By splicing together two models describing normal and rare extreme conditions, KAUST researchers have developed a method to predict the frequency of winds strong enough to shut down wind turbines—even if such winds haven’t yet appeared in observations. The approach could help design and place wind farms in regions with more favorable wind energy resources.
Wind power is rapidly providing a significant proportion of the world’s energy mix and—when wind farms are located in windy areas—can provide relatively reliable power. The average power output of a wind farm at a given location is typically predicted using the average wind speed over time, and there are many mathematical models for this purpose.
However, wind turbines can be damaged when wind speeds exceed their engineered limit, causing them to shut down by turning or “feathering” the blades when the wind becomes too strong. Yet although wind farm outages due to high wind speeds could be a critical factor in the design and placement of wind farms for optimal power output, predicting the frequency and intensity of such events is far more challenging than the estimation of averages.
Daniela Castro-Camilo and KAUST colleagues in Raphaël Huser’s group have now developed a method for predicting both average and extreme wind speeds, even if such high winds have never been measured.
“Our model is designed to provide forecasts beyond observed wind speeds,” says Castro-Camilo. “Our approach essentially corrects the extreme ‘tail’ in the data distribution, which is useful for obtaining better predictions for parameters like wind speed.”
The researchers combined two models, one designed to predict average wind speeds based on the variations under normal wind conditions and one that is mathematically suited to predicting very rare extreme events.
“Wind speeds are strongly fluctuating and intermittent and so it is not easy to come up with a model that can capture these aspects,” says Castro-Camilo. “We tested many different models and found that a single mathematical model was unable to describe the key features of our data. By splicing these two models together, we realized a new way to tackle the situation.”
As well as being able to provide valuable insights that could be used to optimize wind farms, the method uses a fast computational approach and is flexible enough to be used for other types of data.
“As a general framework, our method can also be applied in different contexts, such as hydrology, by modifying the distribution for normal conditions to suit the data,” says Casto-Camilo.