A New Toolbox for Analyzing Wind Plant Data

In order to confirm performance of turbines at a particular site, wind farm developers employ power curves for individual turbines, which rely primarily on hub-height wind speeds measured by a single meteorological (met) tower. Many wind plants underperform due to the effects of atmospheric conditions such as turbulence, shear, waking, and spatial variability in resource. Once the wind plant is built, it is possible to develop an equivalent plant power curve, which is useful for forecasting power production. In contrast to an individual power curve, an equivalent plant power curve indicates performance of the plant as a whole and considers interactions among the turbines. The development of an equivalent plant power curve requires processing and synthesis of large amounts of data from an operational wind farm.

    In this project, a new Python toolbox was developed to ingest data from an operational wind farm and filter the data for use in a variety of applications. As an example application of the toolbox, data from met towers and the SCADA system of a wind farm in the Southern Plains of the United States were used to develop equivalent plant power curves. These data were then used to train a random forest model to predict the response of the full plant to turbulence, wind shear, and wind direction.