The Impact of Large-scale Solar Photovoltaic A4option on Prices and Reliability in the New England Power Pool by Katherine C Martin B.A. Physics, Reed Cqllege, 2000 Submitted to the Engineering Systems Division in Partial Fulfillment of the Requirements for the Degree of Master of Science in Technology and Policy at the Massachusetts Institute of Technology February 2006 ©2006 Massachusetts Institute of Technology. All rights reserved Signatureof Author___ Engineering Systems Division January 12, 2006 .. Certified by Accepted by - -- - v II. X " A , --- , _ Cri _--- David H. Marks Morton '42 and Claire Goulder Family Professor of Civil and Environmental Engineering and Engineering Systems Thesis Supervisor (T~~ Dava J. Newman v Professor of Aeronautics and Astronomics and Engineering Systems Director, Technology and Policy Program _ ..... .l -KansasNs~t-tUTE, OF TECHNOLOGY ARCHI/ES LIBRARIES . An 2 The Impact of Large-scale Solar Photovoltaic Adoption on Prices and Reliability in the New England Power Pool by Katherine C. Martin Submitted to the Engineering Systems Division on January 12, 2006 in Partial Fulfillment of the Requirements for the Degree of Master of Science in Technology and Policy ABSTRACT The potential role of renewable energy in producing electricity in industrialized countries has been gaining attention as issues like climate change and high fossil fuel prices more frequently occupy the minds of the public and policymakers. The higher levels of these technologies in power systems that are anticipated, or at least believably envisioned, make inquiries about their likely impacts on power systems relevant. The intermittency of technologies like wind and solar photovoltaic (PV) systems drives debates about whether larger scale adoption of these technologies will increase the cost of maintaining power system reliability. A more general question is simply whether these technologies will be like conventional capacity in their interaction with the larger power system, both physically and economically. These issues relate to the higher-profile debates about standby charges and if they are warranted. To address these questions, in part, for the New England power pool this study uses detailed historical data in two ways: to analyze the potential impact of large-scale adoption of PV on wholesale power prices and to compare the generation of PV systems in New England to the typical operation of conventional peaking units. This study identifies incentives for owners of incumbent, conventional generators to resist largescale adoption of PV and how these relate to debates about intermittency. In the short term, the effect of PV on price-spikes would be similar to the installation of more natural gas-fired combustion turbines. In this sense, incentives of owners of base-load generators to resist PV, though real, are not specific to PV. Comparison of PV generation to typical combustion turbine operation, however, suggests that growth of PV could exacerbate problems with ability of combustion turbines and other "peaking" units to cover longterm costs. This comparison shows how PV could increase the costs of maintaining power system reliability in the long-term and why owners of peaking units might resist large-scale PV adoption. It also suggests solutions in the form of considerations for policy design for policies aimed at encouraging the use of PV systems. Thesis Supervisor: Professor David H. Marks Title: Morton '42 and Claire Goulder Family Professor of Civil and Environmental Engineering and Engineering Systems 3 4 CHAPTER I - INTRODUCTION AND SUMMARY .................................................................................. 7 PVand price spikes- summary.................................................................................................................. 8 Comparison of PV to CT units - summary..............................................................................................10 Incentivesfor incumbentsto resist large-scalePVadoption- summary...............................................11 SUMMARY OF THESIS AND DATA SOURCES ..................................................................................................... 13 CHAPTER 11- THEORY ................................................................................................................................ 15 Price spikes................................................................................................................................................. 15 PV in the context ofprice spikes ..............................................................................................................16 Normal and high prices ............................................................................................................................. 17 PV in the context of normal and high prices ............................................................................................ 18 How could PV reduce power prices? ........................................................................................................ 19 CHAPTER II111 - THE NEW ENGLAND POWER POOL ......................................................................... 21 CHAPTER IV - METHODOLOGY AND RESULTS ................................................................................ 25 PV AND PRICE SPIKES ..... ............................................................................................................................... 25 Supportfor the methodology's assumptions............................................................................................. 26 CYCLES OF PRICE SPIKES AND INVESTMENT: OBSERVATIONS OF NEPOOL ................................................. 27 Demand, net revenues, and capacity in New England ............................................................................ 28 Net revenues and capacity......................................................................................................................... 29 Price statistics............................................................................................................................................ 31 Number ofprices above peaking-unit costs.............................................................................................. 33 Number of scarcity events.......................................................................................................................... 35 PV ANALYSIS: EMPIRICAL RESULTS ................................................................................................................ 35 Would PV impact price spikes like the installation of conventional capacity?.....................................35 PV results in context of conventional capacity ........................................................................................ 37 How much P V would make an impact?.................................................................................................... 38 Conclusions40Conclusions ................................................................................................................................................ COULD PV REPLACE COMBUSTION TURBINE UNITS IN NEW ENGLAND?9 ...................................................... 40Conclusions40 41 Why compare P V to combustion turbine units?....................................................................................... 41 PV's impact on price in non-price-spike hours........................................................................................47 CHAPTER V - INCENTIVES HINDERING PV ADOPTION.................................................................51 Impact ofP V on baseload unit revenues .................................................................................................. 51 Standbychargesand the impactof PV on operationof CT peakingunits............................................. 54 Recommendations for problems presented by incentives to resist P V ...................................................57 CHAPTER V - CONCLUSION 59 Further w60Further work............................................................................................................................................... w60 60Further BIBLIOGRAPHY .............................................................................................................................................. 62 APPENDIX A63 A ..................................................................................................................................................... A63APPENDIX 63APPENDIX SOLAR RESOURCE AND PV GENERATION IN NEW ENGLAND ........................................................................ 63 5 6 CHAPTER - INTRODUCTION AND SUMMARY The potential role of renewable energy in producing electricity in industrialized countries has been gaining attention as issues like climate change and high fossil fuel prices more frequently occupy the minds of the public and policymakers. The manufacturing costs of these, still expensive, alternative technologies have been decreasing partly as a result of policies in a few industrialized countries that encourage the their use on a larger scale.' In the United States, a few specific proposals have garnered publicity and the number of similar projects will likely grow.2 With this attention come questions about the interaction of non-dispatchable electricity technologies in electric power systems. At present levels of penetration in the United States these technologies are of little concern with regard to overall power system operation. But, higher levels that are anticipated, or at least believably envisioned, make inquiries about increased penetration relevant. This study explores whether large penetration of photovoltaic power systems (PV) in the New England Power Pool (NEPOOL) would interact in the power system in a similar way to the installation of more conventional peaking capacity. It does this by using historical data in two ways: 1) it examines the relative timing of price spikes and PV generation and estimates the magnitude of impact of PV on price spikes, in the short term, for a series of hypothesized amounts of PV capacity, and 2) it compares the typical Agreements between the Danish government and utilities starting in 1985 and the implementation of a green certificates market through the Electricity Reform Agreement and Electricity Act in 1999 helped the level of wind energy reach about 15 percent of annual electricity consumption (up from around zero percent in 1985) in Demark by 2000 with slight increases after. See "Wind Energy in Denmark" at http://www.opet.dk/windsector/wind-intro.litml last viewed July 18, 2005. Japan and Germany used subsidies to encourage the adoption of PV capacity that reached levels of about 500 and 200 MW installed capacity in 2001. See "Photovoltaic Industry Statistics: Countires" at http:/Avww.solarbuzz.com/StatsCountries.htm last viewed July 18, 2005. 2 In February of 2005, the governor's office in California released a plan that would install 3, 000 Megawatts (MW) of solar power in the state by 2018, enough to supply about 5% of the state's peak electricity demand. The bill is SB I see "Official California Legislative Information" at hlttp://info.sen.ca.gov/cgi-bin/postguery?bill number sb &sess=CUR&house=B&site:sen last viewed July 18,2005. Cape Wind proposed to build 420 MW of wind capacity offshore in Nantucket Sound in Massachusetts 2001 and the project is still a controversy in 2005. See "Cape Wind" at !ttp:/www.caewind.org/ last viewed July 18, 2005. 7 pattern of operation of peaking combustion turbine (CT) units to the diurnal pattern of PV production to help understand how the introduction of a large amount of PV capacity would affect.the operation of conventional units and the need for further investment in them. Using this analysis, this study then identifies incentives for owners of incumbent, conventional generators to resist large-scale adoption of PV and how these relate to debates about intermittency. PV andpricespikes- summary In a competitive power system, the impact of the installation of a large amount of PV on prices should not differ from the installation of more conventional capacity, if the PV systems can generate at the right times. One question this study addresses is whether PV, given that it is a non-dispatchable technology, would act like conventional capacity in NEPOOL by interacting with prices through the following mechanisms: 1) The addition of capacity could reduce the frequency and severity of price spikes if the technology can generate during hours of capacity shortages. 2) The addition of capacity could reduce prices during non-price spike hours by causing a lower-cost unit to set the real time marginal price. Using historical data to address these questions is a somewhat limited approach because the system has been evolving and will continue to evolve from the state described by these data. But, a baseline understanding of typical system operation and how PV fits is helpful. Carefully specifying comparisons of PV to the operation of certain types of conventional units leads to conclusions that provide reasonable expectations for what would most likely happen in the future. 8 PV is sometimes put forth as an option to help mitigate the problems that arise from peak and inelastic demand because solar resource, and therefore PV's production curve, matches electricity demand fairly well. This correspondence occurs because humans are active mostly during daylight hours and because hot, sunny days drive air conditioning use. Two qualifications are important to note at the outset. First, PV's production curve does not perfectly match the diurnal curve for electricity demand. This will be discussed further in later Chapters. Second, even if PV generation can reduce peak demands, the effect would only be short-term and therefore a temporary patch to the problems caused by peak demand unless investment in PV grows apace with increases in demand. The primary conclusions from the analysis of PV's relationship to price spikes in New England are: 1) In the short term, the effect of PV on price-spikes would be similar to the installation of more natural gas-fired combustion turbines. 2) A large amount of PV capacity (e.g. 1000 MW) would be needed to impact price spikes similarly to the installation of a smaller amount of CT capacity (e.g. 200 MW). 3) PV would not greatly impact prices in non-price spike hours in New England because the marginal-price-setting fuel is natural gas in over 85% of hours. These conclusions inform debates about whether renewables will increase power system costs and what incentives owners of incumbent generators have to resist the adoption of PV. These topics are discussed later, in conjunction with the analysis that compares diurnal PV generation to the typical operation of CT peaking units. 9 ComparisonofPV to CT units- summary The analysis described above is focused primarily on the short-term impacts of the installation of a large (around 1000 MW) of PV capacity. In the short term, the installation of any technology that can generate during scarcity hours will mitigate price spikes. In general when there is excess capacity in a power system, price spikes will not occur. The above analysis suggests that PV's production curve is well matched with the historical occurrence of price spikes in New England. But, if the goal is to install PV in place of conventional peaking technologies, it is important to consider what would happen in the long term as well. To address this long-term question using historical data, this study compares the typical operation of conventional peaking units in New England to the PV production. The primary conclusions from this analysis are: 1) In the long term, even if PV capacity grew apace with demand, PV's inability to generate in moderately high-demand, evening hours limits its ability to relieve scarcity conditions and therefore price spikes. 2) The magnitude of the mismatch between PV production and the typical operation of peaking capacity is relatively small compared to the correlation between the two between 7 am and 4 pm. 3) Daytime intermittency (e.g. cloudy but hot days) limits PV generation, compared to typical CT generation, during a small number of hours compared to the problem that PV cannot generate in evening hours. 4) Carefully bundling PV and other technologies to (with high probability) fully replace the typical operation of conventional peaking units is a possible way to introduce large amounts of PV into power 10 systems without increasing system costs due to the need for standby units. Little work has been done that carries out very detailed comparisons of PV generation to the typical (not idealized) operation of conventional units in specific power systems. If capturing the environmental and fuel-use-reduction benefits of renewable technologies like PV become priorities, planning and designing models around these detailed analyses could help policy makers and system operators understand how to best encourage the use of alternative technologies. Also, evidence that the installation of certain combinations of renewable technologies would offset the need for new conventional generators could help local communities feel that sacrificing their backyards is worth the effort. Incentivesfor incumbentsto resist large-scalePVadoption - summary One possible reason that owners of baseload capacity might resist large-scale PV penetration is the reduction of revenues caused by the reduction of price spikes. In the short term however, the installation of PV would reduce price spikes, and therefore the revenues that baseload plants and other plants earn from them, similarly to the installation of any new type of capacity. The incentives for baseload plant operators to resist the installation of capacity are not different for the installation of PV than they are for the installation of any capacity in the short term. In the long term, if the installation of PV reduced the number of hours that natural gas set the marginal clearing price, the introduction of PV would reduce revenues to baseload plants more than the installation of new natural gas peaking units. In New England, with the marginal price set by natural gas in over 85% of hours, a large amount of power is supplied by natural gas in the mid-day, high-demand hours in which PV can generate. It is unlikely that PV would change the marginal-price-setting fuel at any reasonable level of penetration. 11 One of the controversies surrounding the adoption of large amounts of renewable capacity in power systems is whether or not or to what extent standby charges should be levied to offset the costs of maintaining conventional generators that are infrequently used but that can be dispatched when the intermittent technologies cannot generate. If redundant conventional capacity is needed to support the use of renewable technologies like PV, then the costs of supplying power increase substantially. This analysis shows that the installation of large amounts of PV in New England would not entirely offset the need for conventional peaking units like CTrunits that generate in less than 1500 hours per year. If large amounts of PV were installed to fill increased demand, new CT units would still be needed for system reliability mostly because of evening hours in which peaking capacity is needed but PV cannot generate. Thus, CT units would operate less than at present and would be less able to make up their fixed costs, which they already struggle to cover. This analysis suggests that planning to couple PV with storage or other generating technologies could offset most of the need for back-up capacity because most of the problem with intermittency stems from evening hours rather than hours stochastically throughout the day. Designing ways to fill increases in demand with portfolios of PV and other technologies in such a way that the portfolio would cover, with very high probability, the typical operation of the conventional units that would be installed otherwise seems a promising approach to planning for the increased use of alternative technologies in power systems. In this way, arguments that standby units would be needed that would increases costs could be met with analyses of why they would not be necessary. 12 Summaryof thesis and data sources This study uses hourly historical data on prices, total demand, combustion turbine unit generation, and PV system generation. The hourly price and demand data for New England for 2000 through 2004 are available on the system operator's (ISO-NE) website.3 Capacity data by fuel and technology type for this time period are also available on ISO-NE's website.4 Hourly generating unit operation data are available from the Environmental Protection Agency's Continuous Emissions Monitoring System (CEMS).5 Hourly PV generation data are from five PV sites installed throughout New England for 2000 through 2002.6 Scaling the generation from these sites produced the generation from hypothesized amounts of PV capacity. Appendix A reports the details of the PV site locations, characteristics, generation, and scaling. This study first uses these data to find evidence to support the applicability of the key price and capacity relationships, outlined above and in Chapter II, in New England between 2000 and 2004. Chapter III outlines relevant characteristics of the New England Power Pool. Chapter IV then analyzes PV generation the context of the first two chapters and finds that PV could impact prices spikes similarly to conventional capacity. Because the installation of PV could thus impact the market signal that more capacity is needed, the extent to which PV could supply power when needed is analyzed. Recent investments in capacity in New England have predominantly been combustion turbine and combined cycle generators. PV generation is compared to the typical operation of combustion 3 ISO-NE website "Hourly Historical Data" at http:/,www.iso-ne.com/markets/hstdataindex.html last viewed July 18, 2005 4 Data derived from New England's "Capacity, Energy, Load, Transmission" (CELT) and "Seasonal Claimed Capability" (SCC) reports available at http://www.iso-ne.comni, trans/celt/report/ and http://www.iso-ne.cornm/genrtion.. resrcs/snl clmd cap/index.html respectively, downloaded July 7, 2005. 5 Environmental Protection Agency's Continuous Emissions Monitoring System (CEMS) (unit generation and heat input data) at htt ://www .epa.gov/airmarkets"ein 6 issions"#prelim Schott Applied Power provided the PV system generation data for 2000 through 2002. 13 turbine (CT) generators in New England and the analysis found that PV could not replace their operation entirely. If PV could not replace combustion turbine operation, it certainly could not replace units that operate more frequently. These results suggest that if PV were installed in place of conventional capacity, its impact on price spikes would be limited to the short term. Once PV was installed, scarcity conditions would likely arise when PV could not generate but:when new conventional capacity would have, causing price spikes in a different set of hours than those in which they now usually occur. The comparison of PV to CT units suggests PV. would have little impact on non-price spike prices. Chapter V discusses how these analyses highlight important considerations regarding the integration of PV into power systems and lead to two insights regarding incentives that incumbent operators of conventional power plants have to resist the largescale installation of PV. The first is whether PV would, by reducing high prices, reduce the revenues of baseload generation owners significantly more than does the installation of additional conventional capacity - or if baseload owners might perceive this to be a problem. The second is that current and future owners of conventional peaking generators may fear that the mismatch between PV and the typical operation of their generators would leave their units necessary for reliability but lacking the revenues needed to cover fixed costs. This analysis suggests that intermittency, a commonly sited challenge, may not be the most substantial reason behind arguments that standby charges are needed. Chapter VI concludes and suggests further research. 14 CHAPTER II - THEORY The extent to which PV can reduce demand and how this will affect generator operation and prices is the subject of this paper's empirical analysis. To aid this analysis, this chapter summarizes the most relevant points of the theory and operation of competitive power systems. Price spikes Price spikes are periods in which the wholesale price rises and falls quickly and when it exceeds the variable cost of the highest-cost generating unit in the system.7 Spikes occur when available capacity cannot cover electricity demand. In reality this does not mean a power outage, but rather that operating reserve margins for spinning reserve or tenminute reserve cannot be maintained. In New England, price spikes generally occur on extremely hot days in afternoon hours when demand for electricity is the highest or during unexpected cold snaps. Extreme weather and the subsequent strain on the power system can cause or be accompanied by equipment failure that amplifies the problems caused by the shortage of available capacity. The revenues that high-cost peaking units, and other units, collect during price spikes helps cover their fixed costs. These revenues are often referred to as short-term profits or inframarginal or scarcity rents reflecting the fact that the competitive price exceeds the variable costs of the generating units. As long as the strategic withholding of capacity, the exercise of market power, does not cause the high prices, they are an innate and important characteristic of competitive power markets. When installed capacity is relatively low, price spikes will occur frequently enough to encourage investment in new capacity. After the installation of new capacity, price spikes will subside as long as that capacity is available during times of highest 7Stoft 2002 pg. 451 15 demand or when other capacity is off-line. When the frequency and duration of price spikes is low, potential revenues for new plants will not be high enough to encourage new investment. If demand grows, price spikes will again encourage investment. PV in the context ofprice spikes Although the relationship between demand, capacity, and price spikes is complex and unpredictable there are certain relationships that hold. First, price spikes are more likely to occur, and occur more frequently, when capacity margins are slim. Second, if PV is to act like conventional capacity in bolstering slim capacity margins, then it must be able to generate when capacity margins are small because of high demand.This is not trivial as with the installation of a new combustion turbine generator because PV is not dispatchable. Its availability during critical times relies on solar resource. Whether PV generation usually occurs during the most critical times, or at least during those that occurred historically, is a testable question. Third, and again due to variability of solar resource, the generation from an amount of installed PV capacity will vary diurnally, somewhat predictably, and from day to day. It is possible to estimate how much PV capacity, in a given geographic region, is comparable to the installation of a given amount of new conventional capacity. Would one need to install 150 MW or 1500 MW of PV to (approximately) avoid the need for 150 MW of new combustion turbine capacity? This is also an answerable question. The above questions focus on the short-term. In power systems, the short-term refers to the ability of the system to generate enough to fill demand in any given hour. If PV can help mitigate capacity shortages or cause a generator not to generate in an hour, it could impact prices in the short term. These are important concerns. It is, however, essential to consider the long-term, which means understanding the potential impact of PV on the cycles of capacity installation, demand growth, and prices. 16 One way to pose the long-term inquiry is to ask: could PV capacity replace the need for new conventional capacity? Answering this question requires understanding the typical operation of the type of conventional capacity PV would replace and how the PV production curve for New England fits with this typical operation. Additionally, one must again understand how much PV would replace a given amount of conventional capacity as 150 MW PV capacity will not necessarily be equivalent to 150 MW of combustion turbine capacity even if they operate during exactly the same hours of the day. Normaland highprices In this analysis the term "normal prices" means those in which investment worthy units are setting the price. In New England in 2002, for example, the ISO identified prices between $20/MWh and $70/MWh as normal.8 The term "high" prices loosely means those around the high-side of "normal" (e.g. around $70/MWh) and those where peaking units that are no longer investment-worthy (inefficient) set the RTMP. High prices occur more often than price spikes, they occur when the level of demand requires generation from peaking units on the steep right side of the supply curve. These are the units with highest variable costs that include the old, inefficient ones and units that bid "emergency" blocks of generation beyond rated capacity. If capacity is added in the form of new peaking units that are more efficient than some of the old peaking units, non-price spike prices will decrease in hours when the new capacity sets the marginal price in place of the older capacity. With the addition of new baseload capacity, prices are reduced when the same level of demand can be met without the operation of the most expensive units previously operating to fill that level of demand. 8 ISO-NE 2003b pg. 40 17 PV in the contextof normaland highprices Adding PV capacity to the system-can be thought of in two ways. PV is a peaking technology because it only generates during hours in which the sun is shining. PV will usually not set the marginal price because its variable costs are effectively zero and units with higher variable costs will always be operating and needed to fill the nextMW of demand. Thus, in the hours PV generates, one could think of the PV adding to the supply curve at the base (lower left). The addition of enough PV generation to replace that of the highest-cost unit could reduce the price: the lowest-cost unit that does not generate, the unit that has slightly higher variable cost than the last one utilized, sets the marginal price. PV generation added to a power system can also be thought of as a decrease in demand. Thus, the PV generation could cause the demand curve to intersect the supply curve at a point where a lower-cost unit would set the marginal price. When comparing the addition of PV capacity to the addition of a new conventional peaking unit (like a natural gas-fired combustion turbine) it is useful to think of PV generation as a reduction in demand. New peaking capacity will generate when the marginal price is greater than its variable costs. It will reduce prices when it generates if it is more efficient than other generators in the system and thus its addition causes a lower-cost unit to set the marginal-price. But, a new peaking unit can only reduce prices in hours when the marginal-price is greater than its variable cost. PV, like demand-management, has the potential to reduce prices in all hours that it generates, regardless of the marginal-price. That said, it is most likely that PV could impact the price when demand was on the high side of normal. Generator's bids for the supply of energy are in blocks that can be hundreds of megawatts in size. When demand is lower, larger units (intermediate or even baseload) units set the marginal price. For PV to impact the price in these hours, it would need to add a lot of energy to the system. PV systems are small and these lower demand hours do not generally occur in the middle of the day when solar resource is best. 18 Another important observation regarding the possibility of PV generation impacting normal and high prices is that there are two central reasons that variable costs between generating units vary appreciably. The first is fuel costs and the second is efficiency. Coal plants' variable costs are lower than natural gas and oil plants' variable costs because natural gas and oil are so much more expensive than coal. This discrepancy may worsen if natural gas prices continue to rise. PV's impact on prices would be most notable if PV generation caused natural gas to no longer be the marginal price setting fuel in some hours. As discussed for the case of New England in Chapter IV, this is not likely because of the large amounts of gas-fired generation that operates during the day. Inefficiency impacts prices most during high levels of demand when the oldest, not investment worthy, units must run to maintain system reliability. The generation from these units is usually in smaller blocks and therefore could more easily be offset by PV generation. How couldPV reducepower prices? The addition of PV capacity to a competitive power system could reduce prices in two ways: by reducing the number and severity of price spikes and by causing a lower-cost unit to set the real time marginal price at a given level of demand. These routes through which PV could reduce prices are not unique to PV. Installation of other generating capacity or use of demand side management technologies or methodologies could have the same effect. The question at hand is whether PV, given that it is a non-dispatchable technology, could cause reductions in prices through these mechanisms and, if so, how substantialan effect should be expectedfor a given amountof installedPV capacity. For the purposes of this analysis I separate the above two routes through which PV could reduce wholesale power prices into three: 19 1) The addition of any capacity will reduce price spikes by reducing the frequency and severity of scarcity conditions if the technology can generate during the hours of capacity shortages. 2) The addition of capacity will reduce prices during non-price spike hours if the PV generation causes a lower real time marginal price by altering dispatch. a. When demand is at peak levels (usually corresponding to high prices); b. When demand is not at peak levels (usually corresponding to normal prices). This analysis posits that PV capacity would act like conventional capacity in competitive power systems by interacting with prices through these mechanisms. One way to test this is to compare PV's generation pattern with historical patterns of demand and conventional unit operation, Using historical data is shortsighted by nature because the system has been evolving and will continue to evolve from the state described by these data. But, a baseline understanding of system operation and how PV fits is helpful in itself and carefully specifying the conditions for comparison, like specifically considering investment worthy conventional units, increases the generality of the conclusions of the analysis to include reasonable expectations for the fiture. 20 CHAPTER III - THE NEW ENGLAND POWER POOL The Independent System Operator of New England (ISO-NE) is responsible for operating markets for electric energy and reserve services in the six New England states.9 Wholesale power markets started operating in New England in May of 1999. ISO-NE cooperates with the voluntary association NEPOOL to develop market rules for the wholesale electricity market. As with other power pools and exchanges, the Federal Energy Regulatory Commission (FERC) reviews all rules and tariffs. ° The New England system is a summer peaking system with typical peak hourly demand between 19,000 MW and 23,000 MW in the summer and between 17,000 MW and 19,000 MW in the winter. The highest hourly demand occurred on the system in August 2005, at just over 26,000 MW. Between 1999 and 2004 about 30% of generation in New England came from natural gas-fired generation, 28% from nuclear, 12% from generators burning oil and gas, 11% from coal, and 4% from oil-fired generation, with other fuel sources making up the remainder. Wind, solar, and other "small generation" made up only 0.6% of generation." The Energy Clearing Price (ECP), hourly wholesale price, reported by the ISO" is the time-weighted average of the Real Time Marginal Prices (RTMP) that the ISO calculates about 6 times an hour.3 The lowest-cost MW of generation that would be dispatched to fill an incremental increase in demand or an incremental change in needed operating reserves sets the RTMP. 9 The New England power pool includes Connecticut, Maine, Massachusetts, New Hampshire, Rhode Island, and Vermont. 10ISO-NE, "Overview of ISO New England," at http://www.isone.com/iso news/Information Kit/01 Overview of ISONewEngland.pdf last viewed June 23, 2005 " ISO-NE, "Sources of Energy in New England, 1999-2004," at http://www.isone.com/iso news/Information Kit/08 Energy Sources in New England.pdf last viewed June 23, 2005 12 2 ISO-NE website "Hourly Historical Data" at http ://ww.iso-ne.com/mnarkets/hstdata/index.htnl viewed July 18, 2005 13 ISO-NE website "Five-minute Prices" at http://www.iso-ne.com/markets/Smindata/index.html last and see Bushnell and Saravia 2002 pg. 4 for some discussion. 21 One of the ISO's actions to reduce the abuse of market power is to set prices during periods when demand plus reserve margins near available capacity. Prior to 2001, the ISO determined the price during scarcity periods on a case-by-case'.basis. In 2001, under FERC's encouragement, they implemented a price cap, similar to' markets like PJM and California, of $1000/MWh during scarcity periods when reserve margins cannot be maintained at all or without importing power at $1000/MWh from neighboring regions. In 2000 there were 10 hours at or above $1000/MWh, in 2001 there were 15 hours, 3hours in 2002, and in 2003 and 2004 the price did not reach the cap."4 ISO-NE made a few important changes to market design between 2000 and 2004. A notable one is the adoption of Standard Market Design (SMI)) in March 2003. This made NEPOOL's wholesale markets consistent with the FERC's recommendations intended to increase efficiency and reliability in wholesale power markets nation wide. The ISO implemented locational marginal Pricing (LMP), a pricing system to encourage efficient use of the transmission system when congestion is a problem. Congestion costs are assigned to locations throughout the system via marginal costs calculated at each location. In the following discussion, the electricity clearing prices (ECP) from 2000 through February of 2003 are compared to system load-weighted real time-prices starting in March 2003. The locational price differences in New England were modest in 2003 and 2004 and accounting for them would not affect the results of this analysis.'5 The ISO's resource adequacy standard is the common "one day in ten years" criterion. The ISO designs and operates the bulk power system such the likelihood that a lack of generating resources will cause an outage is no more than one day in ten years on average for non-interruptible customers.' The ISO schedules and operate power plants to ensure energy and reserves to meet this goal. The ISO maintains about 1,700 MW of '4 Calculations from hourly historical price data posted by the ISO at http://www.isone.coin/M arkets/hstdata/index.html 5 ISO-NE 2005 pg. 112 16 ISO-NE, "Overview of ISO New England," at http://www.iso- ne.comn/iso news/Information Kit/()I Overview of ISO New England.pdflast viewed June 23, 2005 22 reserves. This is enough to replace the sudden loss of energy from the largest power source in 10 minutes and the subsequent loss of the second largest generation source within 30 minutes.1 7 In New England, one of the procedures designed to maintain reliability during unexpected conditions is Operating Procedure Number 4 (OP-4), "Actions During a Capacity Deficiency". The procedure.is a number of steps to increase available supply or reduce real-time demand for electricity so that reserve margins can be maintained. If all the steps of the procedure are implemented, some of which include public notice to encourage consumers to conserve power, load relief of between 2,600 and 3,600 MW results.18 The ISO has also implemented small demand-side response programs to help create some interaction between those with interruptible load and the supply side. '9 As a means to encourage investment in generating capacity when and where it is needed, ISO-NE is in the process of implementing an improved capacity market. This improved market is called LICAP, which stands for "Locational Installed Capacity Market". This type of capacity market will reward capacity that is available during scarcity conditions by basing the price in the capacity market on reserve margins.20 The smaller the reserve margins, the higher the price. In this way, the capacity market should encourage the installation of capacity that can generate when and where it is most needed for reliability. This type of capacity market is constructed to bolster reliability and to correct a problem with previous capacity markets: that they did not result in much of a market.' If PV and wind could enter into this type of capacity market, it might help encourage their installation when their production curves best match the demand pattern of the regional electricity system. 17 bid 18 19 Ibid. ISO New England - Demand Response at http:_www.iso-ne.con/enrtioi resrcs/drindex.htnl last viewed September 25, 2005 20 See Cramton and Stoft 2005 for a full explanation of LICAP. 21 Cramton 1999 23 24 CHAPTER IV- METHODOLOGY AND RESULTS PV and Price Spikes To analyze PV in the context of price spikes, I developed an algorithm to compare historical PV generation from varying levels of capacity to historical changes in demand and price during price spikes. The algorithm steps through the historical data hour-byhour for each year. In each hour, it subtracts the hypothetical PV generation from the region's total load. It tests whether the hour's price was a price spike (over a threshold) and then tests whether both the load and price were increasing up to that hour from three hours before and whether the price and load decrease after the hour for two hours ahead. If the price is increasing with load up to that hour or decreasing with load after that hour the algorithm moves on to set the new price in that hour. It does this by assessing whether the new load with PV generation would have either prevented a load increase that occurred in the previous hour (and the associated price increase) or caused the load to decrease sooner as in the next hour. The algorithm compares the load decrease from PV to the entire load decrease in either the previous or next hour. The price is only changed to the lower one of the previous or next hour if the PV generation is enough to lower the load completely below the level of either the previous or next hour's lower price. Partial decreases in load do not change the price. This keeps the algorithm conservative. If a price is changed, in the next hour, the new load and price are used. In hours in which the price is below the threshold, the methodology assumes that the reduction in demand caused by the PV generation does not affect the price. This also means the analysis underestimates PV's potential impact on wholesale prices because decreases in demand in some hours when the price was below the threshold would also have resulted in lower prices. It is not a serious understatement, as discussed below, 25 because PV at the level of about 1000 to 2000 MW would have little effect on normal power prices (e.g. those below $70/MWh in 200222). This methodology creates a "new" set of hourly data for price and load for the power system. Changing the amount of "installed" PV capacity will alter the amount of hypothesized PV generation by scaling the generation of the PV sites accordingly. Because this methodology uses real, historical, PV generation that is contemporaneous with the load and price data, it reflects how well PV generation matched historically with load and high prices in New England. Some of the matching is coincidental and not necessarily indicative of what might happen in the future. But, New England will likely remain a summer-peaking system. It is also likely that when capacity margins again become slim, price spikes will mostly occur in the.summer as they did in 2000 through 2002. In this sense, this method is useful in determining whether PV generation might be expected to impact price spikes at all and, if so, how much PV capacity would have an appreciable effect. Because the analysis uses real PV data, it also takes into account the typical capacity factor for PV systems in New England given the available solar resource. Supportfor the methodology's assumptions The methodology for estimating.PV generation's impact on price spikes assumes that, in hours above a threshold, if demand were reduced in an hour to or below the demand in the previous hour, then the electricity clearing price would not have changed from the previous hour. This assumption holds reasonably well if, in these highest price hours, an increase in demand is the most important driver of the increased price. A number of factors support this assumption for the hours with the highest prices and loads. 22 ISO-NE 2003b pg 40 The ISO considers prices between $20/MWh and $70/MWh (in 2002) as "normal". 26 First, capacity shortages cause high prices. Because demand is inelastic and supply is fairly consistent from hour-to-hour (barring the relatively rare unexpected generator outage), high prices are well correlated with high demand. The hours with the highest levels of demand are those in which demand intersects the steepest part of the supply curve or those when demand exceeds supply. A small change in demand in these hours can drastically affect the price and variables like the spot price of fuel that day have a smaller impact in comparison. Some of the ISO's actions stress the importance of the level of demand in these situations. For example, the voluntary and emergency demand-side response programs are designed to alleviate demand during these conditions. About 350 MW participated in ISO-NE's demand response programs in 2004.23Additionally, the OP-4 procedures are a measure designed to allow the ISO to decrease demand quickly, typically by about 1000 MW 2 4 but possibly up to 3,600 MW25, when capacity margins are thin. The high prices are a result of scarcity conditions that threaten the reliability of the power system and the ISO would presumably not put so much effort into being able to reduce demand or increase supply in these hours if it would not help. Cyclesof price spikes and investment:observationsof NEPOOL Three types of evidence from the data for NEPOOL between 2000 and 2004 support the applicability of the key relationships between price and capacity. The first is trends in price statistics between years as new capacity was added and demand statistics remained fairly constant. Second is the number of times, annually, the price exceeded the variable costs of investment worthy peaking units. The third metric is the number days on which a scarcity event occurred. 23 ISO-NE 2005 pg 92 24 Analysis 25 ISO-NE, of historical changes in demand before OP-4 procedures were implemented. "Overview of ISO New England," at http://www.iso- ne.coniso. news/Information KitO I Overview of ISO New. Enland.pdf last viewed June 23, 2005 27 Demand, net revenues, and capacity in New England As Table 1 shows, peak demand increased slightly from 2000 to 2002 and then decreased in 2003 and 2004. Average demand remained fairly constant during the five-year period. shows box plots of the demand in New England for the five years. The variation Figure in the highest demand levels is driven mostly by hot weather in the summer. Heat waves and cold snaps with fairly unpredictable severities are one of the major drivers of price. spikes along with, but less frequently, unscheduled power plant outages and transmission and distribution system contingencies. 25000 o-%20000 A x 15000 C i 10000 E aQ WO 5000 2000 2001 2002 2003 2004 Figure 1 Box plots depict the distribution of hourly demands over the five years. Box plots show mean (cross), median (central line), 25 th and 7 5 th percentiles, upper and lower fences, far out values, and maximum demands (triangles). Most statistics are stable over the five-year period. Table I Demand statistics for New England between 2000 and 2004 Average Minimum Maximum 1st Quartile Median 3rd Quartile 2000 2001 14182 14376 8528 8772 21916 24959 11980 12069 14600 14643 16007 16316 2002 2003 2004 14546 14754 14881 8747 8820 9020 25344 24330 23750 12128 12537 12635 14828 15016 15175 16410 16696 16830 MW MW MW MW MW MW 28 Net revenuesand capacity ISO-NE's annual State of the Market Reports (SOM) and other analyses26 found, in each year, that the market would not have covered the fixed costs and competitive return on investment of new gas-fired combined cycle or combustion turbine units in New England. In 2002, for example, they found that a representative natural gas-fired combustion turbine unit with heat rate of 10,500 Btu/kWh would have net revenue from the energy market of $25,000/MW, revenue from the capacity market of about $10,000/MW and ancillary services revenue of about $5,000/MW. The total revenue is thus about $40,000 but annual fixed costs for a new combustion turbine were between $60,000 to $80,000/MW at that time. 27 This analysis of net revenues suggests that no new capacity should have been installed in New England between 2000 and 2004. But, between 2000 and 2003, generating companies brought 17 new gas-fired or gas/oil capable combined-cycle units and 12 new gas-fired or gas/oil capable combustion turbines (CT) on-line. These additions represent about 7000 MW, which is roughly the amount total capacity increased in New England between 2000 and 2003 (there were other additions and units that shut down or were de-rated). Total capacity and combined-cycle and combustion turbine capacity remained about the same between 2003 and 2004. Capacity increased from about 24,200 MW28 in 2000 to 31,100 MW in 2003 and then slightly to about 31,300 MW in 2004. Table 229shows the changes in capacity by generator primary fuel and technology between 2000 and 2004. Figure 2 shows the trends in capacity additions by fuel and technology type compared to the installed capacity in 26 ISO-NE 2003a and 2005 and Joskow 2003 pg. 57-68. In each year, the ISO considered the relevant ECP or LMPs and found that the costs of a new, representative, gas-fired combined cycle or gas-fired combustion turbine would not have been covered in that year. 27 ISO-NE 2003a pg 44 28 Summer-rated without net purchases and sales and non-participant capacity. 29 Data derived from New England's "Capacity, Energy, Load, Transmission" (CELT) and "Seasonal Claimed Capability" (SCC) reports available at http://www.iso-ne.con/trans/celtreport/and http://www.iso-ne.cornm/genrtionresrcssnl .clmd. cap/index.html respectively, downloaded July 7, 2005 29 2000. This figure depicts the rise in gas-fired combustion turbine and combined cycle capacity. Table 2 Capacity additions in New England from 2000 to 2004 (internal resources only and excluding net interchange).; (MW) Capacity 2000 Capacity 2001 Capacity 2002 Capacity 2003 Capacity 2004 4359 3089 2814 4466 3479 4344 3053 2805 3657 3947 4340 3064 2803 3535 3963 4368 3437 2786 3301 3533 4347 3386 2779 3077 3570 10758 10410 10301 9620 9427 1412 3171 5 3457 2570 216 4536 4473 460 6455 4650 541 5873 5647 520 -299 Nuclear ydro Coal Steam Oil Steam Gas/Oil Steam, Sub Total Gas Combined Cycle Gas/Oil Combined Cycle Gas Combustion Turbine Gas/Oil Combustion Turbine 305 296 297 299 Turbine 872 822 851 668 680 Sub Total 5766 7361 10616 12613 13019 Combustion Combustion Refuse Wood Steam 123 28 0 105 107 26 0 153 101 26 0 139 100 40 576 420 101 . 25 583 373 Wind 0.5 0.5 0.7 0.7 0.7 Other 0 0 0 7 38 24228 25455 28588 31181 31299 Oil Combustion Oil Internal Gas/Oil Internal TOTAL 1%nn 'MW .- .... 1. - v'(GCT) . .......... . Us 0o 200. ,,f 8 0 Cu~~~~~ 100 In And . . (Total) baead (OCT) 0 LL. tiv I---- 2000 ... . - - 2001 . . , __ 2002 -1 2003 2004 Figure 2 Trends in capacity by technology and fuel type compared to levels of capacity in 2000. GCT=Gas Combustion Turbine, CC=Combined Cycle, OCT=Oil Combustion Turbine. 30 Price statistics Given the consistent demand, annual price statistics responded as expected to the addition of capacity in New England between 2000 and 2004. Three types of evidence point to this conclusion. First, fuel-cost-adjusted average prices decreased in 2003 and 2004 with the addition of capacity (and the less extreme weather). Second, median prices remained roughly constant and average prices 'conditional on the demand being in higher quartiles decreased more than average prices when demand was in the lower quartiles. Third, volatility decreased. Table 3 reports unadjusted price statistics.30 Table 4 shows price statistics adjusted for fuel prices.3 1' 32 Comparisons of Table 4 columns one to two and three to four show that the increase in unweighted and load-weighted average prices between 2000 and 2004 is largely due to the increase in fuel costs between the years. The ISO noted this in their 2004 State of the Markets (SOM) report. As expected with the addition of new, more efficient, generators in 2002 through 2004, the fuel-cost-adjusted average prices decreased in 2003 and 2004. The improved efficiency of marginal-price setting units contributed to these decreases.3 3 The relatively small changes in median prices between years and the changes in standard deviation, suggest that the reduction in volatility in later years occurred because 30 The ISO reports these in their State of the Markets (SOM) report and uses them to analyze price volatility (ISO-NE 2005). The numbers I report differ slightly from those the ISO reports. I calculated the statistics from the hourly numbers posted on the ISO's website and do not know why the statistics differ. They do not differ enough to change any conclusions in either study. In their 2004 State of the Markets Report, the ISO reports unadjusted price statistics and those adjusted for changes in natural gas and oil prices. They calculate these by altering the five-minute real time prices 3' according to which fuel was marginal and the spot price for that fuel compared to its price in 2000. The prices in Table 4 were adjusted by factors reported in the 2004 SOM (ISO-NE 2005). In that report the ISO normalized the bids of marginal-price setting units for each 5-minute real time price between 2000 and 2004. The annual scale factors that resulted from the SOM analysis are used here. These tables and subsequent charts and tables report the Energy Clearing Price (ECP) from 2000 through February 2003 and the Real Time Locational Marginal Price at the Hub (RT-LMP) after March 2003. In March 2003, New England implemented locational marginal pricing (LMP) and, because locational differences in price during this period were moderate, the Hub RT-LMP is comparable to the ECP before March 2003. ISO-NE 2005 pg. 112 notes that accounting for locational differences would not appreciably change the results of this types of analysis. 32 33 ISO-NE 2005 pg. 31-34 31 of the reduction of high prices to lower prices that Were still above the median. The addition of combustion turbine capacity would be expected to affect prices when demand and prices were high enough that these new units operated. Figure 3 shows the relative changes in prices compared to 2000 conditional on the demand being within various percentiles. The graph shows the decrease in fuel-adjusted mean ECP between 2001 and 2004. It also shows that there was a larger relative decrease in mean ECP given that demand was above the 75th or 9 0 th percentile than in the unconditional mean ECP or the ECP given that demand was below the 2 5 th percentile. The new generating units affected prices more when demand was high because that is when they operated. Table 3 Price statistics for New England between 2000 and 2002 (unadjusted) Median Inter 2000 2001 2002 2003 ^ 2004A Avg 43 40 36 51 52 Stdev 132 45 24 23 23 Max 6000 1000 1000 998 920 quartile Median range Min 39 21 -3 19 35 -6 1.6 33 1 17 48 0 50 17 0 $/MWh absolute deviation 16 14 12 13 ..12 - - - Table 4 Price statistics adjusted for fuel prices and load Adj.* Avg 43 Load- Load- Weight Avg** 46 Weight Adj. Avg 46 SOM- Stdev 30 Adj. Stdev 30 based weights 1.00 2000" Avg 43 2001 40 45 43 49 45 50 0.89 2002 2003 2004 36 51 52 44 42. 42 38 53 47 44 24 23 23 31 19 19 0.80 1.23 1.25 ^ ^ 54 43 I/MWh *Adjusted for changes in natural gas and oil prices in proportion to the fraction of hours each fuel was marginal ** The load-weighted average of electricity clearing prices ^ The real-time locational marginal prices for the New England hub are used after the implementation of LMP in 2003 "Prices over $1000/MWh were dropped to $1000/MWh 32 a 1. i 1.2 8o~ 0 .- ECP IDem % 1 He~~~~~~~~~~ . .......................-.- :D .8.. ~- . ................ _____ -- .- 0.4 2000 Mean ....... ECP Mean ECP I Demanc>75th % C' 0.6 .-- 0 ~~Mean 1-5 2001 2002 23 Mean ECP I Demand>90th % 2004 Figure 3 Comparison of mean energy clearing prices between years conditional on the demand being above a given percentile. The volatility, as measured by unadjusted standard deviation in column two of Table 3, was $132/MWh in 2000 and by 2004 was about $23/MWh. In the year 2000, the ISO still set the price cap during price spikes rather than capping all spikes at $1000/MWh. Adjusting the spikes above $1000/MWh (one reached $6000/MWh) in 2000 to $1000/MWh makes the statistics more comparable between years. When this and adjustment for fuel prices is done as in Table 4, column six, the standard deviation in 2001 was about $50/MWh and it fell to about $30/MWh in 2002 and down to around $19/MWh in 2003 and 2004. This is consistent with the trends in added capacity. Number ofprices above peaking-unit costs Another way to observe the changes in price spikes is to observe the number of prices over a certain threshold each year. Setting the threshold above the variable cost of the highest cost investment worthy peaker in the system in the earliest year and adjusting the threshold for natural gas prices gives a comparable measure of how often price spikes 33 occur in each year. Figure.4.shows the relationship between capacity margins34 and number of prices over fuel-price adjusted thresholds of $70/MWh and $100/MWh in 2000. Capacity margins are measured as the difference between capacity and peak demand in a year as a fraction of peak demand. As expected, the year with tightest. capacity margins, 2001, was the year with the most price spikes above the threshold. 700. X # of Hours> $70MWh 2001 600. S W 500 2003 2002 X 0 X X 'r 400. 2000 2 300. It 20015 E # of Hours > $100/MWh Z 200 2001 10001 0 2003 U ,21)02 82004 2000 f - 0.1 0.2 - 0.3 . o0.4 Capacity Margins (Cap - Peak)/(Peak) Figure 4 Number of hours above fuel-price adjusted thresholds of $70/MWh and $100/MWh in 2000 versus capacity margins in New England. Capacity margins are measured as the difference between capacity and peak demand in a year divided by peak demand. 34 The capacity numbers used in this analysis indicate the amount of internal resources, in summer-rated capacity. New England is a net importer of power from the surrounding areas of New York and Canada. Although total net imports into New England changed during the period, the maximum amount imported in a single hour did not change. This amount remained close to 2600 MW for all years. Because imports can only alleviate capacity shortages relative to the amount that can be imported in a given hour, and that number was constant for all years in consideration, imports do not appreciably affect the results of this analysis. 34 Numberof scarcityevents As capacity increased, the number of days with scarcity-events decreased, as measured by implementation of OP-4 procedures. The system operator implemented OP-4 actions on eleven days in 1999, six days in each year between 2000 and 2002 and only on two days in 2003 and three in 2004. On 3 of the 5 days that the system operator implemented OP-4 in 2003 and 2004, scarcity conditions occurred in the winter as opposed to three of the 29 days in 1999 through 2002. The 2004 SOM suggests that the OP-4 conditions that occurred during these winters did so because of unexpected cold snaps when capacity was off-line for scheduled maintenance. The scarcity conditions were not due to capacity shortages that reflected overall levels of capacity like those in summertime hours when nearly all units are available. The conditions were not severe and prices did not rise to the $1000/MWh cap as in the summers of 2000 through 2002. PV analysis:empiricalresults WouldPV impactprice spikes like the installationof conventionalcapacity? The first goal of this analysis was to determine if PV could impact price spikes in a manner similar to conventional capacity. If, for example, PV never generated or only generated small amounts during hours that price spikes occurred, then it would have no or limited impact on the occurrence of price spikes. If this were true, installing PV could not impact investment cycles as they now occur. The results of the price spike analysis for 1000 MW of PV with a threshold of $70/MWh in 2000, adjusted for fuel prices for later years, show that PV could have historically lowered demand in price spike hours. It is worth stressing that this analysis shows that, historically, 1000 MW of PV in New 35 England could have generated enough.to stop the increases in demand that occurred ... between some of the highest priced hours from 2000 and 2002. Table 5 Number of prices above fuel-adjusted thresholds with and without PV for all year and just the summer. PV data were only available for 2000 through 2002. . All year_ _- ___ # Prices above* # Prices above* # Prices above* 200 /MW 150 $/MW 100 $/MW 2000 Actual 13 .2001 40 2002 2003 2004: With PV Actual 5 23 With PV Actual 18 66 70$'MW With PV Actual 59 430 With PV 357 650 .552 472 51 20 77 6 33 13 453: t10 113 4 274 62 _ 28 13 _ *Thresholds are adjusted or fuel prices, numbers are for 2000. 353 28 62 51 143 106 Summer only (May 1st through September 30th) # Prices above # Prices above # Prices above 200 /MW 2000 2001 2002 2003 2004 # Prices above* _ Actual 10 21 12 1 0 150 $/MW With PV Actual 15 2 26 10 5 30 2 0 i $100/MW With PV! Actual 34 11 66 17 17 63 14 3 , # Prices above 70 $1MW With PV Actual 30 103 40 195 37 248 41 42 With PV 65 141 169 Table 5 shows the number of hours with prices over varying thresholds for 2000 through 2002 with and without the added PV capacity. The thresholds.are reported for 2000, all other thresholds are adjusted for fuel prices. That is, if the threshold is $100/MWh this means $100/MWh in 2000 and the other years are adjusted by the weights shown in Table 4. As expected, based on the rough correlation between PV's production curve and electricity demand, the PV reduces demand enough to prevent increases in a higher percentage of the hours over $200/MWh. Table 6 shows the percentage change in number of prices above the thresholds with the addition of PV. Most of the hours above $200/MWh occurred in the summer and in the middles of the day between 2000 and 2003. 36 Table 6 Percent reduction in number of hours over thresholds caused by the addition of PV for all year and the summer All year Threshold 150 $/MWh 100 $/MWh 200 $/MW 2000 2001 2002 70 $/MWh 17 15 25 11 26 34 22 18 39 62 30 54 % Reduction in # of hours over fuel-price adjusted threhold Summer only (May 1st through September 30th) 200 $/MW. 150 $/MWh 100 $/MWh Threshold 2000 2001 2002 70 $/MWh 37 28 32 12 39 41 27 35 43 80 52 58 % Reduction in # of hours over fuel-price adjusted threhold Table 7 shows price and load statistics for the years with and without PV. The median prices change little because the algorithm affects only prices originally above the median (above the threshold) and reduces some of those prices to new prices that are still above the median. Table 7 Price statistics, adjusted for fuel prices, with and without PV Adj. Average 2000 41 45 2001 44 2002 42 2003 2004 42 PV Adj. Average 40 44 44 Adj. - Median 39 40 41 39 PV Adj. Median 39 *40 41 Adj. Stdev 40 30 50 31 19 19 PV Adj. Stdev 19 38 - 21 $/MWh PV resultsin context of conventionalcapacity Comparing the results of the above analysis to the variation in capacity levels and frequency of high prices in the years 2000 through 2004 is a way to determine when the impact of PV might be considered appreciable. A PV-induced change in number of high prices or in price statistics that is similar in magnitude to that caused by historical 37 fluctuations in conventional capacity between years should be considered significant enough to affect the investment economics of the market. The historical fluctuations in years are also due to weather and other variations in conditions. These are iot taken into. account in an analysis that estimates how much an historical year would have differed with the installation of PV.,Nevertheless, changes from year-to-year do give a benchmark .. for order of magnitude of change thatis significant.' The changes between the years 2001 and 2002, which included the addition of . capacity as well as variations in weather and other factors resulted in a86 percent reduction i the number of prices over $200/MW, about 45 percent reduction in prices.,. above $150/MW and $1.00/MW, and about a 25 percent decrease in those above $70/MW. 3 5 These numbers are similar in magnitude to those from the.PV analysis shown in Table 6, Although the numbers in Table 6 are not directly comparable to the changes between years that actually occurred as weather and capacity changed, the order of magnitude is roughly comparable and roughly the same. . . ... ... How much PVwould make an impact? . The second goal of this analysis is to determine the level of PV capacity that would start to have an appreciable effect on high prices. Table 8 shows sensitivity analysis for the amount of hypothesized installed PV. About 1000 MW of PV in each year impacts prices similarly to the fluctuations between years that occurred historically. 1000 MW of PV has considerably more impact than 500 MW in most years and 2000 MW has a similar impact to 1000 MW. This is because the typical increase in load between hours is around 800 MW so generation from 1000 MW of PV is just enough to prevent many of these increases. 35 As usual, these are adjusted thresholds so $1 00/MWh is $1 00/MWh in 2000 or $89/MWh in 2002. 38 Installing 1000 MW of PV would represent about three percent of the total installed capacity in New England. In Germany where aggressive subsidy programs have encouraged the entry of PV, installed PV capacity (grid-connected) is about 800 MW36 and this is about 0.7% of the total installed capacity of about 120 GW37 . Notably, in 2004, Germany installed about 370 MW of PV systems. It is not out of the question that areas like New England might, at some point for some reason, install hundreds of MW of PV per year. Were this to happen, even a cursory understanding of the expected affect of this new capacity on power system dynamics would be useful. This analysis shows that the installation of large amounts of PV could affect the dynamics of price spikes and investment in a regional power system. Table 8 Number of hours over prices of $200/MWh, $100/MWh, and $70/MWh adjusted for fuel prices for varying levels of PV capacity. Percent differences from actual number of prices over these levels are also shown. MW PV 2000 1000 Results 2000 Results for Results for Results for 2002 f Hours above $200/MWhII 5 62% 5 62% 11 11 Actual 13 2000 1000 26 28 35% 30% 94 106 34% 26% 507 .553 22% 15% 500 50 34 35 15% 13% 123 125 14% 13% 569 591 12% 9% Actual 40 2000 1000 2 6 85% 54% 33 51 57% 34% 304 353 36% 25% 9 31% 65 16% 401 15% 13 0% 67 13% 420 11% 50 Actual 62 63 70/MWh 333 23% 357 18% Sor 500 50 500 15% 15% $100/MWh 43 35% 59 11% 6% 5% 66 77 14% 7% 433 143 13 374 401 650 472 Number of hours above threshold Percent change from actual 36 Solarbuzz, "Fast German Solar Energy, Power Industry and Market Facts," at Ittp://www .solarbuzz.corn F astFactsGernany.h 37 Energy tmn,last viewed September 22, 2005. Information Agency, "International Electricity Information: International Electricity Installed Capacity Data," at http ://www.eia.doe. ov/pub/internationaliealfftalle64.xls, last viewed September 22, 2005 39 Conclusions . The conclusions of this analysis are simple. Installing a large amount.of PV capacity would have airoughly similar effect on price spikes as installing more conventional capacity. This is to say, installing PV would not result in a special type of long-term reduction of price volatility or of price spikes. If PV were added in a time when capacity margins were slim, then PV would likely reduce the frequency of price spikes. If demand increased or capacity decreased from that point, price spikes would reoccur as part of the natural cycle of investment economics in the power system. What this analysis does show is that PV would impact investment economics; the timing of solar.resource in New . England is compatible to that of the historically typical occurrence of high prices due.to slim capacity margins. Although this result was expected, it is not entirely trivial because it would be possible for a non-dispatchable technology to be poorly matched. contemporaneously with the occurrence of most price spikes in a regionalpower system. This simple finding results in another question. If the installation of PV would reduce price spikes, which are the market's signals that more capacity is needed to maintain reliability, would PV also fill the role of providing enough generation at needed times to result in the requisite level of reliability? Most of the new capacity that has been installed in New England since 2000 was natural gas-fired combined cycle and combustion turbine generating units. Given the typical operation of the New England power system, if the installation of PV were to fill the role of the installation of more conventional capacity in improving capacity margins during critical periods, then the PV should be able to replace (or at least nearly replace) the typical operation of the new conventional unit that would otherwise be installed in its place. If PV cannot replace the typical operation of the unit, a problem arises' PV could change the market signals such that no new conventional capacity was installed but then -not be able to generate when that new capacity would have. This could compromise 40 reliability unless the system was able to compensate. If this resulted in price spikes occurring in hours that PV did not generate and thus encouraged the installation of more conventional units, then the total cost of providing power would increase. The new conventional generation would be needed in fewer hours, so in order to cover long-term costs prices would need to be higher or the units would require compensation for their availability. Also, the addition of PV would not accomplish the environmental benefit that one might hope because the new conventional unit would still be installed. CouldPV replace combustionturbineunits in New England? The analysis that follows compares PV generation to that of the typical operation of natural gas-fired combustion turbine generators in New England. The two goals of the analysis that follows are (1) to understand whether added PV capacity could replace the need for new combustion turbine units and (2) to characterize the likely impact of PV on "normal" prices (i.e. those below $70/MWh3 8). Why comparePV to combustionturbine units? Most new capacity in the last five years in New England has been natural gas fired combustion turbine and combined cycle units. Comparing PV to the type of capacity that will likely be installed in the near future can help determine whether PV could replace the need for more conventional capacity. To perform this analysis, it is best to compare PV to a type of generator that is investment worthy and a type that PV could most easily replace. If PV could not replace the need to install more of this type of generator, it could not replace one that operated in manner poorly matched to PV. 38 ISO-NE 2003b pg 40 The ISO considers prices between $20/MWh and $70/MWh (in 2002) as "normal". 41 A difference between PV systems and peaking units is why they generate when they do. Peaking units generate whenever the revenues from doing so, from energy and. other markets, are larger than the costs. This generally means that these units operate when prices are high because peaking units have low fixed costs and high variable costs. A well functioning PV system, however, generates whenever the sun is shining. This usually means that PV systems generate when prices are high. Combined cycle units and combustion turbine cogeneration units usually generate upwards of 90% of annual hours, they are not peaking technologies. In New England, combustion turbine (CT) units operate in a number of different ways. As Figure 5 shows, those with the highest heat rates (lowest efficiencies) operate around 1% or less of annual hours. In the graph, these make up the cluster of points near the y-axis representing units with heat rates above 18,000 BTU/kWh. These units are located within large coal and oil plants and are most likely operated to aid start-up. Other CT units operate for a high percentage of annual hours because they are used for cogeneration of heat and power. CT units that operate around 10% of annual hours, like those shown in Figure 5 are typical of new CT units installed in New England and were therefore chosen for comparison to PV. The units chosen for comparison to PV generation operated for about 10 to 20% of annual hours each year. Figure 6 shows the consistency in operation between years of combustion turbine units with this mode of operation. It displays the distribution of operating hours over the hours of day for the five Devon CT units (170 MW total capacity) in Connecticut from 2000 to 2003. Figure 6 also shows the same distribution for the Wallingford combustion turbine units that started operation in 2002. Both in the summer and during the entire year, when these units generate they do so mostly between 5 am and 10 pm. The Devon units operated for fewer hours in 2000 and 2003 than the other years because capacity margins were higher in New England. The clearing price less frequently reached the level 42 of the units' variable costs where it was economic for them to generate during these years. Annual Average Heat Rate versus the Fraction of Possible Hours unit operated during 2002 (New England) + Combined Cycles x Combustion Turbines I. ... .. 4) .0 d: ~ + ~~~~~~~+ ,t~~~* + +~'~ 12000 I . (I : 0 ++ + 6000- 4.. 00 * 4' '. 0.2 04 0.6 0.8 Fraction of Possible Hours 1 Figure 5 Combustion turbine units with heat rates around 1 000 BTU/kWh, used as peaking units, operated around 10% of annual hours in 2002. Those with higher heat rates that operated far fewer hours are combustion turbine units located within large oil and coal power plants. These are most likely just used for start-up and other support situations. A number of combustion turbine cogeneration units operated for a larger fraction of the year. Distribution.of PV and Peaker Operating Hours (New England) D2 L1.5[ Annual - i ~~~~~~~I 8't-. ·-.. ....... 0.5 0 4 2 8 Hour of Day N¥? , 16 2 Hour of Day Figure 6 The pattern of peaker operation over hours of day is consistent over years although the units operated less during 2000 and 2003 when fuel-adjusted prices were lower on average and less volatile and capacity margins were larger. 43 The most straightforward comparison between PV generation and combustion turbine operation is the hours of the day in.which each typically generates. Availability of sunlight constrains generation from PV and whether the clearing price at or above the variable costs of a combustion turbine constrains its generation. Figure 7 shows the distributions of PV generation and.CT generation over the hours of the day. What these graphs clearly show is that, especially in the winter, the CT units :operate during evening hours when PV cannot. Also, there are not typically hours that PV generates that the CT units do not, at least on days when the CT units operate. Distribution of PV and Peaker Operating Hours (New England) Devon2001 - I0 - :3 Devon 2002 Devon 2003 ........... WaRlingford 2002 C &: Hour of Day 16 12 Hour of Day 20 Figure 7 PV cannot generation during evening hours when combustion turbines generally do. PV generates during more hours annually (4000 compared to about 1000) than do combustion turbines that are operated as peaking units. The next step in comparing the relative effects of PV generation and CT generation on wholesale prices is to consider the hours in which both PV and CT generate. During these hours, how much PV capacity would be needed to replace the generation from a C[ unit? For this analysis, the summed generation from the five Devon 44 CT units (170 MW total capacity) is compared to the generation from varying levels of PV capacity. In the hours that both the P V system and the CT units generated, the PV generation from 1000 MW capacity was larger in about 70% of hours. Results for other levels of PV capacity between 200 and 2000 MW are also shown in Table 9. Table 9 Fraction of hours that PV(MW)-CT(MW)>0 given that PV and CT units generated (PV Capacity) 200 MW 300 MW 500 MW 1000 MW 2000 MW 2000 0.32 0.45 0.59 0.71 0.77 2001 0.30 0.47 0.61 0.73 0.80 2002 0.25 0.46 0.61 0.73 0.80 Fraction of total hours when PV and CT units generated that PV generation larger These results show that even with 2000 MW of PV capacity, PV generation would only be larger than generation from 170 MW of CT capacity during 80% of the hours that both PV and the CT units were operating. This reflects the early morning and late evening hours when solar resource is not strong as well as those hours during the middle of the day when it was cloudy but still hot enough to drive the demand for CT generation. Table 10 shows the fraction of hours that PV generation was larger during all the hours that the CT units generated regardless of whether or not the PV systems were generating. The CT units' generation was often more than PV generation in hours after 5 p.m. on the days the CT units operated. PV generation from 1000 MW was larger than the CT units' generation in about 50% of the hours that the CT units operated. If only the hours between 7 am and 4 pm are considered, the numbers in Table 10 increase by about 30% for 500 MW of PV and above. That is, during the middle of the day, 1000 MW of PV generated more than the CT units in about 83% of hours in 2001 and 2002 and 2000 MW of PV generated more in just over 90% of these hours. 45 Table 10 Fraction of hours that PV(MW)-CT(MW)>0 given that the CT units generated 1000 MW 2000 MW 500 MW 200 MW 300 MW (PV Capacity) 0.41 0.44 0.18 0.26 0.34 2000 0.54 0.32 0.41 0.49 0.20 2001 0.57 0.52 0.33 0.43 0.18 2002 Fraction of total hours when CT units generated that PV generation larger On many days, the CT units did not generate but PV did: generation from 1000 MW of PV was greater than peaker generation in about 85% of the hours PV generated. Table 11 shows the fraction of hours when the PV units were generating that PV generation was larger than that of the CT units, including all the hours that the CT units did not generate at all. Table 11 Fraction of hours that PV(MW)-CT(MW)>0 given that the PV units generated (PV Capacity) 2000 200 MW 0.89 300 MW 0.90 500 MW 0.91 1000 MW 0.92 2000 MW 0.92 2001 0.80 0.83 0.85 0.87 0.88 ~~~~~~. . . 0.84 0.87 0.88 2002 0.77 0.81 Fraction of total hours when PV generated that PV generation larger The differences in Table 9 through Table 11 support the conclusion that Figure 7 suggests: PV generation is not perfectly matched to the way combustion turbines are operated in New England. Although diurnal solar resource does match diurnal electricity. demand patterns generally, the match is not close enough to warrant thinking of PV as a replacement for combustion turbines. Doing so would be misleading from a system operations standpoint and would undervalue PV because it generates during many days when combustion turbines do not.3 9 . . 39 This result may vary in other geographic regions depending on patterns of demand and prices due to the type of installed capacity and on patterns of solar resource. 46 PV's impact on price in non-price-spikehours The operation of peaking units diurnally and seasonally is consistent over time as is the pattern of solar resource in each region. Comparing the scaled PV generation to peaking units in the region that often set the marginal clearing price during hours with average demand provides a benchmark for understanding how PV capacity might affect prices. The logical argument based on the supply curve is that decreases in demand, when prices are in the normal range, do not affect prices very much because the supply curve is not very steep. In the technical supplement to the 2002 SOM, the ISO found that natural gas prices explain most of the variation in prices when they are in the normal range between $20/MWh and $70/MWh.4 ° In order to affect prices at all in this range, PV would have to cause the marginal unit to turn off so that a new marginal price would be set in that hour. This is difficult for PV not only it must generate enough to cause this unit to turn off but, in reality, many units are changing output in response to changes in load. This occurs because of complexities like transmission system constraints, the reserve markets, and unit-commitment problems. Figure 8 depicts this complication: in some hours peaking units turn on while baseload and intermediate units generate at less than full capacity. In the case that PV generation actually did cause the most costly unit to turn off in an hour, then another natural gas plant would probably become the marginal price-setting unit (i.e. PV would cause a natural gas unit to turn off and it would then become the marginal price setting unit). In other words, PV would not cause a change in the marginal price settingfuel and thus would not drastically impact prices. This is especially true during periods of the highest normal prices, when one concerned with high prices would 40 ISO-NE 2003b pg.40. In this range, ISO-NE models average peak electricity clearing prices (averages of ECPs in hours ending in 12 through 19) as a function of fuel costs weighted by the price-setting units' efficiencies, net capacity, energy exchanges with other regions, amount of self-scheduled energy as a fraction of load, and changes in market rules. They find that efficiency-weighted fuel cost explains about 60 percent of the variation in average peak clearing prices. In this range of normal prices, it is difficult to pinpoint the effect of a reduction in demand of say, a 5% reduction in demand because of added PV generation. The change in demand would change the price-setting unit and the self-scheduled energy and perhaps some of the other variables that determine ECP as well. 47 most want PV to reduce them, because more natural gas and oil plants operate at these times. Also, it is more difficult for PV to impact the most expensive,unit that is operating during hours of lower demand because these units have larger capacities and other . barriers to turning off like start-up and shutdown costs. When the amount of PV. or demand side management and.other renewables is enough to change the frequency with which natural gas sets the real time marginal price, then these technologies will have a . more appreciable effect on prices.. Time Series Generation for Five Units and Total Load in Juty 2002 New Enaland; C 2 2( :3 I0 t 2412 -r'8 1-Jul 4-Jul 7-Jul e, 10-Jul 13-Jul , ..... , 16&Jut 19-Jul Date (July, 2002) Figure 8 Hourly generation from five units, two coal-fired steam turbines (ST-Coal), one natural gas-fired combined cycle (CC-NG), and two natural gas-fired combustion turbines (CT-NG). The figure notes the units' states, and capacities. The bottom graph shows New England hourly load. 48 Wholesale prices in hours when both PV and the CT units generated were mostly on the high side of the overall distribution of prices. This is expected because CT units operate when the clearing price is high enough to cover their variable costs (around $60/MWh in 2002 because of high natural Distributions of CT and PV operating hours over ECP {(NewEngland) gas prices). Figure 9 shows the distribution of electricity clearing prices in 2002 as well as the distributions when the PV systems and CT units generated and when the PV generation was larger than that of the five ~ O I 6. C 4 winter of 2002. This was probably due to fuel i PV i Devon -- Wallingford a TC X, ... i, l-- 0. . '--0 40 80 120 Devon CT units. The Devon units operated some at a lower clearing price during the 8A ... i d'Ail =fe 160 200 ' -t[ Lo 0 I - PV Greater C switching or possibly because the units were needed when other units were down for . maintenance or for transmission-related ref . e JgjAA.2. D reasons in which case the electricity-clearing price would not have reflected the fact that 0 I C 4: they were operating.4 1 15 it (1 Q. 4 0 ECP ($/MWh) Figure 9 Distribution of prices when PV and CT units were operating and when PV generation was larger than CT generation. The bottom graph shows the distribution of prices in 2002. 4 In New England in 2002, before the implementation of SMD, when units operated out of merit order for reasons related to transmission constraints, they were paid the difference between their bid and the clearing price in an "uplift" payment. 49 50 CHAPTER V - INCENTIVES HINDERING PV ADOPTION The physical and economic operational intricacies discussed above make it challenging, though not impossible, for a power system to adopt large amounts of PV or other new technologies. Other related and important challenges stem from incentives that incumbent power producers have to resist the widespread adoption of these new technologies. This analysis highlighted two examples of this type of incentive. The first is whether PV would, by reducing high prices, reduce the revenues of baseload generation owners significantly more than does the installation of additional conventional capacity - or if baseload owners might perceive this to be a problem. The second is that current and future owners of conventional peaking generators may fear that the mismatch between PV and their typical operation would leave their units necessary for reliability but lacking the revenues needed to cover fixed costs. Impact ofPV on baseloadunit revenues One of the conclusions of the preceding analysis was that PV's impact on price spikes would be similar in magnitude to that of the installation of more conventional capacity. It then follows that the impact of PV on the revenues of baseload plants would also be similar. Logically, this means that the baseload plant owners should not have a stronger incentive to resist the adoption of PV than they do to resist the installation of more conventional capacity. There are two reasons this may not be the case. The first is that PV does not use natural gas as a fuel and the second is that incumbents may perceive the change toward large PV penetration as an unnecessary threat while they perceive the addition of conventional capacity as "normal", or under at least their control to some extent. 51 The price differential between baseload fuels and natural gas continues to increase as natural gas prices rise and the other prices remain the nearly the same. If the addition of PV capacity reduced the frequency that natural gas-fired units set the marginal price, revenues of baseload plants would be affected differently than if new natural gas-fired' units were installed. PV is unlikely to change the frequency that natural gas is the .. marginal fuel, however, because natural gas currently set the marginal price in about 85% of annual hours. With the exception of late evening and early morning hours, there are ... many natural gas units generating; turning off one will not usually cause a lower-priced fuel like coal to be marginal. As found earlier, see Table 9, 1000 MW of PV generated more than 170 MW of CT capacity in only 80% of annual hours that both generated. Two hundred megawatts of PV generated more in only 30% of hours that both the CT and PV generated. It would be difficult for PV to replace even 170 MW of natural-gas-fired generation and there is over 10,000 MW of natural gas-fired capacity installed in New England. It is highly unlikely that enough PV could be installed to offset the generation from this capacity. It would even be difficult for PV to offset the 500 MW of installed natural gas-fired CT capacity if it were all generating in an hour. This does not mean, however, that the impact from PV on baseload revenues would be insignificant to the owners of those generators. In fact, there is evidence that a relatively small reduction in revenues from the highest priced hours is enough to draw attention from the owners of baseload generators. Understanding one of the details of why baseload plant owners have resisted summertime emission regulations yield some insight into why they might perceive PV as a problem. The owners of coal plants cited the probable reduction in price spike revenues caused by the summertime cap-and-trade program for nitrogen oxide (NOx) emissions as 42 a reason the regulation should not be implemented. 4 2 The coal plant owners were concerned about the impact auxiliary power loss would have on revenues during high 42 Informal interviews with state air regulators 52 priced summer hours. The auxiliary power loss from running emissions control equipment, like selective catalytic reduction (SCR), is around 5%. Comparison of the magnitude of lost revenues from a 5% reduction in generation during peak hours to the magnitude of PV's likely impact on prices spikes suggests that baseload owners will care about large installations of PV. The impact of the PV is much larger, similar to the changes with addition of more conventional capacity between years, and therefore significant enough to cause resistance. Table 12 The difference in price spike revenue between years for I MWh of generation (defining price spikes as prices above threshold), reduction in revenue from a 5% reduction in summertime generation, and the reduction in revenues from the addition of 1000 MW PV capacity. Prices above $200/MWh Actual 2001 to 2002 changes 2002 to 2003 2000 2001 generation 2002 reduction in summer Addition of PV 2003 $100/MWh -15215 -3939 -342 -812 -284 -20220 5160 -499 -1055 -567 -13 -112 2004 0 -22 2000 -6177 -8836 -8836 -3789 -5931 -5931 -11462 -5873 2001 ~2001 2002 Change in Revenue ($ per MWh generated) Table 12 shows the price spike revenues (from hours with prices above $200/MWh and $1 00/MWh)for three cases. The top section of the table shows the changes in revenues between 2001 and 2002 and between 2002 and 2003. For example, the revenue earned from the generation of one megawatt-hour in hours priced above $200/MWh in 2002 was about $4000 more than that earned in 2003. Similarly, shown in the bottom section of the table, the estimated impact of 1000 MW of PV capacity in 2002 is a reduction in revenues from hours priced over $200/MWh of about $3800 per megawatt-hour generated. A 5% reduction in summertime generation in 2002 would have decreased revenues from hours priced over $200/MWh by only $300 per megawatt-hour generated. When a baseload unit generates around 500 megawatts per hour, and some 53 baseload plants generate thousands of megawatts per hour, even a reduction of $300 for each megawatt-hour is a substantial loss in annual revenue. For example, for a plant that would have generated 000 MWh in each hour of the summer, a 5% reduction in output would have resulted in a reduction in energy market revenues of about $1 millionr from prices over $200,MWh in 2002. There were only eight hours with prices above $200/MWh in the summer of 2002. The reductions from PV are larger than this 5% but similar magnitude to Those that have occurred historically in New England from bolstering capacity margins with conventional capacity. The difference between PV and new conventional peaking. units for baseload owners might be that PV is new and different or that PV's generation, and whether or not it is installed, is not necessarily under their control. Standbychargesand the impactofPV on operationof CTpeaking units The most commonly discussed technical challenge with installing more PV or wind or demand-side distributed generation technologies (or even demand response) is whether or to what extent that capacity can actually be relied upon to provide power when needed. One measure of PV's dependability is a probabilistic measure that correlates peak.load and solar resource regionally across the United States.43 Although analyses like this are useful in generally understanding what can be expected, they do not help us understand the exact consequences of the slight mismatch between PV generation and electricity 43 See Perez 2001 and R. Perez, R. Seals, and C. Herig, "PV Can Add Capacity to the Grid,?'"NERL Brochure DOE/GO-10096-262, 1996 available at http://www.nrel.gov/ncpv/documents/pv util.htm'llast viewed July 18, 2005. 'The probabalistic measure of PV's ability to act as capacity even though it is not dispatchable is called effective load-carrying capacity (ELCC). The definition of ELCC is "the ability of a powergenerator-whether PV or conventional-to effectively contribute to a utility's capacity, or system output, to meet its load." ELCC for PV in a region depends on the relationship between available solar resource and electricity demand. Thus, regions with relatively low solar resource can still have high values of ELCC that mean PV can be depended upon to meet demand especially during summertime peak hours. The ELCC is related to the ratio of summer to winter peak electricity demands in a region, the higher the summer to winter peak ratio, the higher a region's ELCC. In New England this ratio was 1.0 in 2000, 1.3 in 2001, and 1.2 in 2002. The ELCC for New England is around 60%. A PV system with MW capacity could be considered a dispatchable power source with 0.6 MW of capacity. 54 demand or the consequences of intermittency on the need for back-up generators to maintain reliability. This issue is the highly visible because of the high level of reliability expected from power systems and because of the efforts of generation owners and utilities to levy standby charges against large distributed power producers. They also threaten to and do levy such charges against renewable technologies. Recent efforts by regulators in New England are notable in their efforts pushing for the increased use of renewable power, in parallel with Regional Greenhouse Gas Initiative. Massachusetts's regulators have recognized that standby charges are a barrier to renewable generating technologies and have therefore taken action to ensure that renewable generation is exempt from them, although distributed conventional generation may not be.44 Many states have not taken similar action. Standby charges can do a great deal to prevent the adoption of renewable technologies because once these technologies are economic, standby charges are enough to make investment in them unattractive until costs come down even more. Intermittency as it relates to the need for reliability is a commonly stated reason for the need for standby charges. Utilities and generation owners contend that these charges are needed to help conventional generators cover the costs of not generating but being ready to generate if distributed generation fails. Many interest groups supporting PV claim that standby and other charges like exit fees are excessive.4 5 Some policy makers also suspect this to be the case.46 Past legal decisions reveal the tendency for 44 The Massachusetts Department of Telecommunications and Energy (DTE) on July 23, 2004 exempted small DG systems between 250 and 1000 kW and renewable generation that are used for less than 30% of customers' load from standby charges until August 1, 2008. See Interstate Renewable Energy Council, "'DTE Exempts Some DG from Standby Charges in NStar Territory," September 9, 2004 at hlttp://Iwww.irecusa.org/articles/static/1/1094758241 987096450.html last viewed September 25, 2005. 45 See for example the World Resources Institute position on "Integrating Green Power with the Grid," February 2003 available at http://www.thegreenpowergroup.org/WRI position statement integration 2- 20-03.pdf last viewed September 25, 2005 46 See for example Barnett et al 2001 55 utilities to charge excessive amounts for both exit fees and standby charges.4 7 Some states. have acted to exempt renewable generation from at least some types of standby charges for a limited amount of time.48 .. . 1 treat this issue of intermittency slightly differently than many analyses by .... implicitly rather than explicitly treating PV's intermittency. This study uses historical data for PV generation that is contemporaneous with data for power system and generator operation. As mentioned above. the frequency with which PV generation from a given amount of capacity was smaller than that of the CT units' generationtakes into account those hours during the middle of the day when the CT units generated because it.was hot even though it was cloudy. Interrnittency is an extension of the problem that PV simply cannot generate much in the evenings and CT units typically do generate.in the evenings. These results were reported in Tables 9 and 10. In hours when both.PV and CT units are generating, we can expect the generation from 1000 MW of PV to be larger than that from 170 MW of CT generators nearly 80% of the time. But, PV'generation from 1000 MW was only larger than CT generation in 50% of the hours the CT units generated. Although the problem posed by intermittency is difficult partly because cloudy hours are somewhat unpredictable, the larger portion of the hours that CT and PV generation do not overlap is during the evening. Forcing CT units to operate for fewer hours per year than is economic is not a good or tenable solution to this problem. The prospect of losing operating time and thus the ability to cover costs is a reason for owners of CT units (and potential investors) to resist the adoption of technologies like PV. Table 47One example is MIT's battle with Cambridge Electric Light Company over MIT's combined heat and power plant. See International District Energy Association, "Case Study Series: college campus CHP," at http://www.districtenergy.org/CIP Case Studies/n.pdf last viewed September 30, 2005. 48Mascuetsoo 48 For example, the Massachusetts Department of Telecommunications and Energy (DTE) did so on July 23, 2004. Small DG systems between 250 and 1000 kW and renewable generation that are used for less than 30% of customers' load are exempt until August 1, 2008. See Interstate Renewable Energy Council, "DTE Exempts Some DG from Standby Charges in NStar Territory," September 9, 2004 at htt://www.irecusa.org/articles/static/l/1094758241 987096450.htrnl last viewed September 25, 2005. Also, California has similar regulations in place. See "DSIRE: Summary Tables: ncentives"at hltt://www.dsireusa.or/cdsire/Ilibrarv;'includes'secal iincent ivetype.cfn'tvpe::Net&currentpaeid :-:7&back :- regtab last viewed September 30, 2004. 56 10 shows that 200 MW of PV would decrease the need for 170 MW of CT capacity by 20% and 500 MW of PV would decrease the potential operating hours of 170 MW of CT by about 40%. As it is, these CT units barely cover fixed costs or do not cover them at all. A reduction of even 20% of potential generating hours is certainly enough to cause current owners to fight back and it is enough to dissuade the installation of more CT capacity, which would be needed for reliability even with PV. Battles around standby charges are a manifestation of this problem but the upside is that intermittency, in a stochastic mid-day sense, may not actually be the biggest cause of these fights. For example, in 2002, the CT units generated in about 1080 hours but only 570 of these were between 7 am and 4 pm. Of the hours between 7 am and 4 pm, PV generation from 1000 MW was larger 82% of the time (about 465 hours) and that from 2000 MW was larger 95% of the time about (540 hours). PV only generated in about 560 of the 1080 hours (52%,.see Table 10). The biggest problem with regard to PV's intermittency is thus, about 500 evening and early morning hours. These hours may be much easier to create a solution for than if they were randomly strewn throughout the days, rather than predominantly in the evening hours. This problem does not disappear even if technologies like PV are encouraged through a market mechanism like real time pricing (RTP) on the demand side. In fact, real time pricing on the demand side could cause similar problems even if PV was not installed as a result of its implementation. If RTP encouraged effective demand reduction only during a portion of the normal operating hours of CT units, its implementation would have the same effect as that of the installation of PV. Recommendationsfor problemspresented by incentivesto resist PV The incentives to resist PV installation could be countered by introducing new technologies in a way that would better cover the operation of conventional units - at 57 least to start. For example, programs that encourage a portfolio of new technologies and demand management that has a very high.probability of covering the typical operation of a peaking unit might be more successful than programs that only encourage PV. A program with demand-side real time prices could set peak and off-peak periods in accordance with the typical operation, of peaking units. Eventually the system will evolve to provide capacity when and where-it is needed and systems like location capacity markets and well-designed real time pricing on the demand side may be what is needed to accomplish this. For now though, to overcome the incumbents' incentives to resist. change, grouping demand-side management and renewable technologies appropriately to. take the place of new blocks of conventional generation would be a good place to start. It seems a worthwhile endeavor to design policies that do not cause a situation where new conventional units are still needed for reliability after the installation of renewable technologies but are operated so infrequently that they cannot cover their costs. One important challenge of encouraging the adoption of new technologies in complex technical systems is to understand how that technology will fit within the system as it is currently operated. Although the system will likely evolve because of the technology, and as and after it is fully adopted, the initial transitional steps involve overcoming incentives to resist change. These incentives will be rooted in the current operational patterns of the system and their history. Thus, transition could be smoother if the policies enable the new technology to "fit" in the current system in both a technical and economic sense. This description of New England's power system identified essential issues involved in encouraging large-scale PV adoption. It finds that PV, by itself, will not "fit" in the New England power system. Noting how and why PV does not fit suggests ways to overcome the problems created by incompatibilities: encourage the initial installation of PV simultaneously with technologies that can generate when PV cannot but CT units usually do. 58 CHAPTER Vl - CONCLUSION The comparison of historical solar photovoltaic generation to the typical demand and price- patterns and typical modes of operation of conventional units in a regional power system removes some of the guesswork from envisioning and planning for the incorporation of PV in that power system. The correlation between PV generation and peak electricity demand is often cited as a benefit of PV systems. This analysis shows that PV can impact price spikes in the short-term and that this impact is similar in magnitude to that of the addition of new natural gas-fired capacity. This might not be true of other alternative technologies like wind, but it is not a special benefit of PV. In the long-term one must consider PV's effects on resource adequacy. Small capacity margins during any hours will lead to high prices. Price signals have not been sufficient to encourage the investment in capacity for system reliability49 although some capacity has been installed anyway. The introduction of large amounts of PV might exacerbate this problem if PV reduced mid-day price spikes but could not generate as needed in evening hours. Then, although PV might reduce prices in the middle of the day, high prices would start or continue to occur in the evenings. This could conceivably leave the system as it is now but reduce the number of hours that peaking units operate making them just as necessary for system reliability but leave it harder for them to cover their costs. If the installation of PV resulted from the implementation of real time prices on the demand side it would not necessarily mitigate this problem, although then at least the new capacity additions might make sense economically and be a more direct response to high prices and high demand. It is not possible for PV capacity to replace conventional peaking units as they are normally operated: there are a significant number of hours in which the generation PV does not overlap that of investment worthy combustion turbine units. Further work must 49 Joskow 2003 pg. 57 to 62 and ISO-NE 2003b and 2005 59 be done to compare PV generation with the operation of other conventional.technologies and with the production curves of other alternative technologies in New'England. .. Comparing PV to just combustion turbines do not capture the entire benefit of PV. The PV systems generate some everyday but does not always generate while the combustion turbines do. But, if PV systems were installed with the pupose.ofrep.acing the need fobr new peaking combustion turbines in New England the effort would not succeed because the peaking units would still be necessary in evening hours. One solution for helping PV: "fit" in New England's power system is to develop incentives that.encourage a portfblio of resources that could, with high probability, replace the need for new combustion turbine units.. .. Large amounts of PV might complicate the tricky economics of encouraging a sufficient level of peaking generation to maintain reliability in a few hours per year where demand is up to 60% higher than average while still limiting market power. But, in New.. England, it does not appear that PV would reduce the profits of baseload generators any more than installing more conventional peaking generation. If PV generated enough to. reduce the frequency that higher variable cost units burning natural gas or oil set the marginal price, then the PV could impact baseload-plant profits differently.- in a' manner that might impact their ability to cover long-run costs. It would be very difficult for PV to have that level of impact in a region like New England where natural gas-burning units set the marginal price over 85% of the lime. Further work An opportunity for further work would be to incorporate the understanding of how PV production compares to the typical operation of conventional units into a more formal economic model. This could help quantify the affect of adding large amounts of PV to a 60 power system on the profits of owners of different types (e.g. baseload, intermediate and peaking) of conventional capacity. Extending this type of study to other regions is a possibility although initial attempts at this proved difficult because of constraints on available data. With a substantial effort, this analysis could be extended to regions like New York, Pennsylvania-New Jersey-Maryland, Texas, and perhaps California. This study suggests other further work such as comparing PV generation to other types of conventional capacity or comparing wind generation to the operation of conventional generators. If capturing the environmental and fuel-use-reduction benefits of renewable technologies become priorities, planning and modeling around these types of analyses could help policy makers and system operators understand how to best encourage the use of alternative technologies without increasing power system operational costs. Also, evidence that the installation of certain combinations of renewable technologies would offset the need for new conventional generators could help local communities feel that sacrificing their backyards is worth the effort. 61 BIBLIOGRAPHY Barnett, A. et al. "Solar Electric Power: The U.J.S.Photovoltaic Industry Roadmrap,." available at htt://www.sanii.g.ov/pw,docs/'llP)F/PV Road Ma c)f last viewed September 25, 2005. Borenstein, S. "The Long-Run Efficiency of Real-Time Electricity Pricing," Energy Journal, forthcoming, (CSEM working paper), February 2005 Bushnell, J. and C. Saravia, "An Empirical Assessment of the Competitiveness of the New England Electricity Market" Center for the Study of Energy Markets, paper CSEMWP- 101, May 1, 2002 Cramton, P., "Review of the Reserves and Operable Capability Markets: New England's Experience in the First Four Months," White Paper, Market Design Inc., 1999 Cramton, P. and S. Stoft., "A Capacity Market that Makes Sense," Electricity Journal. forthcoming, August 2005 ISO-NE (2003a), The New England Independent System Operator, "2002 Annual Markets . Report," August 13, 2003 ISO-NE (2003b), The New England Independent System Operator, "2002 Annual Markets Report Technical Review," August 13, 2003 ISO-NE, The New England Independent System Operator, "2004 Annual Markets Report," 2005 Joskow, P.L. "The Difficult Transition to Competitive Electricity Markets in the U.S.," for the conference Electricity Deregulation: Where From Here? Bush Presidential Conference Center, May 2003 NYSRC, New York State Reliability Council, "The Effects of Integrating Wind Power on Transmission System Planning, Reliability and Operations: System Performance Evaluation," at http:',www.nyseirda.orgs/INYS Reliability Cncl omments.2df February 3, 2005 Perez, R. and R. Seals, "Mapping Photovoltaics' Effective Capacilty," Proc, of Annual Utility PV Experience, Lakewood, Co. (Published'by UPVG, Washington, DC), 1996 Perez, R., S. Letendre, and C. Herig, "PV and Grid Reliability: Availability of PV Power during Capacity Shortfalls," Proc. ASES Annual Meeting, Forum, 2001 Smith, J.C., E.A. DeMeo, B. Parsons, M. Milligan, "Wind Power Impacts on Electric Power System Operating Costs: Summary and Perspective on Work to Date," Global WINDPOWER Conference, Chicago, Illinois, March 2004. Stoft, S., Power System Economics: Designing Marketsfor Electricity, IEEE Press, 2002. 62 APPENDIX A Solar Resource and PV Generation in New England The solar resource in New England is not as high as in other parts of the country.5 0 But PV generation could still be a valuable resource. Five PV sites in New England were selected for use in this analysis. All five of these generated throughout the period of January 2000 through December 2002 and data for all hours of the years was reported for the systems. The systems were selected for their high data availability rather than their geographic distribution' Two of the systems are located in North Dartmouth, Massachusetts. These systems have ratings of about 14 and kW (DC ratings under standard test conditions, STC-DC). Two other sites are also in Massachusetts, in Cambridge and Lynn with ratings of 23 and 5 kW (STC-DC). The final site is in Middletown, Rhode Island and has a rating of about 40 kW (STC-DC). Table 13 and Table 14 show the PV site characteristics as well as their generation in years 2000 through 2002. These five PV sites were scaled in the analysis so that each site represented one fifth of the assumed capacity. That is, if the analysis was testing the impact of 1000 MW of PV capacity, five 200 MW capacity sites were modeled by scaling each real site by the necessary factor. Although these sites are not evenly spread across the region, they do represent different tilt angles and latitudes so some of the natural variation is accounted for that would be found with the realistic installation of many, distributed PV sites across New England. 50 See, for example, the Energy Information Association's map of concentrated solar resource at ittp://www.eia.doe.2ov/cneaf'solar.renewables/ilands/fig 12.html last viewed June 23, 2005. 51Schott Applied Power provided data for PV systems for 2000 through 2002 for another study at the Laboratory for Energy and the Environment. 63 Table 13 PV site characteristics City Cambridge Rating (kW Tilt angle Azimuth State STC-DC) Latitude (Degrees) (Degrees) MA 22.8 42.400 39 155 Expected SiteMax Generation (kW_ (kWh) 21 .. 31920 Lynn North Dartmouth North Dartmouth MA MA MA 4.6 1.0 13.7 42.460 41.642 41.642 25 75 1 180 180 180 4 1 11J 6384 1344 19152 Middletown RI 40.0 41.000 0 180 38: 56000 Table 14 Actual PV site generation' Rating City Cambridge Lynn North Dartmouth North Dartmouth Middletown State MA MA MA MA RI (kW STC-DC) 22.8 4.6 1.0 13.7 40.0 2000 2001 Generation Generation (kWh) (kWh) 21901 21299 5445 5853 1018 1285 1.3910 14645 40342 53059 2002 Generation (kWh)_ 20223 5567 1104, 14259 50609 64