Objectives:

The objective of this Work Package is to develop accurate models for wind resource forecasting based on advanced statistical and mainly on artificial intelligence methods (i.e. fuzzy logic, neural networks). These techniques permit to combine various types of explanatory inputs like wind direction, wind speed from neighbour sites, numerical weather predictions etc. Statistical techniques are very promising when high-resolution meteorological information is used as input to predict wind production up to 48-72 hours ahead.

Emphasis will be given to the development of statistical power curve models to estimate the relationship between wind power and local forecasts for meteorological variables such as wind speed and direction. Experience so far shows that one of the main error sources in wind power prediction lies in insufficient power curves. The use of certified power curves does not guarantee that the relation between wind speed and power output is accurately described. Models based on artificial intelligence will be developed for statistical downscaling. This consists in using a non-physical model for interpolating metrological forecasts from the nodes around wind farm to the level of wind farm and height of wind turbines. Statistical modelling can be an alternative technique to explicit terrain and roughness modelling.

Tasks will be dedicated on the development of approaches for regional or national forecasting of wind power (upscaling). Robust approaches for the on-line tuning of prediction models will be developped and evaluated. Such methods will permit adaptive models to avoid producing outliers in case of erroneous data input or extreme wind conditions.

Description of Work:

Project flow-chart