Reducing the Cost of through Learning By Doing Edward S. Rubin Department of Engineering and Public Policy Department of Mechanical Engineering Carnegie Mellon University it Pittsburgh, Pennsylvania Presentation to the Clearwater Coal Conference Clearwater, Florida June, 14 Outline of Talk A brief overview of learning curves Application to power plants Application to emission control technologies Implications for future costs T.P.Wright (1936) found that the cost of making airplanes declined with increasing experience Overview of learning curves Learning by doing reduced manufacturing costs exponentially 1
Learning Curve Concept Later Extended to Technology Experience Curves Examples of Technology Cost Trends Used to Derive Experience (Learning) Curves The Common (One-Factor) Learning Curve Model General equation: where, C i = a P i b C i = cost to produce the i th unit P i = cumulative production or capacity thru period i b = learning rate exponent a = coefficient i (constant) t) Fractional cost reduction for a doubling of cumulative capacity (or production) is defined as the learning rate: LR = 1 b Literature Review of Learning Rates for Power Generation Technologies Application to power plants PC plants PC with IGCC plants IGCC with NGCC plants NGCC with NG turbines Biomass plants Nuclear Hydroelectric Geothermal On-shore wind Off-shore wind Solar PV Conc. solar thermal
L Reported Learning 4% % Rates for Wind % s Based 18% 16% on $/kw earning Rate Spain UK Spain UK Spain Spain 14% 35% 1% Denmark Global 1% 3% Spain & Denmark 8% Denmark Germany 5% Germany 6% Denmark Sweden % 4% OECD % 15% Global % Global 198 1985 199 1995 1% Period Covered Global 5% McDonald 1 Junginger 5 Neij 3 IEA % Source: Azevedo, et al., 13 1975 198 1985 199 1995 5 1 15 Period Covered 8 Nemet McDonald 1 Neij 9 Wiser 1 Learning Rate 35% Reported Learning 3% 5% Rates for Wind % 15% s Based 1% 5% on $/kwh % Learning Rate Learnign Rate US California China 1975 198 1985 199 1995 5 1 Period Covered IEA McDonald 1 Qui 1 8% UK 5% 3% % EU 18% 15% 13% Germany 1% Denmark 8% 5% 3% % 1978 198 198 1984 1986 1988 199 199 1994 1996 1998 3% Germany 5% Period Covered Ibenholt IEA Neij 3 Source: Azevedo, et al., 13 Distribution of Reported Learning Rates for On-Shore Wind Distribution of Reported Learning Rates for Solar PV 1 Count 1 8 6 4 Mean = 16% Median = 1% Std Dev = 14% Mean = % Median = 1% Std Dev = 1% 5% to 1% % to 4% 5% to 9% 1% to 14% 15% to 19% Learning Rate % to 4% 4% to 9% 3% to 34% Dependent Variable = $/kwh (n=1) Dependent Variable = $/kw (n=3) Source: Azevedo, et al., 13 Source: Azevedo, et al., 13 3
Distribution of Reported Learning Rates for Gas-Fired Power Plants Range of Reported Learning Rates Frequency 3 1 Mean = 14% Median = 13% Std Dev = 13% 15% to 1% to 5% to 9% 4% 1% % to 4% 5% to 9% 1% to 14% 15% to 19% % to 4% 5% to 9% Learning Rate Dependent Variable = $/kw (n=9) Dependent Variable = $/kwh (n=) 3% to 34% Source: Azevedo, et al., 13 g g Technology Number Number Range of Range of rates Years Number of studies of studies learning rates for learning by covered of studies with one with two for learning by researching across all reviewed factor factors doing (LBD) (LBR) studies Coal* PC 5.6% to 1% 19-6 IGCC 1 1.5% to 7.6% Projections Natural Gas* 8 6-11% to 34%.38% to 17.7% 198-1998 Nuclear 4 4 < to 6% 1975-1993 Wind (on-shore) 35 9 6-3% to 3% 1% to 6.8% 198-1 Solar PV 4 1% to 53% 1% to 18% 1959-1 BioPower Biomass production 4 4 1% to 45% 1971-6 Power generation** 7 7 % to 4% 1976-5 Geothermal power 3 198-5 Hydropower 3.48% to 11.4%.63% to.6% 198-1 *Does not include plants with. **Includes combined heat and power (CHP) and biodigesters. Source: Azevedo, et al., 13 Early Trends in Amine Performance and Cost Measures Application to emission control technologies Decreasing trend of MEA regeneration heat reqm t. Estimates from cost studies, not actual projects Capital cost estimates for an amine system at a 5 MW coal-fired plant (on a gross power basis) Source: Yeh & Rubin, 1 What can be learned from the history of other pollution control systems? 4
U.S. Government Actions Affecting SO and NO x Control Technology Legislation / Regulation Clean Act Amendments of 197, 1977, 199 New Source Performance Standards of 1971, 1979, 199 Funding / Financial Incentives EPA multi-million $ research budget in 197s DOE Clean Coal Technology Program (since 1985) Facilitating Technology Transfer SO Control Symposium (starting 1969) Symposia and workshops on multiple pollutants (starting in 197s) Regulatory requirements established markets for environmental technologies Num mber of Patents Filed U.S. Patenting Activity in SO Control Technology (188 ) 1 11 1 9 8 7 6 5 4 3 1 No Federal Some Federal U.S. Clean Act of 197 CAA Regs + 188 189 19 191 19 193 194 195 196 197 198 199 Year Patent Filed Atop an Early SO Absorber Unit 1 FGD system, Bruce Mansfield Power Station Many of the early flue gas desulfurization (FGD) system designs were complex and costly 5
Adoption of FGD Technology (Coal-Fired Power Plants) Typical Cost Trend for a New Technology Cum mulative Capacity of Wet Scrubbers (GWe) 1 11 1 9 8 7 6 5 4 3 1 197 1974 US Japan Germany Other 1976 1978 198 198 1984 1986 1988 199 199 1994 1996 1998 Capital Co ost per Unit of Capacity Research Development Demonstration Deployment Mature Technology Year Scrubber in Service Time or Cumulative Capacity Source: Based on EPRI, 8 Trend of FGD Capital Cost Patenting Activity in NO x Controls (U.S. Patents, Class-based dataset) ital Costs ($/kw) in 1997$ Capi 3 5 15 1 5 1976 1975 198 198 199 1974 197 (1 Initial MW cost, eff estimates =8-9%) were a bit optimistic 1968 ( M W, eff =87%) (O&M costs also low) 1995 1 3 4 5 6 7 8 9 1 Cumulative W orld W et FGD Installed Capacity (GW ) Source: Rubin et al. 7 Num mber of Patents Filed 18 16 14 1 1 8 6 4 1873 1879 1885 1891 All Methods of NO x Removal 1897 193 199 1915 191 197 1933 1939 1945 1951 1957 Year Patent Filed 1963 CAA 1969 1975 1981 1987 1993 1999 (Post-1976 data based on 9% SO removal, 5 MW plant, 3.5%S coal) 6
Patenting Activity in Post-Combustion NO x Controls Trend of SCR Cost Estimates Number of Patents Filed N 1 U.S. Regs 11 1 German 9 Regs 8 Japanese 7 Post-Combustion Regs 6 NO x Removal 5 4 3 1 188 189 19 191 19 193 194 195 196 197 198 199 Year Patents Filed SCR Capital Costs ($/kw), 1997$ 1 11 1 9 8 7 6 5 1977 1978 1979 198 First Japan commercial installation on a coal-fired power plant (early O&M costs also low) 1983 First German commercial installation 1989 First US commercial installation 1993 1995 4 - -1 1 3 4 5 6 7 8 Cumulative World SCR Installed Capacity (GW) Source: Rubin et al. 7 Historical Learning Curves for FGD and SCR s Standard log-linear models fit to data for declining FGD and SCR capital costs These values reflect the real change in cost of doing the same job at different points in time for the same power plant and fuel specifications Normalized Capital Cos 1% 1% SCR y = 1.41x -. R =.76 FGD y = 1.45x -.17 R =.79 Cost reductions of ~1% per doubling of installed capacity (~ 5% reduction in years) 1 1 1 1 Worldwide Installed Capacity at Coal-Fired Utility Plant (GWe) Implications for future cost of carbon capture and storage () 7
k Cost of for New Coal-Based Plants Using Current Technology Cost for New NGCC Plants (Current Technology) Increase in levelized cost for 9% capture Incremental Cost of relative to same plant type without (based on bituminous coals) % Increases in capital cost ($/kw) and generation cost ($/kwh) Supercritical Pulverized Coal Plant Integrated Gasification Combined Cycle Plant ~ 6 8% ~ 3 5% Capture accounts for most (~8%) of the total cost Retrofit of existing plants typically has a higher cost Added cost to consumers will be much smaller (reflecting the capacity in the generation mix at any given time) Increase in levelized cost for 9% capture New NGCC Cost Measure Cost Increase with % Increase in generation cost ($/kwh) ~ 3 45% (relative to NGCC w/o ) Cost of Avoided: Relative to NGCC: ~$1 /t Relative to SCPC: ~$4 /t Ten Ways to Reduce Cost (inspired by D. Letterman) Application of Learning Curves to Power Plants with 1. Assume high power plant efficiency 9. Assume high-quality fuel properties 8. Assume low fuel price 7. Assume EOR credits for storage 6. Omit certain capital costs 5. Report $/ton based on short tons 4. Assume long plant lifetime 3. Assume low interest rate (discount rate). Assume high plant utilization (capacity factor) 1. Assume all of the above!... and we have not yet considered the technology! PC Plant Oxyfuel Plant Electricity to atmosphere Turbine Generator Coal Pollution CO Controls Capture Mostly PC Boiler N (SCR, ESP, FGD) Amine Amine/ Amine/ to Separation Compression storage Electricity Turbine to atmosphere Generator to atmosphere Pollution Coal CO Distillation to storage PC Boiler Controls H O compression (ESP, FGD) O water to recycle Separation Unit Stack NGCC Plant Separation Unit O Coal H O Gasifier Natural Gas IGCC Plant Quench Combustor H O Four Baseline Plant Designs with Electricity Gas Turbine Compressor Turbine Generator Generator Capture Mostly N Amine Amine/ Amine/ Separation Electricity to atmosphere Sulfur CO Gas Turbine Shift H H Removal Capture Combined Reactor Cycle Selexol Selexol/ Selexol/ Separation compression Sta ack Compression to storage to atmosphere Stack to storage 8
Seven Steps to Project Future Costs Disaggregate plant into major components Estimate current plant cost and component contributions Select learning rate for each plant component Estimate current installed capacity of each component Set capacity additions for start and end of learning Aggregate component results back to plant level Conduct sensitivity analysis on key uncertain variables, e.g., Starting point for learning curve End point for learning curve Choice of and basis for current capacity data Basis for multi-year cost adjustments Plant cost parameters Learning Rates for Plants w/ Capture Power Pla ant Learning Rate (%) (based on improvements in major plant components) TOTAL CAPITAL COST COST OF ELECTRICITY 1 8 6 4 NGCC PC IGCC Oxyfuel Power Pla ant Learning Rate (%) 1 (excluding T&S costs) 8 6 4 NGCC PC IGCC Oxyfuel Learning rates for a given plant type vary by about a factor of 3 Source: Rubin, et al., 7 Projected Reductions in COE after 1 GW of Experience Projected Cost Reductions in 5 for Global Energy Scenarios Projected reduction in cost of electricity generation (COE) for each plant type varies by factors of ~ to 4 Projected reductions in mitigation cost are larger than for the overall plant Percent Re eduction in COE % COE REDUCTION AFTER 1 GW EXPERIENCE 3 5 15 1 5 NGCC PC IGCC Oxyfuel Source: Rubin, et al., 7 Cost reductions, 1 5, 5, based on energy-economic economic modeling with endogenous learning curves for power plants with * Power Plant Reduction in Cost of Electricity ($/MWh) Reduction in Mitigation Cost ($/t avoided) NGCC 1% 4% 13% 6% IGCC % 5% 19% 58% PC 14% 44% 19% 6% * Range based on low and high global carbon price scenarios. Source: van der Brock et al, 1 9
Which capture technologies will have the lowest cost? Learning Curves Can t Pick Winners Programs Seek to Develop Lower-Cost Technologies Cost Reduction Benefit Post-combustion (existing, new PC) Pre-combustion (IGCC) Oxycombustion (new PC) compression (all) Chemical looping OTM boiler Biological PBI Ionic liquids processes membranes Metal organic CAR process Advanced Solid frameworks physical solvents Advanced sorbents Membrane systems Enzymatic membranes Amine chemical ITMs solvents solvents Biomass co- Physical Ammonia firing solvents CO com- Cryogenic pression oxygen Present 5+ years 1+ years 15+ years + years Time to Commercialization Source: USDOE, 1 Magnitude of future cost projected using learning curves depends strongly on assumed initial cost Bottom up engineering-economic analyses offer some insights into cost of new technology designs $/MWh ($9) 175 155 135 115 95 75 55 35 15-5 No IGCC Today IGCC Technologies with No 7% below no with IGCC 31% w/ reduction IGCC w/ Today with Pulverized Coal Technologies No Supercritical PC Today with No 9% above no with Supercritical 7% PC reduction Adv Combustion w/ w/ Source: DOE/ NETL, 1 www.iecm-online.com Can Alone Achieve the Big Cost Reductions We Seek? A More Realistic Model The Linear Model of Technological Change Invention Adoption Diffusion Innovation Invention Innovation (new or better product) Adoption (limited use of early designs) Diffusion (improvement & widespread use) Learning By Doing Learning By Using Deployment of new technology is required to achieve biggest cost reductions 1
Government policies are needed to create market demands for Direct Gov t Funding of Knowledge Generation contracts with private firms (fully funded or costshared) Intramural in government laboratories contracts with consortia or collaborations Policy options that can foster innovation Technology Policy Options Direct or Indirect Support for Commercialization and Production tax credits Patents Production subsidies or tax credit for firms bringing new technologies to market Tax credits, rebates, or payments for purchasers/users of new technologies Gov t procurement of new or advanced technologies Demonstration projects Loan guarantees Monetary prizes Knowledge Diffusion and Learning Education and training Codification and diffusion of technical knowledge (e.g., via interpretation and validation of results; screening; support for databases) Technical standards Technology/Industry extension program Publicity, persuasion and consumer information Regulatory Policy Options Economy-wide, Sector-wide, or Technology- Specific Regs and Standards Emissions tax Cap-and-trade program Performance standards (for emission rates, efficiency, or other measures of performance) Fuels tax Portfolio standards Source: NRC, 1 Conclusions from Learning Curve Studies There is significant potential to reduce the cost of but Realization of that potential will require significant ifi commercial deployment of in addition to sustained A Final Word of Wisdom It s tough to make predictions, especially about the future Thank You - Yogi Berra rubin@cmu.edu 11