Spatially seamless forecasting of wind power generation with generalized power curves

WAIMANN, C.; PINSON, PIERRE; SAULO, C.

Montreal

Conferencia; World Weather Open Science Conference; 2014

Denmark has the world largest wind power capacity in proportion to the size of the electricity consumption and in 2013, covered 33.2% of its electricity demand by wind power. It also has a long tradition in forecasting wind power generation for optimal power system operations. A challenge is to build computationally-efficient and high-quality forecasting methodologies for a large number of sites based on a combination of numerical weather predictions (NWP) and local observations, for lead times up to 48-72 hours ahead.The aim of this research is to examine different multi varying linear models for wind power prediction, describing their particularities and testing their performance. A spatial generalization of wind power prediction models is proposed, permitting to predict wind power generation at a large number of sites with a single model. Model coefficients are seen as spatial stochastic processes estimated by ordinary Kriging. Such an approach may also allow to readily provide forecasts at any new location where turbines are deployed. The ideas and methods are demonstrated on the test case of Denmark, based on a set of more than 200 wind farms. We observe that the usage of wind speed forecasts at 10m above ground level yields more accurate results than the multi varying model using wind speed forecasts at 100m. Besides, the spatial modeling approach provides a good tool to start issuing forecasts, particularly in coastal areas, where predictions exhibit a significant quality improvement with respect to forecasts made using a unique (average) power curve for Western Denmark