The primary objective of the project is to devise a method for integration of CFD and mesoscale meteorological modeling techniques for use in wind farm optimization. One of the secondary objectives is to improve mesoscale model accuracy by employing Machine Learning algorithms.
To improve the wind farm performance it is crucial to obtain an accurate predictions of wind turbine production and loading. Accurate wind characteristics on the turbine scale can be determined by integrating mesoscale models with the fine-scale fluid dynamics models employed in wind farm optimization by using mesoscale forecast of improved accuracy as in input for computational fluid dynamic (CFD) modeling. Recent advances in mesoscale meteorological models make it possible to tighter integrate the mesoscal meteorological modeling with fine-scale CFD. Nonetheless, the accuracy of CFD simulations is related to forecasts and if mesoscale forecasting is run for areas characterized by a complex terrain the results are expected to be of a poor quality. In such cases it is necessary to recognize the salient processes underlying the forecast error and to improve the accuracy of the mesoscale model before running fine-scale fluid dynamics models. Thus, a model that improves forecast accuracy for wind farms optimization is foreseen and will be a key output of this project.
A new mesoscale-microscale coupling model is thus proposed: trained on historical observations, the model uses mesoscale forecasts output to issue a high quality site-specific forecast. A new procedure for coupling mesoscale and fine-scale fluid dynamics and application of machine learning techniques enables the inclusion of realistic regimes and boundary conditions in CFD simulations of the relevant site and therefore provides a fast and reliable solution for a wind farm optimization in a situation of complex terrain. As the simulations of the wind farms are in some cases more representative of local terrain conditions with respect to average standards, idealised profiles with wider range of stratification regimes are here considered and their individual effects are accounted for in the optimization procedure.
Regarding the employment of ML techniques, several models and approaches are used to
- improve the quality of NWP (numerical weather prediction) and/or
- predict the total wind park energy yield and the wind speed 30- , 60- , or 180-minutes ahead.
ML-based model are built on real park production data provided by WindSim AS.