February 15, 2011 | A new NCAR study compares two different methods for forecasting power production at wind farms: turbine-based versus farm-based.
The turbine-based method involves forecasting winds at each individual turbine on a given farm, converting the forecast to power, and summing the results to produce an overall forecast for the farm. The farm-based method involves making a forecast of mean wind for the entire farm and performing the wind-to-power conversion by modeling farm power against winds.
As part of an agreement between NCAR and Xcel Energy, NCAR makes wind forecasts for numerous wind farms with a total of roughly 3,500 turbines. Scientist Julia Pearson and colleagues wanted to see if it would be possible to simplify the forecasting system by making farm-based forecasts rather than forecasting for each turbine, as they currently do. They ran an experiment in which they created models based on the two approaches and then made wind power forecasts using the same wind input data over a four-month period.
The study’s results show that while both methods can be applied successfully, they have different strengths and weaknesses that may benefit different types of forecast systems.
One advantage of the turbine-based approach is that it can be applied at all wind farms and enables new or expanding farms to issue power forecasts immediately. It is also less sensitive to data quality control issues. A disadvantage is that making turbine-level wind forecasts requires a more complex forecasting system than forecasting for an entire farm.
In general, the farm-based approach performs slightly better at short-term forecasts than the turbine-based approach, with the error rates for the two methods converging for forecasts with longer lead times. Farm-based models can only be applied to farms with wind data, however, and because the forecasts are more sensitive than turbine-based models to quality control issues, it may be necessary to also run a turbine-based approach.