Production Estimates

Production estimates for a new wind farm are critical to building a financial model, and of course obtaining financing.  But what is the track record of production estimates?  Where are the errors in estimates to be found?

One wind resource assessment specialist estimated that as many as 85% of projects overestimated their production.  Of course this estimate in itself was an estimate, and so of course is inaccurate.  Natural Resources Canada has complained that they have wind farms have overestimated production, and so funds for the Wind Power Production Incentive have been set aside that will never be used.
When building a wind farm, you first erect a meteorological tower, and measure winds for at least a year.  You need to measure for at least a year because of the seasonal nature of wind.  You want to measure the wind as close to hub height as you can, to improve accuracy.  Of course, you may have a small amount of measuring error, although good calibration of the equipment can reduce this.  You will prepare a “wind rose”, which tells you the predominate wind direction, and the percentage of power from each wind direction.  Data is collected in 5 or 10 minute increments, so an accurate production forecast can be made.  Outside investors will normally want the study conducted by a reputable outside firm, and sometimes will ask to have the data reviewed by a third party.

After you have measured for at least a year (2 years is better), then you conduct a correlation study with the nearest source of long term wind data – usually an Environment Canada weather station.  This is done to see if you were measuring in a good wind year, or a bad one.  The output from a wind farm can vary as much as 20% from year to year, depending on wind events.  You then adjust your estimate up or down, depending on the outcome of your correlation study.  Of course, Environment Canada stations are not designed to obtain data for the wind business.  They are weather stations.  Sometimes, their data is incomplete, or the wind is blocked in a particular direction.  Most of them measure at 10 m height, and of course your hub height is usually more than 65 m.  So there is as much art as science in a correlation study, and so there is error.

Once you have your estimate of wind, you then apply it to the power curve of the wind turbine you select.  A power curve is simply the manufacturer’s estimate of the production of the wind turbine at a particular wind speed.  Wind turbine manufacturers provide this data, and your estimated annual production is the sum of the outputs at each wind speed.  But while manufacturers go to great lengths to estimate power curves, and usually provide a warranty on the power curve, it is still just an estimate.

Then you estimate wake losses.  These are losses associated with an upwind turbine reducing the wind available to the down wind turbine.  There is good software to do this, but there is some error in the forecast, and it will only be as good as your wind rose, which is based on one or two year’s data, not the 20 year life of the wind turbines.
It is sometimes useful to estimate the wind shear using sodar.  A sodar sends a loud ping up in the air, and records the echo, which takes longer to return from a given height in higher winds.  This allows you to determine the difference in wind speeds at different heights.  The wind is speed is faster the higher you go, and today’s turbines may sweep the air from a range of 40 – 120 m, so the energy captured through that distance can vary by quite a bit.  Sodar can allow you to estimate your wind shear with greater accurace than just measuring at 40, 60 and 80 m.

You must estimate downtime for equipment failure, loss of grid or grid problems like brownouts, substation problems, icing events, blade degradation from dirt and insects reducing the blade efficiency etc.  Blades sometimes require repairs due to lightning damage, and the equipment cannot operate until the repair is done.  Wind turbines shut down in high winds (75-90 km/hour), and so the extent of these events must be estimated.

The roughness class of the area is estimated, as trees and buildings cause turbulence, which has less power.

Wind has a different amount of power depending on its density, which is a function of temperature and air pressure.  So you need to adjust for the averages in your area.  Usually these adjustments are done based on data from a nearby weather station, as it is highly unusual to gather both temperature and pressure data at your meterological tower.  But applying an average doesn’t adjust for some small factors, such as the possibility that the wind blows more when the air is dense, or when it isn’t.  Again, error is introduced.

Another loss you will have is in transformers.  Turbine manufacturers usually provide power curve data at the turbine generator, which is at low voltage (600 V or so).  To attach to the grid, you must boost the voltage, and your transformers consume some power, and so you have losses.  These losses are again just estimates, as your losses as a percentage of your output are higher at low power levels than high, and so the loss estimates are a function of the accuracy of your power output.  The wind farm collector system will also have small losses.  Transformer and collector losses will be between 1 and 2%.

There is one key variable that I don’t think anyone can estimate properly.  And that is when the downtime occurs.  Wind is variable, and the output of a wind turbine varies between zero, and its rated capacity.  Rated capacity is achieved when winds are 40-50 km/h, depending on the model of turbine.  But when will turbines fail?  And when will the grid fail? Both will occur far more during high wind events.  So if you have downtime due to equipment and grid failure of only 5%, you may lose 10% of your power output for the year, or more.  I have seen no model to provide a good estimate of this impact.  In June, the Ferndale Turbine produced only 60,000 kWh, or 10% of its rated capacity for the first 2 weeks of the month.  From June 15-20, the turbine produced 121,000 kWh, or 46% of its rated capacity.  The wind was much stronger.  Wind turbines and grids tend to fail only when winds are strong, so operating during wind events is critical to overall annual production.

Financiers and lenders are well aware of the risk associated with expected production.  So, to cover their risk, many of them use “P90″ as the maximum amount they will lend.  P90 is a probabalistic estimate that accounts for the expected errors in the overall estimates.  Essentially they prepare a probability distribution, where P50 is the midpoint.  P90 is the level where the estimate is that the wind turbine will produce power above that amount 90% of the time.  Of course, estimating the probability curve is just an estimate too.  I wonder if the financial community uses P90 more to claim they have done due diligence, and exercised caution – I hope they don’t really believe a P90 estimate any more than a P50.

Once a wind farm is built and operating for a couple of years, the estimates become much firmer.  Measuring error goes away.  Wake effects become real instead of just models.  Roughness estimates convert to real production data.  Grid conditions, icing events, downtime etc. become real live operating experience.  At this point, the estimated production from a wind farm becomes much firmer.  The P90 moves closer to the P50.  Perhaps at this point you can renegotiate a higher amount of leverage on the wind farm, as the uncertainty of output is reduced.

The risks in production estimates are significant.  But understanding them, and estimating them, will ensure you can put together a project that is viable financially.

Leave a Reply