WIND FORECASTING

The keys to an accurate forecast of wind energy output are high quality and reliable current and historic data from turbines, on-site and off-site anemometers and the entire project. The uncertainties in the physical/numerical weather modeling influence the accuracy of the forecasts. However, there are other project-related sources of inaccuracy such as wind speed sampling error and turbine power curve estimation. The impact of these and other issues must be quantified accordingly.

Also, reliable communication channels (such as SCADA systems) are necessary to get all the necessary data to the forecast provider in a reliable and timely fashion. The provider must also ensure that these forecasts are provided to the wind farm owner in a timely and reliable manner.

Finally, a clear statement of the forecast objectives by the project’s owner is crucial together with an understanding of the trade-offs of accuracy versus computational expense and data requirements.

Methods for Short-range Wind Forecasting

Statistical methods based on observations are the foundation of short-range forecasts, with a timeframe from minutes to hours into the future. The first phase in developing this type of forecast consists of identifying, compiling and integrating regional assets (location of turbines and anemometers, available records, etc).

The second phase consists of developing and training various self-learning forecasting methods using all available data. The final product provides a timely, relevant and accurate forecast for making the best operational and marketing decisions.

Users of short-range forecasts should expect hour-ahead forecasts to be 10 to 30 percent better than persistence when averaged over one month. However sometimes, especially during low power generation periods, persistence may outperform sophisticated forecasting methodologies.

Methods for Medium-range wind forecasting

Numerical simulations of the weather prevail in medium-range forecasts, with a timeframe of days into the future. The development of a day ahead forecast system consists of identifying, configuring, optimizing and operating a series of numerical weather prediction systems, followed by statistical post-processing and power curve application to translate the numerical-simulated winds into energy output from the project.

Timely and relevant forecasts are presented to the users in graphical format and as input to optimization and Energy Management Systems.

Numerical weather simulations include a hierarchical overlap of grids, from a wider, coarse resolution grid to finer resolution nests covering the area of interest. Finer grids are usually of resolutions from 1 km to 5 km depending on the complexity of the local terrain.

Users of medium-range forecasts should expect numerical weather prediction models to perform better than persistence for time periods beyond four hours into the future. Typically, day-ahead forecasts have errors that are half of those of forecasts based on climatology or persistence.

Statistical models prevail in long-range forecasts, with a timeframe of weeks to months. The inputs for the models are based on forecasts or observations from various climate oscillations. Various adaptive or regressive statistical methods are developed to create a monthly deviation-from-normal forecast.