Fraunhofer researchers win wind forecasting competition

In the EU research program WIRE: „Weather Intelligence for Renewable Energies“, within the European Cooperation in Science and Technology (COST) algorithms of the Fraunhofer IWES generated the best wind park forecasts in a worldwide forecasting benchmark.


Due to its close cooperation with the Fraunhofer IWES, enercast is the single provider of forecasting services, to dispose over Fraunhofer IWES’s state of the art research results in the field of power forecasts. The latter are exclusively utilized in the enercast algorithms and offered to direct marketers, electricity traders and network operators.


Participants from Europe, Japan, India, USA and Australia took part in the benchmark test which was initiated by WIRE COST Action in 2013. The focus was on the performance of individual forecasting algorithms, which are employed to transform the meteorological input variables into the expected power production of the wind park. All participants received the same data sets on historical local weather forecasts and power output measurements of two wind parks. However, only the weather forecasts were provided for the assessment period. The task consisted of forecasting the performance of two wind parks for the given time period. Two wind parks with particularly extreme local attributes were selected, in order to draw conclusions about the applicability of the various forecasting models in different orographic conditions. Both wind parks are located in Italy. The first one disposes over an installed nominal power of 104 MW and is situated in complex terrain. The second one incorporates a plant portfolio of 21 MW and is located in level country.


In order to model the connection between meteorological input variables from the weather forecasting model, and the expected power production of the wind park, the researchers of Fraunhofer IWES are using artificial neural networks. The networks’ topology and learning mechanism were especially tailored to this task. „The use of artificial neural networks for wind power forecasts was already realized in 2001 by the Fraunhofer IWES and heralded a new era in this field. “, stresses Dr. Rohrig. At this juncture, the specific focus lies in the processing of available data. This includes a comprehensive analysis of the existing historical performance measurements and an optimized preparation and selection of the model input variables.


The organizational committee of COST Action evaluated the power forecasts of all participating institutions, for all forecasting horizons, based on the original weather forecast. These comprise lead times of 1 to 48, or up to 72 hours, depending on the type of weather model that was employed.


The results of the benchmark test reflect the IWES’s competence for the generation of power forecasts and statistics- based modeling.


Due to its close cooperation with the Fraunhofer IWES in the field of power forecasts, enercast is the single provider of forecasting services, to use the distinguished algorithm to generate power forecasts for network operators and electricity traders.