Forecasting of regional wind generation by a dynamic fuzzy-neural networks based upscaling approach

Pierre Pinson (1) , Nils Siebert (1) , Georges Kariniotakis (1)

1 CEP - Centre Énergétique et Procédés

Short-term wind power forecasting is recognized nowadays as a major requirement for a secure and economic integration of wind power in a power system. In the case of large-scale integration, end users such as transmission system operators focus on the prediction of regional or even national wind power up to 48 hours ahead. At a European level such predictions will be required in the future for planning power exchanges between regions or countries. The main difficulty for predicting regional wind power is that on-line information is not available for all concerned wind farms. Predictions have to be based on a limited number of representative wind farms for which SCADA data and/or Numerical Weather Predictions are available and then extrapolated (“upscaled”) to predict the total wind power. In this work several approaches were developed for upscaling ranging from simple to more complex ones (i.e. based on artificial intelligence methods such as fuzzy-neural networks). Evaluation results are provided for the case of the Irish power system. Predictions for the output of eleven wind farms are made from a number of one up to five representative wind parks. The performance of the various approaches is evaluated using one year of data. Useful conclusions are derived for the impact of the “smoothing effect” on the performance of Persistence and that of advanced models.

Mots clés: Wind power; short-term forecasting; upscaling; adaptive fuzzy-neural networks; regional forecasting

Domaines: Sciences de l'ingénieur [physics]; Energie électrique

Pierre Pinson, Nils Siebert, Georges Kariniotakis. Forecasting of regional wind generation by a dynamic fuzzy-neural networks based upscaling approach. EWEC 2003 (European Wind energy and conference), Jun 2003, Madrid, Spain. 5 p. - CD ROM. ⟨hal-00530550⟩