Researchers have created an algorithm that is showing to be 5 to 10 percent more accurate at forecasting wind speeds – potentially a key development for energy companies struggling with how to improve the reliability of, and justify investing in, wind power.

The new data is a result of testing the new method against three other algorithms using datasets from three different wind farms in Wisconsin. The research team is hoping to further validate initial results; if they’re successful, it could mean significant savings for power companies looking at wind-forecasting technologies to help keep energy supply as close to demand as possible to save resources. These systems are used to determine when back-up sources of energy like diesel or batteries are needed. The problem, of course, is that wind forecasting can be a very inexact science. Most forecasting systems are based on prediction intervals, which provide an estimate of where future observations (wind speeds) will fall with a certain probability, based on data from previous years.

Because it’s impossible to always exactly predict wind speed before it happens, the researchers felt a prediction method was needed that also included a prescribed confidence level, a percentage that demonstrated the likelihood that the wind would be at a specific speed at a certain point of time. In an article in IEEE Transactions on Power Systems, they detail developing a prediction interval which includes a lower and upper bound that not only provides a range of target values within the two bounds, but also provides an indication of their accuracies.

Prediction intervals are extremely difficult to calculate, so the team split their wind power series into an algorithm with three components to simplify the problem. Using the extreme learning machine technique, they perform the prediction on the separated components and combine them to find the optimal prediction interval.

6936937-fig-1-large

 

Lead researcher Herbert Ho-Ching Iu and his team plan to collect additional data from wind farms in Australia and test their developed approach again for further verification. Initial results, though, show the method could potentially save power companies some serious money in the long run.

“Some power companies are hesitant to invest in wind energy because of its potential to provide inconsistent amounts of power,” said Herbert Ho-Ching Iu, lead researcher of the project. “Our approach helps make wind energy an effective, more reliable power source and it can be used right away. We hope to see our research put to use in the near future.”

You can also find more articles about “Prediction Intervals” in IEEE Xplore.