Response from Professor Gordon Hughes to previous posting

(Professor Hughes has very kindly provided a response to a recent posting on this site. (Electricity output figures show wind turbine performance deteriorates very slowly with age). The original article was also carried on other web sites and Professor Hughes refers to the title and date of publication on the Ecologist blog. My reply to Professor Hughes is carried as a comment below his text.)

 

Wind Turbine Performance Over Time: A Response to Chris Goodall

 

In his blog published on 03.01.14, “Wind turbines – Going strong 20 years on”,[1] Chris Goodall argues that the degradation in the performance of wind turbines with age is much lower than reported in my 2012 study The Performance of Wind Turbines in the United Kingdom and Denmark.[2] The following note explains why I believe that my conclusions are sound.

Mr Goodall has kindly provided me with the data to which he refers to in his work. With the exception of a long series for Delabole wind farm, Mr Goodall’s data is a small subset of the much larger sample of wind farms, several hundred in fact, analysed in my original study. Mr Goodall’s data also adds a few monthly observations that were missing when my data was originally extracted from the source database. Overall, Mr Goodall’s data amount to about 5% of the data that I analysed, and where he has new material it adds very little.

Furthermore, Mr Goodall himself very frankly admits that he does not have the statistical skills required to replicate the methods of my analysis. His work does not constitute a reanalysis or a rebuttal of my paper. In fact, his calculations simply reproduce one feature of the results reported in my paper.  There was a generation of wind farms developed in the early 1990s, both in Denmark and the UK, using turbines of less than 0.5 MW which have experienced a relatively limited decline in performance with age.  By focusing exclusively on these wind farms, Mr Goodall misses the bigger picture.  The performance of wind farms developed from the mid-1990s onward is much worse.  The average size of the turbines and the wind farms increased.  The larger turbines appear to have been less reliable, while my analysis suggests that the siting and maintenance of wind farms may have deteriorated.

Mr Goodall concludes with two challenges/questions which are representative of many comments on my work.  They spring from a lack of understanding of the statistical reasoning involved.  I will begin with his second question, since it is central to the analysis. Mr Goodall wonders how it is possible to estimate the decline of load factors over time when we have less than twenty years of data for any wind farm. This is where the mathematical/statistical specification described in the Appendix to my paper is crucial.

The load factor for any wind farm in any period is expressed as the sum (or product in the multiplicative version) of components associated with the age of the wind farm (held constant over all wind farms of the same age), the period (constant over all wind farms in one period), the site of the wind farm (constant over time and age), and a random error. This is a standard formulation used by statisticians, including for the analysis of data from a wide range of medical and biological trials. The age effects can be identified from the variation in output across wind farms of different ages for each month. So long as each wind farm is tracked for a number of periods, the site characteristics of the wind farm can be separated from age effects which are common to all wind farms of the same age.

In his first question, Mr Goodall challenges me to produce a counter-example to the case of Delabole, which he claims demonstrates a much lower rate of degradation with age than that reported in my paper (in fact it is similar to the overall rate I report for Denmark). This is a recurrent theme among critics of my work. As an argument it is equivalent to someone claiming that smoking cannot harm anyone’s health because their “Uncle Jack” has smoked a pack a day for 60 years and is still fit and well at an age of 80. Of course there are apparent counter examples, and these can be found in the REF load factor database: www.ref.org.uk. It would be invidious to name them, and in any case they no more prove my analysis than Delabole disproves it. Individual cases prove nothing about population epidemiology, a point which is as true for wind power as for public health. The proof is in the statistical analysis itself.

As a separate point, I am struck by how selectively critics report the results of my work. As noted above, the experience of Delabole and other wind farms built in the period 1991-93 is consistent with my analysis of wind farms in Denmark, where load factors seem to decline more gently with age. That may reflect the robustness of wind turbines built in the early 1990s, site choice, how they have been maintained, and other factors. For the avoidance of doubt, I do not argue that the performance of wind farms must, inevitably, degrade rapidly with time. My observation is that the average performance of wind farms in the UK has, as a matter of fact, fallen as they have aged, a fact that is probably the result of both the physical characteristics of wind power and the economic characteristics of the financial incentive regime, the Renewables Obligation subsidy.

My results have important and obvious implications for both investors and policymakers. But the response of advocates of wind power is rather interesting. For the most part, it has involved an attempt to shoot the messenger rather than trying to understand the underlying phenomena. Yet, none of the statistical analyses of my or other data have demonstrated that there is no degradation in performance in age. The issue is not whether degradation occurs, but how much. There can be reasonable disagreement about that, as the comparison between Denmark and the UK illustrates (which is why I included that in my original study). The key point is to identify the causes of changes in load factors over time revealed by statistical analysis, and whether and how these may be addressed.

The willingness of the owner/operator of Delabole to provide unpublished data on output from the wind farm is to be commended, but, though welcome, it is only a small step in the right direction. Any investigation in this area is hampered by the unwillingness of operators to provide the wind speed data collected by the anemometers which are installed at all wind farms. Let me briefly indicate why this matters. One explanation for performance degradation over time would be an increasing frequency (or length) of mechanical failures of turbines. An alternative explanation is that the power curve (the relationship between wind speed and power output) changes due to gradual erosion of the blades, a phenomenon well known in the industry. An assessment of the relative contribution of these – and other – factors can be used to improve both turbine designs and maintenance regimes for existing wind farms, but such work cannot happen until the anemometry data from individual wind farms is made publicly available.

An ostrich-like approach of denying that there is a problem helps no-one. A lack of transparency leads to the suspicion that wind operators are unwilling to be accountable for the large sums of public money which they are currently receiving, and certainly makes it difficult to ensure that subsidy policies give good value for money to the consumers who foot the bill. But even the wind industry does not benefit in the long run, because it is foregoing the opportunity to learn from and build on the lessons from detailed analysis of performance.

Gordon Hughes

05.01.14

About the Author

Dr Gordon Hughes is a Professor of Economics at the University of Edinburgh, where he teaches courses in the Economics of Natural Resources and Public Economics. He was senior adviser on energy and environmental policy at the World Bank until 2001.



[1]http://www.theecologist.org/blogs_and_comments/commentators/2221532/wind_turbines_going_strong_20_years_on.html

[2] Gordon Hughes, The Performance of Wind Turbines in the United Kingdom and Denmark (Renewable Energy Foundation: London, 2012). Available for download at www.ref.org.uk.

  1. Chris Goodall’s avatar

    I want to thank Professor Hughes for his comments.

    Like many others, I entirely agree with him that wind farm operators in receipt of billions of pounds of yearly subsidies should be obliged to provide much more data about their turbines. In my view, but probably not Professor Hughes, the UK needs tens of billions of further investment in onshore wind farms and better availability of data about existing turbines would help investors decide where such farms should be most productively sited. And, importantly, whether it makes sense to invest in wind at all. Measurements of wind speeds at publicly funded wind farms should be available to all.

    To move on to his points of criticism. He says that my analysis focuses on just 14 wind farms, representing about 5% of the total set of information that he used. He’s correct. I selected these farms because they are the oldest operating in the UK, all of them being at least 18 years old. According to Professor Hughes’ model, they should therefore be averaging output of less than 11% of their maximum. My point was they are not. They are doing over twice as well and show no signs of declining power. Whatever the statistical sophistication of his model, it contradicts what all the operators of old turbines have found.

    If I understand Professor Hughes correctly, he does admit that none of these farms show a rate of output deterioration consistent with his model. He says that this is because all these farms employ small wind turbines, and because of superior maintenance. I should mention that nowhere in Professor Hughes’s paper does he mention his theory that smaller turbines age better than big ones. (He does indicate that small UK wind farms are better than big ones, but this is not the same point).

    I asked Professor Hughes to give us an example of an old UK wind farm whose output has declined at the rate that he suggests is typical. He has chosen not to do so, saying it would be invidious. Of course he is right to say that one single set of observations would neither invalidate nor confirm his hypothesis. That’s not what I am asking: I’d just like to see a single illustration of any onshore wind farm – out of the 380 or so in the UK – that performs as he suggests. A sophisticated econometric model that is at variance with every single set of observations used in that model has clearly failed.

    To try to make this point more clearly, I’ll use the Uncle Jack analogy mentioned by Professor Hughes. The fact that the old chap smoked like a chimney but lived to a great age does not disprove the hypothesis that cigarette use reduces life expectancy. I agree. This was not my point. I suggested that if no smokers (wind farms) ever die young (lose power with age) then any model that proposes that smoking causes disease must be wrong. Professor Hughes cannot give us any Uncle Jack – not one – who died young as a result of smoking. Any model that says smoking cuts life expectancy but cannot show a show a single example of a smoker dying early is not credible.

    Professor Hughes rightly says that I do not have the competence to understand his statistical techniques. He’s right: I tried for several hours, but nevertheless utterly failed to comprehend what he was doing in his statistical work. (Despite attending his lectures at Cambridge University in the ’70s, it shames me to say). I’ve talked to some specialists since and am amazed to learn just how Professor Hughes gained his results.

    My failings are not unusual. Professor David MacKay FRS, the DECC Chief Scientist had similar problems. In layman’s language, Professor MacKay suggested that Hughes’ model couldn’t reliably indicate the expected rate of deterioration with age. If MacKay cannot support the statistical techniques that Professor Hughes used, I wonder who can? Hughes’s methods must be exposed to analysis from professional econometricians.

    Last point. The people who built and operate wind farms aren’t stupid. They invest their money to earn a return. Their income comes entirely from the electricity generated by their turbines. There’s not a cent of profit if the blades aren’t turning. So if wind turbines deteriorate rapidly, the operators earn less. If Professor Hughes is right, they earn a lot less.

    I’ve yet to hear a single wind farm operator ever voice a concern about the loss of income resulting from the ageing of turbines. As far as I can tell, severe deterioration of wind farm performance with age is not a phenomenon recognised by anybody who works in the business. Economists doing statistical analysis should, in my opinion, test their hypotheses with people working in the industry before releasing papers.

    Once again, I thank Dr Gordon Hughes for the contribution he continues to make to the close analysis of whether the subsidies for wind power are an appropriate use of bill payers’ money.

  2. Michael Knowles’s avatar

    Chris & Prof Hughes

    Whoever is correct needs establishing beyond doubt as soon as possible by independent evidence -based analysis. We don’t want to wait 20 years to find .

    May be the wind turbine manufacturers and operators as well can produce their version of the analysis? With all the £billions we consumers are being made to pay to their industries, we deserve this surely?

    Mike CEng

  3. Dr Mike Patching CEng’s avatar

    I’m a wind energy engineer living and working in Cornwall for the last 15 years and pretty conversant with all the local wind farms which are some of the oldest in the country. Without giving away any commercial confidentiality I can state that this predicted level of reduction in output is completely at odds with the actual experience on the ground. As mentioned above academics just need to get out of their warm offices and actually talk to the engineers on the ground. Maybe even climb a wind turbine with an engineer ! But that would be a first and not just in wind energy! Much easier to play around on your computer trying to bamboozle everyone with complex statistical models with no real world experience at all. You can tweak statistical models to give you any answer you want and you need to look at who is funding this ‘research’. There are no subsidies without actual output so you can rest assured that those of us actually building wind farms have looked at this in detail and are completely confident that a well designed, constructed and maintained wind farm will have no problem delivering to its specification throughout its design life.

  4. Paul D’s avatar

    You only need one engineer who knows his theory and has at least fifteen years experience in his field!

    It is a pity that DECC do not employ them; even a handful would be a big improvement. Of course, DECC do employ one or even two but the ministers and civil servants prefer not to listen to him/them. All of this stems from the privatisation of the CEGB in 1990. Government lost it’s own design authority at that time and I notice that at least one engineering institution has proposed the re-establishment of exactly that – a new, central, design authority.

  5. Paul D’s avatar

    1: ‘Once again, I thank Dr Gordon Hughes for the contribution he continues to make to the close analysis of whether the [SUBSIDIES] for wind power are an appropriate use of bill payers’ money.’

    2: If I understand this Daily Telegraph article correctly, the SUBSIDIES are going to stop :-
    ‘EU Orders Britain To End Wind And Solar Subsidies
    The Daily Telegraph, 3 January 2014
    James Kirkup and Bruno Waterfield
    The European Commission is to order Britain to end wind farm subsidies. Officials have told ministers that the current level of state support for renewable energy sources must be phased out by the end of the decade. 
    … ‘

    This is all about electrifying heat. If you can reduce the amount of generation to drive heat pumps by a factor of ten and still do 90% of the job, then 90% of these wind turbines (plus their backup generators and grid reinforcement) are not needed at all.

    This policy of electrifying heat was conceived when heat pumps were big and only achieved a COP of 3 in the best cases. Pumps can now do 6 and a 1kW air-to-air can do 90% of the space heating in a standard retrofit scenario (27 million of those). A 1kW pump in each of 27 million houses requires 27GW of generation. Current maximum system demand 56GW. Current maximum wind generation now above 6GW (maybe 6.41GW according to National Grid) in the recent gales.

    By 2020, we will have more than enough wind capacity to run the experiment against an installed heat pump load based on 1kW per house. If mass market methane (mains gas) fuelled fuel cells (generating 1kW) take off, then the need for these wind turbines is drastically reduced. If these fuel cells are cost competitive with a replacement gas boiler, then there is a market of one million per year for these cells, just as ageing replacements alone. A million 1kW fuel cells installed per year is a generation capacity increase of 1GW per year. This fuel cell generation will REDUCE the demand on the electricity grid, particularly at the winter peak demand.

  6. Paul D’s avatar

    Mea Culpa:
    I should have read ALL the article, not just the headlines :-
    ‘The commission, which oversees the European single market, is preparing to argue that the onshore wind and solar power industries are “mature” … ‘

    What is the logic for removing the subsidy from onshore wind and leaving it for offshore?
    Put that another way. Why is onshore wind mature and offshore immature?

  7. Paul D’s avatar

    I am not a statistician, but …

    … I have opened ref_wind_production.xls as supplied from a certain website,
    I then sorted the data by year,
    I then summed the generation and capacities by year and
    produced the following data table :-
    Year_______Gen/Cap____line no of summation
    2002_______0.1714_________587_____
    2003_______0.1910________1461_____
    2004_______0.1943________2471_____
    2005_______0.1925________3772_____
    2006_______0.1827________5327_____
    2007_______0.1909________7167_____
    2008_______0.1834________9360_____
    2009_______0.1938_______11901_____
    2010_______0.1556_______14756_____
    2011_______0.2024_______17848_____
    2012_______0.2418_______18852_____
    Gen/Cap = Generation / Capacity = Load Factor
    The line number is the spreadsheet line number where the year changes.
    This data is for UK onshore wind turbines.

    The 2002 data excludes the data for Jan, Feb and March, so is not representative of a full year.
    The 2012 data is for the year to and including March, so is not representative of a full year.
    This exercise could be repeated for monthly results or for years based on April to March inclusive.

    The lowest load factor is 0.1556 in 2010 and the highest is 0.2024 in 2011 for a full year summation, January to December inclusive.

    I am not seeing any deterioration in the wind turbine load factors.
    Can anyone peer review my work? I would like to be sure about this.
    I have not tackled any of the clever stuff yet.

  8. Paul D’s avatar

    I have opened a new copy of ref_wind_production.xls and selected the sheet ‘ref_data_original’.
    The data is for UK onshore wind turbines.
    The summation of all ‘Mwh’ and ‘Ofgem IC(kW)’ data gives a load factor of 0.1879 for all turbines for all years.

    I added two columns for year and month functions to read the original date format and sorted on those, month first :-
    All results for each month
    Jan __________0.2471_____
    Feb _________0.1875______
    Mar _________0.2048______
    Apr _________0.1641______
    May _________0.1814______
    June _________0.1122______
    July _________0.1287______
    Aug _________0.1406______
    Sept _________0.1844______
    Oct _________0.2248_______
    Nov _________0.2444_______
    Dec _________0.2206_______
    Lowest wind speeds in summer and highest in winter.

    The January results for all years :-
    2003 0.3085
    2004 0.2635
    2005 0.3335
    2006 0.1857
    2007 0.3426
    2008 0.2618
    2009 0.2473
    2010 0.1831
    2011 0.1949
    2012 0.2949
    Not far off a max/min ratio of 2.

    Years by April to March inclusive :-
    2002/3 ___0.1909__
    “3/4 _____0.1915__
    “4/5 _____0.1975__
    ”5/6 _____0.1828__
    ”6/7 _____0.2030__
    “7/8 _____0.1898__
    “8/9 _____0.1815__
    ”9/10 ____0.1776__
    “10/11 ___0.1635__
    “11/12 ___0.2157__
    The key feature here is the last two data points.
    The regression line data points for the first and last years were 0.1918 (2002/3) and 0.1870 (11/12).

    I isolated the last two years’ data and sorted by identification code, year and month. This generates pairs of data points from the two different years. The two data points were subtracted to give a difference value which could be positive or negative. These data points for the same wind farm were summed and so was the capacity. The summed difference / capacity gives a load factor change between the 10/11 year and the 11/12 year. All the samples I took (this is a fiddling job putting in the data ranges; it really needs a macro) were in the range of +0.04 to +0.10. This suggests to me that the increase in load factor from 10/11 to 11/12 was widespread, if not universal. No negative values, indicating a deterioration in load factor were present in this limited sample, although a few turbines returned a negative.

    The data for 12/13 should be available but the Ofgem source reference is protected by a login screen and I cannot reach it. If the author of the paper would wish to explain the increase in load factor (0.2157-0.1635) = +0.0522 between the years 10/11 and 11/12, then I would be interested to see it. These two numbers are composed of the summation of thousands of other numbers from individual wind turbines, so something systemic must be happening here, in my view. The obvious explanation to me is a significant increase in wind speeds. This last data point in 11/12 is the highest of the ten years’ worth of data.

Reply

Your email address will not be published. Required fields are marked *

You may use these HTML tags and attributes: <a href="" title=""> <abbr title=""> <acronym title=""> <b> <blockquote cite=""> <cite> <code> <del datetime=""> <em> <i> <q cite=""> <strike> <strong>