Forecasting method AVDE

The ECN wind power forecasting method AVDE is a physical forecasting method with an output statistics module [Lange & Focken, 2005]. In an operational sense it is a post-processor to the atmospheric model HiRLAM, or any weather prediction model that delivers the required input data (two horizontal wind speed components, temperature and pressure in two vertical levels on a horizontal grid covering the sites to be considered) in the required format (GRIB). If wind speed and/or wind power realisations are available, the output statistics module can be employed in order to compensate for systematic errors in the forecasts. Technical aspects and application of AVDE are described separately [Brand & Kok, 2003][Brand, 2008].

AVDE can be operated in two modes: power forecast and wind forecast.

• In the power forecast mode AVDE delivers the expected value of the 15-minute averaged power of a solitaire wind turbine, wind turbine farm or cluster of turbines and/or farms at 15-minute intervals. In addition it gives the confidence interval of the power and the expected value of power variation intensity (ratio of wind power standard deviation and average), plus the expected value of wind speed, wind direction and air density at turbine hub height. All power values are based on the expected value of the air density.
• In the wind forecast mode AVDE gives the expected value of 10 or 15-minute averages of wind speed, wind direction, Obukhov stability length, temperature and pressure as well as turbulence intensity (ratio of wind speed standard deviation and average) at the given height above ground or mean sea level at 10 or 15-minute intervals up to 48 hours after initiation of the underlying run of the atmospheric model.

The feature of the forecasting method AVDE is that it takes the local influences of roughness, obstacles and stability into account. For sites with fetches less than 10 km open water upstream surface roughness length is determined by using 0.1x0.1 km² patches from the roughness map of the Netherlands developed by the Dutch met office KNMI [Verkaik, 2000]. For the other sites upstream surface roughness length is determined from the wind speed by using the Charnock relation. Upstream obstacles are modeled by employing a direction dependent displacement height. The effect of atmospheric stability is determined by using the Holtslag speed and temperature profiles (in moderate to very stable conditions) or the Businger-Dyer profiles (in the other conditions) in combination with the Obukhov length determined from the speed difference and temperature difference in two vertical levels. As to power forecast AVDE employs the directional standard (that is: normalised to the standard value of the air density) speed-power curve of a wind turbine or wind farm.

Data from HiRLAM

The High-Resolution Limited Area Model HiRLAM is an atmospheric model operated by KNMI (and other met offices). Like all atmospheric models HiRLAM is a numerical approximation of the physical description of the state of the atmosphere in the near future. The daily runs are initiated at 0, 6, 12 and 18 h Universal Time, starting with initial conditions originating from recent measurements. The output consists of the expected average value of physical quantities like wind speed at various vertical levels up to 30 km above mean sea level in a horizontal grid, to date with a size of 0.2x0.2 deg2, covering Europe and vicinity and with a time step of 1 or 3 hour up to 48 hours after initiation.

The hierarchy of the consecutive runs is as follows:

ECN obtains the HiRLAM data from KNMI under agreement 2001/265 for the daily delivery of meteorological data, and has been maintaining a database with HiRLAM data files since June 2001.

Compensation for systematic forecast error

The systematic forecasting error is a difference between the average measured power and the average forecasted power. Usually, this error is site dependent. If sufficient data in the range between zero and nominal power is available, the forecasts can be compensated for the systematic error. To this end a linear regression is applied between the data of each wind turbine or wind farm, yielding the slope and the offset to be applied on the forecasted power in order to obtain the correct average (that is: equal to the measured average when applied to the same data set).