Wind Power Forecasting Using Ensembles

Short-term prediction of wind energy is by now an established field of wind power technology. For the last 15 years, our groups have worked in the field and developed short-term prediction models being used operatively at the major Danish utilities since 1996. The next step in the development of the models is to use ensemble forecasts. The ensembles are produced routinely at US NCEP (National Center of Environmental Protection) and at ECMWF (European Centre for Medium Range Weather Forecasts). Both are collected since late 2002, and are investigated for their properties in regard to wind speed and power forecasting for Danish wind farms and meteorology masts. Additional ways of creating ensembles of forecasts are to use ensembles based on lagged initial condition, formed from the overlapping forecasts created at different starting points, and multi-model ensembles, which stem from using the output of different models, preferably with different input models as well. In our project, we use the operational model runs of DMI-HIRLAM and DWD-Lokalmodell. The idea behind the use of additional input for short-term forecasting is the connection between spread and skill of the forecast. The basic assumption here is that if the different model members are differing widely, then the forecast is very uncertain, while a close model track means that this particular weather situation can be forecasted with good accuracy. Landberg et al have shown that this is not necessarily true for Poor Man's Ensembles (defined as ensembles based on lagged initial condition), when just using the spread for a single point in time, but Pinson and Kariniotakis, and Lange and Heinemann showed that using the temporal development of the forecasts, some assertion can be made on the uncertainty. The paper will start with a literature overview on ensemble wind power forecasting. Then, the current Danish project will be described, including first results. Two additional papers on this conference point out the more technical aspects of the work. The project is funded by the Danish PSO funds under the reference no. ORDRE-101295 (FU 2101).