Greenbyte Documentation

Learned Power Curve

The learned power curve is calculated daily. On the first day of every month, the current learned power curve is saved for future reference. Greenbyte uses data from the most recent month to generate the power curve.

  • A valid power curve has a power value from the wind turbine's cut-in wind speed for each 0.5 m/s up to at least 15 m/s.

  • The minimum data set needed to generate a learned power curve is 10-minute data for wind speed and power.

  • If available, the default power curve, status events, and minimum power measurements (per 10-minute period) are also used.

If it is not possible to generate a valid power curve using only one month of data, another month’s data is added until it is possible (up to a maximum of 12 months of data).

Before the power curve is calculated, the wind speed and power data is filtered to only include data that represents normal operation. Unwanted and deviating data is removed. The filters used are:

  • Filter out periods with stops or curtailment – Data during status events might deviate from normal operation.

  • Filter out periods with high and low wind speeds – Data above/below cut-out/cut-in wind speed is not interesting.

  • Filter out periods with low power – Data with power lower than 20% of the default power curve is deviating from normal operation.

  • Filter out periods with start-up and shut-down – Data with minimum power of 0. A minimum power of 0 during a 10-minute period means that the turbine was stopped at some point during the 10-minute period. A start or stop event is not normal operation.

  • Filter outliers – Target outliers in the data to remove deviating data. The outliers are found using the interquartile range (IQR) and a lower fence of Q1-0.8×IQR  and an upper fence of Q3 + 0.8 × IQR.

Note

Note that the time with active warnings is not filtered out.

When the data has been filtered, a best-fit line is drawn through the remaining wind speed and power observations. The best-fit line is found by taking the median of the power observations for each 0.5 m/s wind speed bin. If there are any single gaps in the best-fit line, they are filled using linear interpolation. Gaps above 15 m/s are filled using the previous wind speed bin's power value.