Nonparametric Wind Power Forecasting under Fixed and Random Censoring

We consider nonparametric forecasting of wind power for individual wind turbines, allowing for random right censoring as well as two-sided fixed censoring. We propose a very fast estimation algorithm and show that this estimator of the unknown regression function is uniformly consistent. We argue that the key statistical features of the proposed nonparametric regression framework, such as nonlinearities and fixed and random censoring, are all needed in order to properly capture the main characteristics of wind power production functions. We show by a simulation study that the asymptotic properties of the estimator also holds in finite samples. We provide an empirical illustration comparing the forecasting accuracy of the proposed nonparametric regression model to some of the existing and popular forecasting devices used for predicting short to medium term wind power production at the individual turbine level. The empirical results are generally very encouraging.