Lueder von Bremen

Enhanced regional forecasting considering single wind farm distribution for upscaling

With increasing wind power penetration the need for more accurate wind power forecasts increases to raise the market value of wind power. State-of-the-art wind power forecasting tools are considered either statistical or physical. Fundamentally new techniques are rare, thus it is tried to establish a new approach. The spatial decomposition of wind power generation in Germany can be done with principle component analysis to extract the main pattern of variability. They have a physical meaning when linked with typical weather situation. The first four eigenvectors explain about 94 % of the observed variance. The time-evolving principle components are linked with the total wind power feed-in in Germany and are used for its estimation. A new wind power forecasting model has been implemented with this approach and shows very good results that are comparable with state-of-the-art commercial wind power forecast models. The day-ahead forecast error for a common intercomparison period Jan-Jul 2006 is 4.4 %. The suggested approach offers wide ranges for future developments (e.g. several NWP models), because it is computationally very cheap to run.

Wind power forecast error smoothing within a wind farm

Smoothing of wind power forecast errors is well-known for large areas. Comparable effects within a wind farm are investigated in this paper. A Neural Network was taken to predict the power output of a wind farm in north-western Germany comprising 17 turbines. A comparison was done between an algorithm that fits mean wind and mean power data of the wind farm and a second algorithm that fits wind and power data individually for each turbine. The evaluation of root mean square errors (RMSE) shows that relative small smoothing effects occur. However, it can be shown for this wind farm that individual calculations have the advantage that only a few turbines are needed to give better results than the use of mean data. Furthermore different results occurred if predicted wind speeds are directly fitted to observed wind power or if predicted wind speeds are first fitted to observed wind speeds and then applied to a power curve. The first approach gives slightly better RMSE values, the bias improves considerably.

The role of predictability in the investment phase of wind farms

Abstract Within the SAFEWIND project, wind power predictability is evaluated to suggest potential benefits during the planning phase of wind farm deployment. Predictability assessment can be useful, for instance, to anticipate operational costs due to market imbalance for energy traders, costs related to prediction of offshore weather windows for turbine installation or maintenance, or to benefit from portfolio effects when aggregating wind farms that operate in uncorrelated wind climates. While the weight of predictability for a wind farm developer is low in the decision-making process compared to the capacity factor, it remains a useful indicator of the quality of wind for grid integration purposes. A series of case studies are presented in various European regions to present tools for wind power predictability assessment at different scales.