European University Viadrina, Germany
Keywords: Censored Regression, Wind Energy, ForecastingPrecise short term forecasts (one to 36 hours ahead) are crucial in the wind energy industry for all market participants. Power producers as well as power distributors (energy traders) require reliable forecasts to achieve market clearing and fair pricing in the power markets. Transmission system operators (TSOs) are also in need of efficient forecasts of the power production since they have to manage the network load. We present a forecasting model that focuses on the non-linear relationship between wind speed and wind power production, called “power curve”. We utilize wind direction as an explanatory variable and observe the non-linear relationship as a two-sided censored set of data. Therefore, our model uses an important additional ex-ante available information: The power range of the turbine(s) that determines the two censoring points. Our model returns an efficient and unbiased forecast of wind power production that is more precise than models currently used. It should be emphasized that the model mainly aims at running at a turbine specific level, but can also be used at a more macro-oriented perspective WLOG, e.g. at a wind park level including several turbines.