Time-adaptive quantile-copula for wind power probabilistic forecasting

This paper presents a novel time-adaptive quantile-copula estimator for kernel density forecast and a discussion of how to select the adequate kernels for modeling the different variables of the problem. Results are presented for different case-studies and compared with splines quantile regression (QR). The datasets used are from NREL’s Eastern Wind Integration and Transmission Study, and from a real wind farm located in the Midwest region of the United States. The new probabilistic prediction model is elegant and simple and yet displays advantages over the traditional QR approach. Especially notable is the quality of the results achieved with the time-adaptive version, namely when evaluated in terms of prediction calibration, which is a characteristic that is advantageous for both system operators and wind power producers.

► Description of a novel time-adaptive quantile-copula estimator for kernel density forecast.
► Discussion of how to select the adequate kernels for modeling the different variables.
► Evaluation for two real datasets of wind generation.
► Flexible representation of wind power uncertainty.
► More reliable wind power probabilistic forecasts.