APPLICATION OF A KALMAN FILTER FOR IMPROVING WIND AND SOLAR IRRADIANCE FORECASTS FOR THE RENEWABLE INDUSTRY

Wind forecasts from numerical weather prediction (NWP) models exhibit significant systematic errors. This is particularly the case at locations with difficult topography. In addition, certain meteorological phenomena such as the forming of low-level jets may be systematically ill-timed or biased in strength. Likewise, solar irradiance forecasts may be biased due to static local conditions such as the presence of tall buildings or mountains that block the radiation at low solar elevation angles. Low elevation angles are also associated with long optical path lengths through the earth's atmosphere. Current deficiencies in the various NWP models regarding the description of atmospheric cloud, water vapor and aerosols are therefore also likely to contribute to larger relative errors at these angles.

MeteoGroup has developed a Kalman filter for correcting the type of systematic errors described above. Kalman filtering as a means for post-processing numerical weather prediction model output has been around for several decades. Traditionally, Kalman filters have been mainly applied on weather parameters that have errors that are normally distributed. However, the current work will show that this filtering technique is also successful in removing systematic errors in wind and solar irradiance forecasts.

A recursive Kalman filter does not require the storage of large training data sets. The system is ‘self-learning' and will adjust automatically if NWP models are upgraded or local conditions at the site of interest change.