Power Curves

Wind turbine power curve analysis

The purpose of this repository is to provide code, papers, and sample data for power curve analyses.

Bootstrapping Power Curves

The following is an outline of an outline of a paper still to be written.

Data

• Mean hubheight wind speed, U.
• Shear exponent, alpha.
• Turbulence intensity, I.
• Mean power, P.
• Standard deviation of power, sigma_p.

Methods

• Method of binning, MB.
• Machine learning, ML.
• Quasi static, QS.

A method maps a dataset to a model. A model maps the triple (U, alpha, I) to the pair (P_pred, rho_pred) consisting of the predicted mean power P_pred and one or more measures rho_pred of dispersion of the mean power, e.g. the sample standard deviation of several suggested mean powers used interally in the model.

Part I: Synthetic data

1. Compare predicted dispersion with sample dispersion. The purpose is to give the reader some faith in the models' ability to provide a credible measure of dispersion. Divide the parameter space of U, alpha, and I into small regions that still contain a decent number of points. For each method, obtain a model from the entire dataset. For each point in each region, obtain a prediction of the mean power. From these predictions of mean power, obtain a sample-based measure of the dispersion of the predicted mean power. Compare this dispersion with the dispersion obtained from a single representative point in the region, e.g. the center.

2. One training dataset and one verification dataset. Make a training dataset and a verification dataset. For each method, obtain a model from the training dataset. From U, alpha, and I in the verification dataset, predict the mean power and form the error epsilon = P_pred - P_true, where P_true is the corresponding mean power from the verification dataset. Make various plots of epsilon and rho_pred, merging U and alpha into the rotor equivalent mean wind speed, W, which is the same as the mean driving wind speed in the quasi static model.

3. Bootstrap point 2 to obtain a distribution of errors. The mean of these errors is the bias and it should be small. Make plots as appropriate.

Part II: Field data

• Obtain field data.
• Repeat part I using the field data.