On-line adaptation of confidence intervals based on weather stability for wind power forecasting

Pierre Pinson (1) , Georges Kariniotakis (1)

1 CEP - Centre Énergétique et Procédés

Existing literature or tools for wind power forecasting do not consider online estimation of confidence intervals for the output of the forecasting models. Uncertainty is estimated either based on error estimations coming form weather forecasts or on the inappropriate assumption that the error distribution is Gaussian and the intervals symmetrical around the spot predictions. This situation reveals the necessecity to develop formal methods for on-line uncertainty estimation adequate for the problem of wind power forecasting. The paper introduces an advanced method for this purpose. The aim is to compute intervals for wind power forecasts with a confidence level defined by the end-user. The intervals are derived after an analysis of wind power prediction error characteristics. The error distribution parameter is estimated in an adaptive way after appropriate exploitation of past errors and using fuzzy set modeling. Then, an index named as MRI, expressing the expected weather stability is used for fine-tuning the intervals. This index reflects the spread of poor man's ensemble weather forecasts. A relation between the MRI and the level of power prediction error is shown: the linear trend is used for narrowing the intervals when weather situation is considered as stable. Evaluation results of this methodology over a three-year period on the case study of a Danish wind farm and over a one-year period on the case study of nine Irish farms are given. The proposed methodology has an operational nature and can be applied to all kinds of wind power forecasting models.

Mots clés: Wind power; short-term forecasting; confidence intervals; weather stability; on-line software; numerical weather predictions; ensemble forecasting; uncertainty; energy meteorology

Domaines: Energie électrique

Pierre Pinson, Georges Kariniotakis. On-line adaptation of confidence intervals based on weather stability for wind power forecasting. Global WindPower Conference, Mar 2004, Chicago, United States. ⟨hal-00529488⟩