Wind Turbine Performance: A Dance of Data
Image by liuzishan on Freepik
We live in a world of increasing specialism and that means a world where we increasingly don’t understand other people’s jobs (if they’re not in the same industry as you). I’m often asked by relatives to explain what I actually do. They are from a farming background and my day-to-day activities are a world away from their experiences. I tell them - wind energy data analytics. Ok - but what is the data and why do you analyse it? The ‘what’ (measurements from operational wind farms) is easy enough to explain, but the ‘why’ and the ‘how is this valuable to anyone’ needs a little more elaboration and a patient audience!
The way I put it is this – if you think of understanding the performance of a power plant, you can understand the performance of the plant by looking at the power outputs relative to the fuel input. More power for less fuel and your power plant is more efficient or better performing. With wind turbines, our measurement of the “fuel” is typically very bad – we have 3 cups on a stick called an anemometer placed right in the place that there is most influence from the turbine itself (behind the rotor). On top of this, the fuel for a wind turbine is not a single point wind speed measurement at hub height, but instead a large and complex mass flow field comprising of varying shear, veer, turbulence and air density.
This is not to take away anything for the anemometer manufacturers and installers who have spent years improving and calibrating these to make them as good as possible. We just don’t really have any better affordable options. There are devices that can measure the wind more accurately such as hub-mounted ultrasonic anemometers and nacelle mounted Lidars. However, these are both expensive and still somewhat imperfect. Even if you were able to measure the mass flow field perfectly, to derive how much energy the turbine should optimally produce is not easy.
This presents such an issue for measuring wind farm performance, that it’s not easy to determine if a turbine is operating to even within 5% of optimum. This means that quite a bit of lousy performance can go undetected. There’s a litter of issues that frequently affect turbine performance that fall within this – static yaw misalignment, aerodynamic imbalance, mistuned controller settings to name a few. And then what about performance upgrades? There are many turbine upgrade packages available on the market that promise performance gains in the region of 1-5%. So if an owner wants to try an upgrade out on one turbine and check if it really worked before applying it on the whole fleet, how can they validate it?
The answer is that we need some data analytics and data science gymnastics. We have a few methodology groups available to us here:
If there’s a meteorological mast near the farm, its wind speed measurements (still using anemometers, but due to their location they’re a lot more accurate than those on nacelles) can be used to validate the performance of a few nearby turbines. Unfortunately, there are very few turbines that this can be applied to.
Foregoing the wind speed measurements all-together and doing turbine-to-turbine power comparison, accounting for factors such as wakes and terrain.
Using machine learning to learn power trends from a combination or choice of wind speed, power and other measurements throughout the farm.
Bitbloom’s primary approach so far been on the second method but with an increasing emphasis on bringing in machine learning to our approaches. On a site where the terrain is not too complex, our methodologies can get that 5% uncertainty down to below 1%. That’s still not perfect and not the kind of accuracy a nuclear power plant manager would accept, but it gets us a lot closer and allows us to validate performance for most purposes.
Of course, power performance is not the only role for data analytics in wind energy, and I’ve also glossed over the importance of data cleaning, flagging and filtering that is a prerequisite to be able to do any of this. But power performance and its related topics are behind a significant proportion of the data analytics on wind energy operational data, and all because it’s so difficult to measure our wild and natural fuel. With the right approach, what we lack in good measurements we can make up for in cutting-edge data science and analytics. To ensure the most valuable data analytics, it’s vitally important for the wind industry to continue to invest in it.
Find out more in our latest whitepaper - download now - and get in touch to speak to the Bitbloom team.