Reducing the number of wind turbine prototype measurement campaigns for power curve model validation using a model-validity predictor

Prototype turbine measurement campaigns are required for turbine type certification and the associated validation of aero-elastic wind turbine models when wind turbine designs slightly differ from previous measurement campaigns. Small changes in the turbine design are allowed without additional testing campaigns being required. We propose a model-validity predictor that is able to conservatively predict the validation results on new turbines before a measurement campaign is carried out. This predictor is trained based on previous measurement campaigns and uses a set of variables to describe the wind turbine design space. The model-validity predictor demonstrates the potential to reduce the number of prototype campaigns required to validate a model, and it is shown that a change of variables can expand the predicted safe domain by 72%. Detailed uncertainty propagation is carried out for each wind turbine measurement campaign using Monte Carlo simulations. We separate aleatoric uncertainties and parameter epistemic uncertainties. This separation enables us to quantify the degree of validity of a model with respect to the measurements. This study demonstrates the potential benefits of calibration, with the qualification that the ideal workflow will separate data used for validation and calibration, perhaps using a sophisticated cross-validation routine.