Accounting for the effect of turbulence on wind turbine power curves

Wind turbines require methods to predict the power produced as inflow conditions change. We compare the standard method of binning with a turbulence renormalization method and a machine learning approach using a data set derived from simulations. The method of binning is unable to cope with changes in turbulence; the turbulence renormalization method cannot account for changes in shear other than by using the the equivalent wind speed, which is derived from wind speed data at multiple heights in the rotor disk. The machine learning method is best able to predict the power as conditions change, and could be modified to include additional inflow variables such as veer or yaw error.