for dealing with variablewind power than a system that is a net importer of power. In thisscenario with higher demand, each of the multi-mode CCGTunits experienced increased total production (combined-cycleplus open-cycle) when multi-mode operation was allowed, suggestingthat offering multi-mode capability may prove moreprofitable on a system with a smaller capacity margin.Given the low deployment of the multi-mode functionalityand the high capacity factor in combined-cycle mode for CCGT1 and 2, as seen in Fig. 3, it would appear that there is insufficientincentive for all CCGTs capable of multi-mode operationto offer this flexible capability. Thus, given that CCGTs 3, 4, and5 have low capacity factors in combined-cycle mode, additionalsimulations were conducted to investigate the benefits yieldedif these units alone, and if CCGT 5 alone, offered multi-modecapability. Table XI shows the total system cost (for Ireland andBritain) and the magnitude of the replacement reserve shortfallover the year for these configurations (in addition to other configurationsexamined in the paper). Examining the shortfall inthe replacement reserve target for the different configurationsreveals that the majority of the reduction in replacementreserve shortfall due to multi-mode capability is attributableto CCGT 5, while CCGTs 1 and 2 are seen to have noimpact on the replacement reserve shortfall. Thus, CCGTs capableof open-cycle operation, which have very low output incombined-cycle mode, have value in providing replacement reserve.As seen in Table VIII, the multi-mode CCGTs may experiencea reduction in total production as a result of offeringmulti-mode capability to the market. This was also observedto be the case for CCGTs 3 and 4, when only three units offeredmulti-mode operation. This indicates that a system seekingto increase its flexibility via multi-mode operation of CCGTs,possibly to facilitate integration of variable renewables, mayneed to reward these units either through ancillary service paymentsor another market mechanism to restore their revenueto original levels (i.e., when multi-mode operation was not allowed).The subsidy or “top-up payment” required to restore therevenue of these units to their original level is estimated hereas the loss in total production multiplied by the average electricityprice. The average “top-up payment” required is shown inTable XI with the number of units requiring this payment shownin parenthesis. However, it should be noted that this representsthe worst-case figure given that the multi-mode CCGT unit offeredthis capability in all time periods, rather than when it wasprofitable for them to do so, as would likely be the case in reality.V. CONCLUSIONSThis paper examines if allowing CCGT units to operate inopen-cycle mode, when this is technically feasible and costoptimal, could deliver benefits to a system with a high windpenetration or to the generators themselves. It is shown that the

This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.TROY et al.: MULTI-MODE OPERATION OF COMBINED-CYCLE GAS TURBINES WITH INCREASING WIND PENETRATION 9extra fast-starting capacity available from multi-mode operationof CCGTs can reduce the replacement reserve shortfall,indicating an opportunity for increasing system reliability.Low-merit CCGTs will utilize the multi-mode function moreas they are frequently offline and available for dispatch, whilethe increased competition among generators, typical at higherlevels of wind generation, results in multi-mode operation ofCCGTs being utilized less frequently. Peaking production fromCCGTs in open-cycle mode can displace peaking productionfrom OCGTs, potentially reducing the need for such units tobe built. Sensitivity studies reveal that usage of the multi-modefunction is dependent on the level of flexibility inherent ina system. Optimizing the system stochastically or allowingintra-day trading on interconnectors reduces the need for flexibilityto be extracted from generators and consequently resultsin less frequent deployment of the multi-mode function.ACKNOWLEDGMENTThe authors would like to thank A. Mahon and A. Barnes ofESB for their helpful contributions.REFERENCES[1] R. Kehlhoffer, Combined-Cycle Gas & Steam Turbine Power Plants,2nd ed. Tulsa, OK: PennWell, 1999.[2] W. J. Watson, “The success of the combined cycle gas turbine,” inProc. IEEE Conf. Opportunities and Advances in International ElectricPower Generation, 1996, pp. 87–92.[3] U. C. Colpier and D. Cornland, “The economics of the combined cyclegas turbine—An experience curve analysis,” Energy Policy, vol. 30, no.4, pp. 309–316, 2002.[4] A. Shibli and F. Starr, “Some aspects of plant and research experiencein the use of new high strength martensitic steel P91,” Int. J. PressureVessels and Piping, vol. 84, pp. 114–122, 2007.[5] F. Starr, “Background to the design of HRSG systems and implicationsfor CCGT plant cycling,” Oper. Mainten. Mater. Issues, vol. 2, no. 1,Apr. 2003.[6] R. Anderson and H. van Ballegooyen, “Steam turbine bypass systems,”Combined Cycle J., 2003.[7] Winning With European Wind, European Wind Energy Association,2009. [Online]. Available: http://www.ewea.org.[8] H. Chandler, Empowering Variable Renewables, Options for FlexibleElectricity Systems. Paris, France: International Energy Agency,2008.[9] F. Van Hulle and P. Gardner, Wind Energy—The Facts, Part 2Grid Integration, 2008. [Online]. Available: http://www.wind-energy-the-facts.org/.[10] P. Brown, J. Lopes, and M. Matos, “Optimization of pumped storagecapacity in an isolated power system with large renewable penetration,”IEEE Trans. Power Syst., vol. 23, no. 2, pp. 523–531, May 2008.[11] V. Hamidi and F. Robinson, “Responsive demand in networks withhigh penetration of wind power,” in Proc. IEEE/PES Transmission andDistribution Conf. Expo., 2008.[12] L. Göransson, “Wind power in thermal power systems—Large-scaleintegration,” Licentiate thesis, Dept. of Energy and Environment,Chalmers Univ. Technology, Goteburg, Sweeden, 2008.[13] B. Lu and M. Shahidehpour, “Short-term scheduling of combined-cycleunits,” IEEE Trans. Power Syst., vol. 19, no. 3, pp.1616–1625, Aug. 2004.[14] B. Blevins, Combined-Cycle Unit Modeling in the Nodal Design.Taylor, TX: ERCOT, 2007.[15] Multi-Stage Generating (MSG) Unit Modeling, CAISO, 2010. [Online].Available: http://www.caiso.com/2078/2078908392d0.html.[16] N. Troy, E. Denny, and M. O’Malley, “Base-load cycling on a systemwith significant wind penetration,” IEEE Trans. Power Syst., vol. 25,no. 2, pp. 1088–1097, May 2010.[17] Growing Wind—Final Report of the NYISO Wind Integration Study,NYISO, 2010. [Online]. Available: http://www.nyiso.com.[18] Integration of Renewable Resources—Operational Requirements andGeneration Fleet Capability at 20% RPS, California ISO, 2010. [Online].Available: http://www.caiso.com/2804/2804d036401f0ex.html.[19] Western Wind and Solar Integration Study, National Renewable EnergyLaboratory, 2010. [Online]. Available: http://www.nrel.gov/wind/systemsintegration/wwsis.html.[20] L. Göransson and F. Johnsson, “Dispatch modeling of a regional powergeneration system—Integrating wind power,” Renew. Energy, vol. 34,no. 4, pp. 1040–1049, 2009.[21] N. Troy and M. O’Malley, “Multi-mode operation of combined cyclegas turbines with increasing wind penetration,” in Proc. IEEE Powerand Energy Soc. General Meeting, 2010.[22] Wind Variability Management Studies, All Island Renewable GridStudy—Workstream 2B, 2008. [Online]. Available: http://www.dcenr.gov.ie.[23] P. Meibom, R. Barth, B. Hasche, H. Brand, and M. O’Malley, “Stochasticoptimization model to study the operational impacts of highwind penetrations in Ireland,” IEEE Trans. Power Syst., vol. 26, no. 3,pp. 1367–1379, Aug. 2011.[24] A. Tuohy, P. Meibom, E. Denny, and M. O’Malley, “Unit commitmentfor systems with significant wind penetration,” IEEE Trans. PowerSyst., vol. 24, no. 2, pp. 592–601, May 2009.[25] P. Meibom, WILMAR—Wind Power Integration in LiberalisedElectricity Markets, 2006. [Online]. Available: http://www.wilmar.risoe.dk/Results.htm.[26] L. Söder, “Simulation of wind speed forecast errors for operation planningof multiarea power systems,” in Proc. Int. Conf. ProbabilisticMethods Applied to Power Systems, 2004.[27] J. Dupacova, N. Growe-Kuska, and W. Romisch, “Scenario reductionin stochastic programming: An approach using probability metrics,”Math. Program., vol. 95, no. 3, pp. 493–511, 2003.[28] R. Doherty and M. O’Malley, “A new approach to quantify reserve demandin systems with significant installed wind capacity,” IEEE Trans.Power Syst., vol. 20, no. 2, pp. 587–595, May 2005.[29] J. M. Arroyo and A. J. Conejo, “Modeling of start-up and shut-downpower trajectories of thermal units,” IEEE Trans. Power Syst., vol. 19,no. 3, pp. 1562–1568, Aug. 2004.[30] Redpoint Validated Forecast Model and PLEXOS Validation Report2010, Commission for Energy Regulation, 2010. [Online]. Available:http://www.allislandproject.org.[31] Generation Adequacy Report 2010–2016, EirGrid, 2009. [Online].Available: http://www.eirgrid.com.Niamh Troy (M’11) received the B.Sc. degree in applied physics from the Universityof Limerick, Limerick, Ireland. She is currently pursuing the Ph.D. degreeat the Electricity Research Centre in University College Dublin, Dublin,Ireland.Damian Flynn (SM’11) is a senior lecturer in power engineering at UniversityCollege Dublin, Dublin, Ireland. His research interests involve an investigationof the effects of embedded generation sources, especially renewables, on theoperation of power systems.Mark O’Malley (F’07) received the B.E. and Ph.D. degrees from UniversityCollege Dublin, Dublin, Ireland, in 1983 and 1987, respectively.He is a Professor of electrical engineering in University College Dublin andis director of the Electricity Research Centre with research interests in grid integrationof renewables.Prof. O’Malley is a member of the Royal Irish Academy.

1Unit Commitment with Dynamic Cycling CostsNiamh Troy, Student Member, IEEE, Damian Flynn, Senior Member, IEEE, Michael Milligan, SeniorMember, IEEE, and Mark O’Malley, Fellow, IEEEAbstract—Increased competition in the electricity sector andthe integration of variable renewable energy sources is resultingin more frequent cycling of thermal plant. Thus, the wearand-tearto generator components and the related costs are agrowing concern for plant owners and system operators alike.This paper presents a formulation that can be implementedin a MIP dispatch model to dynamically model cycling costsbased on unit operation. When implemented for a test systemthe results show that dynamically modeling cycling costs reducescycling operation and tends to change the merit order over time.This leads to the burden of cycling operation being more evenlydistributed over the plant portfolio and a reduces the total systemcosts relative to the case when cycling costs are not modeled.Index Terms—Thermal Power Generation, power system modelingIndices/SetsNOMENCLATUREt, T Time step, set of time stepsg, G Units, set of unitsi, I Interval of cycling cost function, set of intervalsof cycling cost functionj, J Level of ramp, set of all ramp levelsl, L Segment of the piecewise linearization of thevariable cost function, set of all segments ofthe piecewise, linearization of the variablecost functionConstantsa g , b g , c g Coefficients of the quadratic production costfunction for unit gcost S g Cycling cost increment for each additionalstart-up for unit gTh S g (i) i th threshold corresponding to cumulativestart-ups for unit gcost S g (i) Cycling cost increment for each additionalstart-up, while N S g (t,i) < Th S g (i+1) for unit gR gproduction change (MW) over time period tdeemed damaging for unit gR g (j) j th production change (MW) over time periodt deemed damaging for unit gN. Troy (niamh.troy@ucd.ie), D. Flynn (damian.flynn@ucd.ie) and M.O’Malley (mark.omalley@ucd.ie) are with the School of Electrical, Electronicand Communications Engineering, University College Dublin, Ireland.Michael Milligan is with the National Renewable Energy Laboratory, Golden,CO 8041 USA (email: michael.milligan@nrel.gov).This work was conducted in the Electricity Research Centre, UniversityCollege Dublin, Ireland, which is supported by the Commission for EnergyRegulation, Bord Gais Energy, Bord na Mona Energy, Cylon Controls, Eir-Grid, the Electric Power Research Institute (EPRI), ESB Energy International,ESB Energy Solutions, ESB Networks, Gaelectric, SSE Renewables, andViridian Power & Energy. This publication has emanated from researchconducted with the financial support of Science Foundation Ireland underGrant Number 06/CP/E005.cost R g Cycling cost increment for unit g for eachadditional ramp > R gTh R g (i) i th threshold corresponding to cumulativeramps for unit gcost R g (i) Cycling cost increment for unit g for eachadditional ramp, while N R g (t,i) < Th R g (i+1)I gTotal number of intervals in cycling costfunction for unit g¯j gNumber of ramp levels defined for unit g¯P gMaximum capacity for unit gP g Minimum capacity for unit gA gFixed cost for unit g ($/h)NL g Number of segments in piecewise linearizationof the variable cost function for unit gF lg Slope of segment l of the variable costfunction for unit gT lg Upper limit of block l of the piecewise linearproduction cost function of unit j (MW)UT g Minimum up time for unit gDT g Minimum down time for unit g¯TT coldgNumber of hours in the planning periodNumber of hours unit g must be offline,beyond its minimum downtime, before it isconsidered to be in a cold statecc g Cold start-up cost for unit ghc g Hot start-up cost for unit gh up Number of hours unit g has been online forat start of planning period (h)h down Number of hours unit g has been offline forat start of planning period (h)MLarge numberα, β, γ Scaling factorsBinary Variabless g (t) equal to 1 when a unit starts up at time tz g (t) equal to 1 when a unit shuts down at time tv g (t) equal to 1 when a unit is online at time tstep S g (t, i) equal to 1 when N S (t,1) ≥ Th S (i) at time tr g (t) equal to 1 when a unit undergoes ramp >R g between time t − 1 and tr g (t, j) equal to 1 when a unit undergoes ramp >R g (j) between time t − 1 and tstep R g (t, i) equal to 1 when N S (t,1) ≥ Th R (i) at time tPositive VariablesN S g (t) Cumulative start-ups for unit gN S g (t,i) Cumulative start-ups for unit g beyondthreshold Th S g (i)

2C S g (t) Total cycling cost attributed to start-ups forunit gN R g (t) Cumulative ramps > R g for unit gN R g (t,i) Cumulative ramps > R g beyond thresholdTh R (i) for unit gC R g (t) Total cycling cost attributed to ramping forunit gc p g(t) Production cost for unit g at time tc s g(t) Start-up fuel cost for unit g at time tp g (t) Output (MW) for unit g at time tD(t) System demand (MW) at time tδ l (g,t) Power produced in block l of the piecewiselinear production cost function of unit g attime t (MW)I. INTRODUCTIONINCREASED competition in the electricity generation sectorcoupled with the large-scale deployment of variablerenewable energy sources, particularly wind power, has ledto increased plant cycling in power systems worldwide [1],[2]. Cycling may be defined as frequent start-ups or rampingof units. Some generation types (such as hydro or even opencyclegas turbines) are more suited to frequent cycling, butfor others, particularly units designed for base-load operation,cycling can accrue large levels of damage within the plant’scomponents leading to increased maintenance requirementsand forced outage rates. Thermal shock, metal fatigue, corrosion,erosion and heat decay are common damage mechanismsthat result from cycling operation [3]. The wear-and-tear whicharises incurs increased maintenance costs for generators, butin addition to this, loss of revenue due to more frequent andlonger outages, increased fuel costs due to more frequent startupsand reduced plant efficiency, as well as additional capitalcosts due to component replacement can also be expected.Studies indicate that the magnitude of these cycling relatedcosts are high, but accurately quantifying them is challenging[4], [5]. The level of wear-and-tear for a unit that undergoescycling operation will be dependent on many factors includingthe operating history of the plant (i.e. how much creep damageit has accumulated), and the engineering design of the plant.It is also typical to see a time lag of several years from whencycling occurs to when the damage manifests itself [6].Research related to the cost of generation cycling has beenundertaken by EPRI and Intertek Aptech and the approachesemployed can be categorized as top-down (statistical analysis)or bottom-up (component modeling). EPRI carried out atop-down study utilizing multivariate regression models toanalyze the operating regimes of 158 units from NERC (NorthAmerican Electric Reliability Corporation) GADS (GeneratingAvailability Data System) and CEMS (Continuous EmissionMonitoring) data in an attempt to identify patterns relatingplant operation to capital expenditure. However, the inconsistencyin accounting practices between the units complicatedthe modeling and no correlation was found [7], [8]. IntertekAptech employ a combination of top-down models based onhistorical operations, forced outage and cost data as well asbottom-up methods which calculate operational stresses andthe life expenditure of critical components to determine cyclingcosts for individual generating units [4]. Intertek Aptechhave analyzed cycling costs for over 300 generating units andfound that the cost of cycling a conventional fossil-fired powerplant can be as much as $2,500-500,000 per start/stop cycledepending on unit age, operating history and design features,and are often grossly underestimated by utilities [4], [6].Not considering these costs, however, will result in an uneconomicplant dispatch, yet markets currently do not includespecific cycling cost components in their bidding mechanisms,or at best cycling costs are bundled into a generator’s startupor operating costs. Depending on the operating regime ofa plant, these cycling related costs can accumulate rapidlyand are therefore dissimilar to plant characteristics such asheat rate, which typically vary over a much longer time-scale.Therefore, to examine the impact of these costs accurately,they should be modeled in a dynamic manner such that theyaccumulate within the optimization process based on howthe unit is being operated and thereby can influence dispatchdecisions.This paper presents a novel formulation to dynamicallymodel these cycling costs, which can be integrated into a MIP(mixed integer programming) unit commitment and economicdispatch model. This facilitates more accurate modeling ofthese costs and examination of how they accumulate in linewith the operating regime of the plant. The formulation definesa cycling cost which increments with each additional plantstart-up or ramp with the resulting cost function being linear,piecewise linear or step-shaped. A case study is included todetermine how implementing dynamic cycling costs for a testsystem over a period of one year will affect the resultingdispatch, relative to a scenario where cycling costs are notconsidered. This new approach to modeling cycling costs isparticularly suitable for long-term planning studies where itcan be used to reflect the ageing effect on a plant over time.It may also have applications for real-world market modelswhere it can discourage the same unit from being repeatedlydispatched to cycle by incurring an incremental cost to reflectthe wear-and-tear to that unit, which can consequently alterits position in the merit order.The paper is organized as follows: Section II details theformulation of dynamic cycling costs, Section III describes aunit commitment model and economic dispatch model usedto implement the dynamic cycling cost formulation and alsodescribes the test system, Section IV details the results of thecase study and Section V summarizes the findings.II. FORMULATION OF DYNAMIC CYCLING COSTSA detailed formulation for implementing dynamic cyclingcosts which increase in line with unit operation is presented.Cycling costs are subdivided into costs for (A) start-ups and(B) ramps. The formulation utilizes three main steps: (i) abinary variable is set to indicate that damaging operation hasoccurred at time step t, (ii) a counter tracks how much of thattype of operation has occurred up to that point, and (iii) anincrementing cycling cost is incurred at that time step. Linear,piecewise linear and step-shaped cost functions for both startsand ramps are detailed here.

3A. Cycling costs related to start-upsLinear: Constraints 1-3 allow a dynamic, linearly incrementingcost for wear-and-tear related to start-ups to be modeled.Based on the online binary variable, v g (t), constraint 1sets the start-up, s g (t), and shut-down, z g (t), binary variablesequal to 1 appropriately, when unit g is started up or shutdown at time t. Constraint 2 increments a counter, Ng S (t),to track how many start-ups have been performed by thatunit. Constraint 3 determines the start-up related cycling cost,Cg S (t), with the final term ensuring that a cost is only incurredwhen the decision is made to start the unit at time t (i.e. s g (t)= 1). Figure 1 provides an example of this linearly increasingcost function, where the cycling cost increment cost S g is setequal to 100. (It is also possible to initialize the counter Ng S (t)with the number of starts that have been carried out previouslyif this is known).C S g (t) ≥I g∑i(N S g (t, i). ( cost S g (i) − cost S g (i − 1) ))− ( 1 − s g (t) ) .M, ∀ t ∈ T, ∀ g ∈ G(5)s g (t) − z g (t) = v g (t) − v g (t − 1), ∀ t ∈ T, ∀ g ∈ G (1)N S g (t) ≥ N S g (t − 1) + s g (t), ∀ t ∈ T, ∀ g ∈ G (2)C S g (t) ≥ N S g (t).cost S g− M. ( 1 − s g (t) ) , ∀ t ∈ T, ∀ g ∈ G(3)Fig. 2.Piecewise linearly increasing start-up related cycling costStep Function: Alternatively, if less information is knownregarding the shape of the cost function an appropriate simplificationmay be to define a step function, where C S g (t) doesnot increment until T h S g (i) is reached. Again, it is requiredthat T h S g (1) is equal to 1. N S g (t, i) is determined by constraint6 and in this case can be greater than or less than 0 (it waspreviously defined as a positive variable only). Constraint 7sets the binary variable step S g (t, i) equal to 1 when N S g (t, i)has exceeded T h S g (i), and constraint 8 determines the cyclingcost. Figure 3 provides an example of this incrementing, stepshapedcost function, where cost S g (t, 1) is set equal to 100,cost S g (t, 2) is set equal to 150 and T h S g (2) equals 4.N S g (t, i) =()Ng S (t − 1, 1) + s g (t) + 1 − T h S g (i),(6)∀ t ∈ T, ∀ g ∈ G, ∀ i ≤ I gFig. 1.Linearly increasing start-up related cycling costPiecewise Linear: By defining i thresholds, T h S g (i), eachcorresponding to a cumulative number of plant start-ups, atwhich point the start-up related cycling cost, C S g (t), willincrease by incremental cost cost S g (i) for each additionalstart, a piecewise linear incremental cost function can bemodeled. Constraint 4 is a modified form of constraint 2which counts the cumulative number of start-ups. For i >1, the start-up counter, N S g (t, i), will not have a positivevalue until N S g (t, 1) has reached T h S g (i). T h S g (1) must equal1. Constraint 5 determines the total cycling cost. Figure 2provides an example of a piecewise linearly increasing costfunction, where cost S g (1) is set equal to 100, cost S g (2) is setequal to 150 and T h S g (2) equals 4.N S g (t, i) ≥()Ng S (t − 1, 1) + s g (t) + 1 − T h S g (i),(4)∀ t ∈ T, ∀ g ∈ G, ∀ i ≤ I gFig. 3.N S g (t, i) − step S g (t, i).M ≤ 0,∀ t ∈ T, ∀ g ∈ G, ∀ i ≤ I g(7)C S (t) ≥ cost S g (i).step S g (t, i) − ( 1 − s g (t) ) .M,∀ t ∈ T, ∀ g ∈ G, ∀ i ≤ I g(8)Step increasing start-up related cycling cost

4Hot and Cold Starts: Either the linear, piecewise linear orstep formulations can be extended to differentiate between hotand cold start-ups for units. Constraint 9 will set the binaryvariable s coldg (t) equal to 1 only if unit g is started at timet, having been offline for Tgcold plus its minimum downtime,DT g . In constraints 2, 4 and 6 ‘+ s g (t)’ is replaced with ‘+s g (t) + α.s coldg (t)’. A scaling factor, α, is chosen based onthe ratio of cycling damage caused by a hot start relative to acold start, and thus normalizes N S g (t, i) to count in terms ofhot starts.s coldg (t) ≥ v g (t) −T coldg +DT ∑gn=1B. Cycling costs related to rampingv g (t − n), ∀ t ∈ T, ∀ g ∈ G1) Define one ramp level: The simplest form of incurringcycling costs related to ramping duty is to define a changein output, R g , between consecutive time periods, greaterthan which, damaging transients will occur within unit g.Constraints 10 and 11 ensure that the binary variable r(t)is set to 1 when a change in output exceeding R g occurs.To avoid double counting cycling costs when large ramps areexperienced in the start-up or shut-down process, the final termensures that the constraints are non-binding when the unitis in the start-up or shut-down process. If the ramp-relatedcycling costs are likely to exceed the start-up or shut-downcost, constraint 12 is needed to prevent the model setting s(t)and z(t) both equal to 1 in constraint 1, in order to makeconstraints 10 and 11 non-binding.(pg (t) − p g (t − 1) ) − M.r g (t) ≤ R g + M.s g (t),∀ t ∈ T, ∀ g ∈ G (10)(pg (t − 1) − p g (t) ) − M.r g (t) ≤ R g + M.z g (t),∀ t ∈ T, ∀ g ∈ G (11)(9)s g (t) + z g (t) ≤ 1, ∀ t ∈ T, ∀ g ∈ G (12)Utilizing the binary variable, r g (t), a counter N R g (t) isdefined, as before, to incur an incrementing, ramp-relatedcycling cost, C R g (t). Using the formulation from Section II.A,the ramp-related cycling cost function may be linear, piecewiselinear or step-shaped. Constraints 2 and 3 are replaced withthe analogous ramp terms shown in Table I to implement alinearly incrementing cost. Constraints 4 and 5, or 6 to 8, arereplaced with the analogous ramp terms as shown in Table I todefine a piecewise linear, or step shaped, incrementing ramprelated cycling cost respectively.2) Define multiple ramp levels: The previous formulation,where one level R g is set to define a ramp, can be expandedto incur a dynamic ramp-related cycling cost, for j ramps ofdifferent magnitudes, R g (j). Constraint 13 ensures that for aramp less than R g (1), the binary variable r g (t, j) will equalzero for all j. A ramp greater than R g (1), but less than R g (2),will set r g (t, 1) equal to one, and so forth. The final termTABLE IANALOGOUS TERMSStartsRampss g(t) r g(t)Linear cost S g (t) cost R g (t)N S g (t) NR g (t)C S g (t)C R g (t)s g(t) r g(t)Piecewise cost S g (t,i) cost R g (t,i)Linear & N S g (t,i) N R g (t,i)Step Th S g (t,i) Th R g (t,i)C S g (t) C R g (t)step S g (t) stepR g (t)ensures that the constraint is non-binding when the unit isstarting up. A corresponding constraint is needed for downramps, where ( p g (t) − p g (t − 1) ) in constraint 13 is replacedwith ( p g (t−1)−p g (t) ) and M.s g (t) is replaced with M.z g (t).Constraint 14 ensures that the binary variable, r g (t, j), whichindicates that a ramp ≥ R g (j) has occurred, can only have avalue of 1 for one ramp level j, at any given time. As before,constraint 12 is required to prevent s g (t) and z g (t) both beingset to 1, to make constraint 13 and its corresponding downramping constraint non-binding.(pg (t) − p g (t − 1) ) < R g (1). ( 1 −j∑r g (t, j) )j=1+ R g (2).r g (t, 1) + ... + R g (j).r g (t, j − 1)+ ¯P(13)g .r g (t, j) + M.s g (t),where R g (1) < R g (2) < R g (j)... < ¯P g ,∀ t ∈ T, ∀ g ∈ G, ∀ j ≤ ¯j gj∑r g (t, j) ≤ 1, ∀ t ∈ T, ∀ g ∈ G, ∀ j ≤ ¯j g (14)j=1As with hot and cold starts, scaling factors are used tonormalize N R g (t) to count in terms of one ramp level, as shownin constraint 15, where r(t, j) is expressed in terms of r(t, 1).Constraint 16 determines the total ramp-related cycling cost,shown here with a constant cost increment, cost R g , with thefinal term ensuring that the cost is only incurred in a timeperiod when a ramp (> R g (1)) occurs.N R g (t) = N R g (t − 1) + r g (t, 1) + β.r g (t, 2)+.... + γ.r g (t, j), ∀ t ∈ T, ∀ g ∈ G, ∀ j ≤ ¯j g(15)Cg R (t) ≥ Ng R (t).cost R g − ( j∑1 − r g (t, j) ) .M(16)j=1∀ t ∈ T, ∀ g ∈ G, ∀ j ≤ ¯j gTo combine this formulation of j ramp levels with i costthresholds (i.e piecewise linear) constraints 15 and 16 arereplaced by constraints 17 and 18, such that once N R g (t, i)reaches T h R g (i), C R g (t, i) will begin incrementing by cost R g (i).

5N R g (t, i) = ( N R g (t − 1, 1) + r g (t, 1) + β.r g (t, 2)C R g (t) ≥−+.... + γ.r g (t, j) + 1 ) − T h R g (i) (17)∀ t ∈ T, ∀ g ∈ G, ∀ j ≤ ¯j g , ∀ i ≤ I gI g∑i(N R g (t, i). ( cost R g (i) − cost R g (i − 1) ))j∑r g (t, j).M, ∀ t ∈ T, ∀ g ∈ G, ∀ j ≤ ¯j g(18)j=1To include a step-shaped ramp related cycling cost function,constraints 6-8 are replaced with the analogous terms forramping from Table 1.III. DISPATCH MODEL AND TEST SYSTEMTo examine how cycling costs, modeled dynamically, willimpact plant dispatch the new formulation was implementedin a conventional MIP unit commitment model based on [9],[10]. The unit commitment problem was formulated asMinimize ∑ t∈Tsubject to∑g∈Gc p g(t) + c s g(t) + C S g (t) + C R g (t) (19)∑p g (t) = D(t), ∀ t ∈ T (20)g∈Gp g (t) ≤ ¯P g .v g (t), ∀ t ∈ T (21)p g (t) ≥ P g .v g (t), ∀ t ∈ T (22)As per [9], a piecewise linear approximation of a quadraticproduction cost function for each unit was adopted, as representedby:c p g(t) = A g v g (t) +p g (t) =NL g∑l=1NL g∑l=1F lg δ l g(t), ∀ t ∈ T, ∀ g ∈ G (23)δ l g(t) + P g v g (t), ∀ t ∈ T, ∀ g ∈ G (24)δ 1 (g, t) ≤ T 1g − P g , ∀ t ∈ T, ∀ g ∈ G (25)δ l (g, t) ≤ T lg − T l−1g , ∀ t ∈ T, ∀ g ∈ G, ∀ l = 2..NL g − 1(26)δ NL (g, t) ≤ ¯P g − T NLg−1 − T l−1g , ∀ t ∈ T, ∀ g ∈ G (27)δ l (g, t) ≥ 0, ∀ t ∈ T, ∀ g ∈ G, ∀ l = 1..NL g (28)where A g = a g + b g P g + c g P 2 g.Start-up costs which were dependent on the period of timethe unit had been offline were modeled as follows:c s g(t) ≥ ( v g (t) − v g (t − 1) ) .hc g ∀ t ∈ T, ∀ g ∈ G (29)c s g(t) ≥ ( v g (t) −T coldg +DT ∑gn=1v g (t − n) ) .cc g ,∀ t ∈ T, ∀ g ∈ G(30)Minimum up time constraints were formulated by constraints31, 32 and 33. Equation 31 is only included if thenumber of hours a unit must remain online to satisfy itsminimum up time, B g , is greater than or equal to 1.t≤B∑ gtt+UT g −1∑n=t¯T∑n=t(1 − vg (t) ) = 0, ∀ g ∈ G (31)v g (n) ≥ UT g .s g (t), ∀ g ∈ G∀ t = B g + 1... ¯T − UT g + 1(vg (n) − s g (t) ) ≥ 0, ∀ g ∈ G∀ t = ¯T − UT + 2... ¯Twhere B g = max ( 0, v g (T)UT g -h upg +v g (T) )(32)(33)Minimum down time constraints were formulated usingconstraints 34, 35 and 36. Equation 31 is only included ifL g ≥ 1.t≤L∑ gtt+DT∑ g−1¯T∑n=tn=t(vg (t) ) = 0, ∀ g ∈ G (34)v g (n) ≥ DT g .z g (t), ∀ g ∈ G∀ t = L g + 1... ¯T − DT g + 1(1 − vg (n) − z g (t) ) ≥ 0, ∀ g ∈ G(35)(36)∀ t = ¯T − DT + 2... ¯Twhere L g = max ( 0, (1 − v g (T)).DT g − h downg +(1 − v g (T)) )The formulation was applied to the 10 unit test system usedin [9], [11], which was duplicated to give a 20 unit system, thusfacilitating a larger case study. The peak demand (1500 MW)was doubled (3000 MW) and a historical year-long hourlydemand profile for the Irish system was scaled to produce ademand profile with a 3000 MW peak. The model was runfor the test year, optimizing each day at an hourly resolution.Generator cycling costs are difficult to determine and largelyuncertain, as discussed in Section I. The figures used here,

6shown in Table II, to implement dynamic cycling costs for thetest system, are conservatively based on those in [12] and areintended to illustrate how dynamic cycling costs could impactsystem operation, rather than provide an accurate estimate ofsuch costs. Piecewise linear costs for starts and ramps were implementedwith the incremental cost (cost S g (i) or cost R g (i)) increasingby 10% and 20% when the start counter (N S g (t, 1)), orramp counter (N R g (t, 1)), exceeded 100 (T h S g (2) or T h R g (2))and 200 (T h S g (3) or T h R g (3)) respectively. The scaling factor,α, was chosen to be 2, i.e. each cold start incrementedN S g (t, 1) by 2 (while a hot start incremented N S g (t, 1) by 1).Two ramp levels, R g (1) and R g (2) corresponding to 20% and40% of the difference between maximum and minimum outputfor a unit, were modeled. Scaling factors were chosen such thatramps greater than R g (1) or R g (2) incremented N R g (t, 1) by1 or 2 respectively.TABLE IIINCREMENTAL CYCLING COSTS $, (I=1)Units cost S g (i) cost R g (i)1-4 300 155-10 60 311-20 30 1.5IV. RESULTSThis section examines how plant dispatches for the testsystem are affected over one year when (i) a cycling costrelated to start-ups is implemented, (ii) a cycling cost relatedto ramping is implemented, and (iii) cycling costs related tostart-ups and ramping are implemented simultaneously.A. Start-up Related Cycling Costs ResultsImplementing a dynamic cycling cost for plant start-ups, asshown in Table II, was seen to result in an overall reduction inplant start-ups. This is seen in Table III, which reveals reducingstarts for base-load and mid-merit units. For base-load units,the reduction in starts was correlated with increased productionas, having the largest incremental cycling costs, these unitsavoided shut-downs and their online hours increased. This isevident through the average capacity factor shown in TableIV. Mid-merit units, however, which also had reduced starts,saw reduced production indicating that they were utilized lessoften. As these units were started up and shut down, andsubsequently incurred cycling costs, it became more economicalafter some point to dispatch peaking units. Thus, startsand production increased for peaking units when a dynamiccycling cost for start-ups was modeled. Figure 4 illustratesthe cumulative start-ups for the mid-merit and peaking unitsover the year when (i) cycling costs were modeled and (ii)when cycling costs were not modeled. Starts are seen toaccumulate rapidly between 0 and 2000 hours and for hoursgreater than 7000, as these are the winter months and thushave higher demand, requiring more plant start-ups. Beyond1000 hours the cycling costs which are accumulated by midmeritplant begin to have an effect on their position in themerit order and consequently peaking plant are seen to bedispatched more frequently. Modeling dynamic cycling costsrelated to plant start-ups was also found to have the knockon effect of increasing generator ramping. Over the year a22% increase in ramping (N R g (t, 1)) was observed relative tothe case when no cycling costs were modeled as generatorswere more frequently ramped down to minimum output, ratherthan shut-down, in an effort to avoid incurring cycling costsfor starting up.TABLE IIIIMPACT OF DYNAMIC CYCLING COSTS FOR START-UPS ON TOTAL ANNUALSTARTSNo cycling Cycling cost forUnits costs modeled starts modeledBase-load (Units 1-4) 34 12Mid-merit (Units 5-10) 1372 1005Peaking (Units 11-20) 577 838Total 1983 1855TABLE IVIMPACT OF DYNAMIC CYCLING COSTS FOR START-UPS ON AVERAGEPLANT CAPACITY FACTORS (%)No cycling Cycling cost forUnits costs modeled starts modeledBase-load (Units 1-4) 92.59 92.73Mid-merit (Units 5-10) 27.82 25.42Peaking (Units 11-20) 0.85 2.23Fig. 4. Cumulative plant start-ups over the year, shown when dynamic cyclingcosts for starts were (i) modeled and (ii) not modeledUnits within the same class, i.e. base-load, mid-merit orpeaking, were also seen to converge to a similar number ofannual start-ups, as indicated by the reduced standard deviationof annual start-ups seen in Table V. This indicates that oncea unit has been cycled and its cycling cost is incremented,the next time a unit needs to be cycled the costs will havenow changed such that a different unit (most likely the nextin the merit order) may be scheduled. This leads to the burdenof cycling operation being more evenly distributed across theunits. Over a long horizon, i.e. several years, this effect canlead to a shift in the merit order, a trend which can be seenin Figure 4.To facilitate a sensitivity analysis, multiples of the initialincremental cycling costs, cost S g (1), shown in Table II, werealso examined. As the incremental cost was increased thereduction in start-stop cycling that is achieved by modeling

7TABLE VIMPACT OF DYNAMIC CYCLING COSTS FOR START-UPS ON TOTAL ANNUALSTARTSNo cycling Cycling cost forcost modeled starts modeledUnits Avg. Std. Dev. Avg. Std. Dev.Base-load (Units 1-4) 8.5 9.9 3 3.6Mid-merit (Units 5-10) 228.7 75.7 167.5 26.1Peaking (Units 11-20) 57.7 73.1 83.8 27.5dynamic cycling costs quickly saturated as seen in Figure 5,thus indicating that the majority of plant cycling is unavoidable.Table VI shows a breakdown of the total number ofstarts by unit group, which again reveals that increasing startsfor peaking units are correlated with increasing incrementalcycling cost, as it becomes more favorable to dispatch theseunits due to the relatively larger cycling costs associated withthe mid-merit units. (The ripples in the curve shown in Figure5 result from the increasing starts for peaking units, as seenin Table VI.)Fig. 5. Impact of dynamic cycling cost on total start-ups, shown for variousmultiples of cost S g (i)TABLE VIIMPACT OF DYNAMIC CYCLING COSTS FOR STARTS ON TOTAL PLANTSTART-UPS, SHOWN FOR VARIOUS MULTIPLES OF cost S g (i)Base-load Mid-merit Peaking(Units 1-4) (Units 5-10) (Units 11-20)No cycling cost 34 1372 577cost S g (i)*0.5 13 1104 781cost S g (i)*1 12 1005 838cost S g (i)*2 13 941 896cost S g (i)*3 13 907 948cost S g (i)*10 13 869 992A scenario where cycling costs were only modeled for asubset of the total fleet was also examined. The 6 largest unitson the system (units 1, 2, 3, 4, 9, 10) were chosen basedon the assumption that these units would be most impactedby cycling operation and thus most likely to bid a wear-andtearcost into the market if such an option was available. Theresults showed that although the number of annual start-upswas reduced for these units, the start-ups for the other unitsincreased by a much greater amount as seen in Table VII. Thiswould indicate the need for a uniform policy relating to thebidding of cycling costs to be implemented in markets, suchthat all units reflect their cycling costs, or do not, to avoidthe situation where only some generators are bidding cyclingcosts as this leads to inefficient operation and excessive costs.TABLE VIICHANGE IN STARTS WHEN A SUBSET OF UNITS BID CYCLING COSTS FORSTART-UPS∆ StartsUnits 1, 2, 3, 4, 9, 10 -86All other units +256B. Ramping Related Cycling Costs ResultsImplementing a dynamic cycling cost for plant ramping(shown in Table II) resulted in a 90% reduction in rampingoverall, as seen in Table VIII. As described previously, assuminga ramp greater than 20% or 40% of the difference betweena unit’s maximum and minimum output increments the rampcounter, N R g (t), by a value of 1 or 2 respectively. The totalvalue of N R g (t) at the end of the test year, summed for allunits, is shown in Table VIII. Base-load units which carriedout the greatest amount of ramping when cycling costs werenot modeled, saw the greatest reduction in ramping operationwhen cycling costs for ramps were implemented. The dramaticreduction in ramping that was achieved by implementingdynamic ramping costs, however, led to increased start-stopcycling as might be expected, although only by 3.3% overthe year. The most notable change to the overall dispatch thatresulted from the introduction of dynamic ramping costs wasa slight reduction in production from base-load plant allowingfor increased production from mid-merit and peaking units asseen in Table IX, thereby spreading the ramping requirementover more units. Thus, including the ramping cost was alsoseen to result in a slightly greater number of units online(5.94 per hour on average when dynamic ramping costs weremodeled, versus 5.92 when no cycling costs were modeled).TABLE VIIIIMPACT OF DYNAMIC CYCLING COSTS FOR RAMPING ON TOTAL ANNUALRAMPING (N R g (t, 1))No cycling Cycling cost forUnits costs modeled ramps modeledBase-load (Units 1-4) 3717 120Mid-merit (Units 5-10) 2214 1224Peaking (Units 11-20) 795 623Total ramping 6726 1967TABLE IXIMPACT OF DYNAMIC CYCLING COSTS FOR RAMPING ON AVERAGE PLANTCAPACITY FACTORS (%)No cycling Cycling cost forUnits costs modeled ramps modeledBase-load (Units 1-4) 92.59 92.21Mid merit (Units 5-10) 27.82 28.61Peaking (Units 11-20) 0.85 1.02C. Start-up and Ramping Cycling Costs ResultsImplementing dynamic cycling costs (as shown in TableII) for starts and ramping simultaneously, reduced both types

8of cycling operation relative to the case when no cyclingcosts were modeled, as shown in Table X. Base-load units,having the largest cycling costs, see the greatest reductionsin cycling operation. Nonetheless, neither total starts nor totalramps were reduced in this scenario as much as starts aloneor ramps alone were reduced when cycling costs for startsor ramps were modeled individually. However, when cyclingcosts for start-ups only were modeled, ramping operationincreased and likewise when cycling costs for ramping onlywere modeled, starts increased. Thus when the cycling coststhat would have been incurred due to both start-ups andramping are examined (assuming the costs given in TableII), the case in which cycling costs for start-ups and rampingwere modeled simultaneously had the lowest overall cyclingcosts, as shown in Figure 6. This would indicate that modelingcycling costs for starts and ramping simultaneously is the mostcost effective way to reduce cycling and as such one shouldnot be considered without the other.TABLE XIMPACT ON TOTAL ANNUAL STARTS AND RAMPS WHEN DYNAMICCYCLING COSTS FOR BOTH START-UPS AND RAMPING WERE MODELEDNo cycling costs Cycling cost for startsUnits modeled and ramps modeledStarts Ramps Starts RampsBase-load (Units 1-4) 34 3717 12 144Mid merit (Units 5-10) 1372 2214 1003 2069Peaking (Units 11-20) 577 795 855 1456Total 1983 6726 1870 3669Fig. 7.Total system costs shown for various scenariosbeing determined by the level of knowledge of the generator’scycling costs.The formulation for piecewise linear incremental cyclingcosts related to plant start-ups and ramps was implementedfor a test system. Although the incremental costs chosen areapproximations, the results reveal certain trends that are likelyfor power systems where generators undergo regular cyclingand reflect the resulting wear-and-tear costs in their bids. Forexample, dynamically modeling cycling costs for generatorstarts was seen to reduce the number of starts, but causedramping operation to be increased (and vice-versa), whilstmodeling cycling costs for only a subset of the generationfleet was seen to induce much higher levels of cycling in theremaining generation. It was also seen that as cycling costsaccumulated over time changes in the merit order occurred,and that modeling cycling costs led to an overall saving forthe system as cycling operation was subsequently reduced.Fig. 6. Cycling costs (that would have been incurred) shown for variousscenariosFinally, when total system costs are examined for thescenario including cycling costs and compared to the totalsystem cost for the scenario in which cycling costs were notmodeled, but were calculated and added afterwards, it can beseen that modeling cycling costs leads to lower system costsoverall. This is shown in Figure 7. In this example, the costsaving seen is considerable i.e. 14%.V. CONCLUSIONSInterest concerning cycling costs is growing and this papersets out a formulation that can utilize knowledge ofincremental wear-and-tear costs related to plant start-ups orramping, to implement a dynamic incrementing cycling cost.The formulation covers linear, piecewise linear and stepshapedcycling cost functions, the appropriate choice for a userREFERENCES[1] L. Göransson, and F. Johnsson, “Large scale integration of wind power:moderating thermal power plant cycling”, Wind Energy, vol. 14, no. 1,pp. 91-105, 2011.[2] N. Troy, E. Denny and M. O’Malley, “Base-load cycling on a system withsignificant wind penetration”, IEEE Transactions on Power Systems, vol.25, issue 2, pp. 1088 - 1097, 2010.[3] “Damage to Power Plants Due to Cycling”, EPRI, Palo Alto, CA, 2001.[4] Editorial, “Profitable Operation Requires Knowing How Much it Coststo Cycle your Unit,” Combined Cycle Journal [online], Spring 2004,available: http://www.combinedcyclejournal.com/[5] S. Lefton, “Profitable Operation Requires Knowing How Much it Costs toCycle your Unit”, Combined Cycle Journal, pp. 49-52, Second Quarter,2004.[6] S. Lefton, P. Besuner, P. Grimsrud, A. Bissel and G. Norman, “Optimizingpower plant cycling operations while reducing generating plant damageand costs at the Irish Electricity Supply Board”, Aptech EngineeringService, Tech. Rep. 123, Sunnyvale, CA, 1998.[7] “Determining the Cost of Cycling and Varied Load Operations: Methodology”,EPRI, Palo Alto, CA: 2002. 1004412.[8] “Correlating cycle duty with cost at fossil fuel power plants”, EPRI, PaloAlto, CA: 2001. 1004010.[9] M. Carrion and J.M. Arroyo, “A computationally efficient mixed-integerlinear formulation for the unit commitment problem”, IEEE Transactionson Power Systems, vol. 21, no. 3, pp. 1371 - 1378, 2006.[10] J.M. Arroyo and A.J. Conejo, “Optimal response of a thermal unit toan electricity spot market”, IEEE Transactions on Power Systems, vol.15, no. 3, pp. 1098 - 1104, 2000.[11] S.A. Kazarlis, A.G. Bakirtzis and V. Petridis, “A genetic algorithmsolution to the unit commitment problem”, IEEE Transactions on PowerSystems, vol. 11, no. 1, pp. 83-92, 1996.[12] S. Lefton, P. Besuner, D.D. Agan, “The real cost of on/off cycling”,Modern power systems, vol. 26, no. 10, 2006.