by Peter Lang
A revolution could be achieved with nuclear power if we remove the factors that caused the large cost increases during and since the 1970’s, i.e. return to the learning rates demonstrated before 1970.
Main Points:
- Learning rate is the rate costs reduce per doubling of capacity. Until about 1970 learning rates for nuclear power were 23% in the US and 27% to 35% in the other countries studied, except India.
- Around 1970, learning rates reversed and become negative (-94% in the US, -82% in Germany, -23% to -56% in the other countries, except South Korea); clearly something caused the reversal of learning rates for nuclear power around 1970.
- If the positive learning rates from 1953 to 1970 had continued, nuclear power would cost less than 1/10th of current cost.
- If nuclear deployment had continued at 30 GW per year from 1980, nuclear would cost much less than 1/10th of what it does now; furthermore the additional nuclear generation would have substituted for 85,000 TWh of mostly coal-generated electricity, thereby avoiding 85 Gt CO2 emissions and 5 million fatalities.
- In 2015, assuming nuclear replaced coal, the additional nuclear generation would have replaced half of coal generation, thus avoided half of the CO2 emissions and 300,000 future fatalities. If the accelerating rate of deployment from 1960 to 1976 had continued, nuclear would have replaced all baseload coal and gas generation before 2015.
- High learning rates were achieved in the past and could be achieved again with appropriate policies.
Introduction
History is replete with examples of one technology replacing another. The rate that technology transitions take place is highly dependent on the technology learning rates that exist during the transition period. If we want electricity generation that is cheap, clean, safe and has low CO2 emissions, then policies need to focus on ways to improve the learning rates of the technologies that meet the requirements. Historical learning rates provide some insight into what rates are achievable and what could be done to increase learning rates for technologies that meet requirements.
The concept of learning rates is widely used to quantify the ability to reduce costs as experience is gained. Rubin, et al., 2015, ‘A review of learning rates for electricity supply technologies’ defines ‘learning rate’ as “the fractional reduction in cost for each doubling of cumulative production or capacity”. The authors explain how to calculate learning rates and provide a summary of learning rates for a selection of electricity generation technologies. Unfortunately, the paper has little information on nuclear learning rates and none before 1972 or after 1996.
Lovering et al., 2016, ‘Historical construction costs of global nuclear power reactors’ provides a comprehensive analysis of nuclear power construction cost experience of early and recent reactors in seven countries; the analysis covers 58% of the civil nuclear reactors constructed for electricity generation globally between 1953 and 2008 (based on construction start dates). Table 1 (reproduced below) defines periods with different stages of development and different trends in construction costs for the seven countries.
Table 1: Summary of Overnight Capital Cost (OCC) trends by country
Method
I re-analyse Lovering’s data to calculate learning rates. Figure 1 plots Overnight Capital Cost (OCC) ($/kW) versus cumulative global capacity (GW), for the nuclear points in Lovering’s Figure 13. It shows there was a marked reversal in the slope of OCC versus cumulative global capacity. Before cumulative global capacity reached around 32 GW, OCC was decreasing as cumulative capacity increased (i.e. positive learning rates). Then an abrupt change occurred; thereafter, OCC was increasing (i.e. negative learning rates). Trendlines are fitted to the US data points before and after 32 GW cumulative global capacity to highlight the dramatic change.
Figure 1:
Given this clear evidence for two phases, I calculated the learning rates for two periods, before and after the change in slope, or inflection point, for each country. The inflection point did not happen at the same time in all countries. It occurred first in the US, there was a lag to Canada and Europe and a further lag to Asia. The inflection points I selected are: 32 GW for US; 64 GW for Canada, France and Germany; and 100 GW for Japan, India and South Korea. I divided the data points into two series for each country: Period 1 before and Period 2 after the inflection point. I plotted the points on a log-log plot (base 2), fitted trendlines to each period for each country and calculated the learning rate for each. Following Rubin, et al., 2015, ‘A review of learning rates for electricity supply technologies’ I calculated learning rate by regressing Overnight Capital Cost against cumulative global capacity using a power function. Learning rate is equal to 1-2b where b is the exponent of the fitted power function.
Results
Figure 2 has seven charts, one for each of the seven countries; each shows the data points for that country and the trendlines fitted to the periods before and after the inflection point. The power equation for each trendline is shown on the charts.
Figure 2:
To facilitate comparison of the trends for the different countries, Figure 3 shows all on one chart. Japan and France had the fastest learning rate in the 1st phase and Korea had a similar rate since it started building reactors in 1972, but starting from a high OCC after the initial rapid cost escallation in the other countries.
Figure 3:
The learning rates for the first and second period in each country are in Table 2. For each period, learning rates are given for cumulative global capacity and for the country’s cumulative capacity. The 6th column is the inflection point for each country. The last column lists the projected OCC at 500 GW cumulative global capacity of constucted reactors if the 1st period learning rate had continued until now.
Table 2:
Learning rates are affected by the growth of cumulative capacity both globally and in the country building the reactors. That is, experience gained globally and in the country both contribute to the learning rate for that country. As shown in Table 2, the differences in the learning rates calculated by the two methods are relatively small. I have followed the example of Lovering et al. and use global cumulative capacity in the charts above and below. The periods over which the learning rates apply are plotted in Figure 4.
Figure 4:
Discussion
Table 2 shows that until about 1970 learning rates for nuclear power averaged 23% in USA and 27% to 35% in the other countries, except India (7%). Around 1970, learning rates changed abruptly and become negative (-94% in USA, -82% in Germany, -23% to -56% in the other countries), except in South Korea; South Korea started building nuclear power plants after the initial rapid cost-escalation period, achieving 33% learning rate since 1972. The fact that high learning rates existed up to about 1970 suggests they could be achieved again. Something disrupted and reversed progress in the late 1960s and 1970s (the cause will be discussed in another paper).
If the 1st period learning rates had continued until today, when cumulative global capacity of reactors constructed so far is around 500 GW, the OCC of nuclear power would be less than 1/10th of what it is now, e.g., around $260/kW (France), $350/kW (US), $740/kW (Japan). This is much lower than fossil fuels and other alternatives. Clearly, if we can once again achieve the high learning rates of pre-1970’s then nuclear power will become much cheaper than the alternatives.
The US’s learning rate during the second period was the worst of the seven countries. The second period started a few years later in the other countries and the cost escalation was not as severe as in the US. This suggests the US influenced the development of nuclear power in all seven countries (and probably all countries). It also shows technology learning rates and transition rates can change quickly and disrupt progress, in this case delaying it for half a century so far.
Figure 5 shows the cumulative global capacity of constructed reactors versus the construction start date.
Figure 5:
If the rate of deployment from 1970-1976 had continued, cumulative global capacity of nuclear (based on construction starts) would be 1,150 GW in 2015 (linear projection); it would be around 2,900 GW if the acceleration rate from 1960 to 1976 had continued. Figure 6 (from IAEA ‘Technology Roadmap, Nuclear Energy, 2015 Edition’ Figure 2) shows the rate of grid connections peaked at over 30 GW per year in 1984 and 1985.
Figure 6:
If the 30 GW per year rate had continued since 1985, cumulative global capacity of nuclear would have been 1,320 GW in 2015. At this rate OCC would be much less than 1/10th of what it is now and the extra nuclear generation, from 1980 to 2015, would have substituted for 85,000 TWh[i] of mostly coal-generated electricity globally and avoided approximately 85 Gt CO2[ii] and 5 million fatalities[iii] from pollution. For the year 2015, assuming nuclear replaced coal, the additional nuclear generation would have replaced 54% of coal generation, avoided 54% of the CO2 emissions and saved 300,000 future fatalities. If the accelerating rate from 1960 to 1976 (see Figure 5) had continued, nuclear would have replaced all baseload coal and gas generation before 2015.
If we remove the impediments that reversed the learning rates, we could achieve high positive learning rates again. That would lead us to cheap, clean, safe, nuclear power and to decarbonisation of electricity systems globally (over decades). How the impediments can be removed will be discussed in another paper.
Policy implications
Energy is the lifeblood of modern civilization. Policies that increase the cost of energy are unlikely to be politically sustainable and, therefore, unlikely to succeed[iv]. Therefore, to reduce the emissions that are detrimental to health and the environment, countries will need access to low-emissions technologies that are cheaper than high-emissions technologies.
Furthermore, cheap electricity increases productivity and GDP growth rate, drives faster electrification for the people without any electricity or with insufficient and/or unreliable electricity and thus more quickly lifts the world’s population to higher standards of living. As electricity costs decrease the deployment rate increases, capacity doublings occur faster so costs reduce faster (i.e. we progress more quickly down the learning curve). Technology transition takes place faster and the benefits are delivered sooner.
This revolution could be achieved with nuclear power if we remove the factors that caused the large cost increases during and since the 1970’s, i.e. return to the learning rates demonstrated before 1970. These factors, which will be discussed in another paper, represent impediments to transition to cheap, clean, safer electricity and the benefits that flow from that. The cost of nuclear power could decrease at the learning rates achieved in the first period if these impediments are removed.
To remove the impediments I suggest we need a catalyst to get people to reconsider the basis for their fears about nuclear power and to take another look at the costs, benefits and risks of nuclear power for the world. A suggested catalyst and a way to proceed will be the subject of another paper.
Endnotes
[i] Total world nuclear generation (TWh) per year is factored up in proportion to the projected / actual cumulative global capacity from 1980 to 2012 (EIA ‘International Energy Statistics’)
[ii] Assumed CO2 emissions intensity of electricity displaced by nuclear power is 1 t/MWh (= 1 Mt/TWh)
[iii] Assumes nuclear power avoids 60 fatalities per TWh, i.e. “coal electricity – world average” (60) minus nuclear (0.09) (‘Deaths by energy source in Forbes’, 2012)
[iv] Peter Lang 2015, ‘Why carbon pricing will not succeed’; although this is about carbon pricing the main points apply to all policies that would raise the cost of energy.
JC note: As with all guest posts, please keep your comments relevant and civil.Filed under: Energy
