Wind Resource analysts try to lower the uncertainties in their wind resource assessment studies. To do that they use advanced flow models like computational fluid Dynamics (CFD) which per today use artificial boundary conditions. To lower the uncertainty even more and to model the turbulence and wake effects inside a wind farm the best possible way, a statistical-dynamical downscaling of numerical weather prediction models is developed based on CFD. Those numerical weather prediction models are typically run on a horizontal grid of several kilometers which is too coarse to describe the detailed flow field inside a wind farm. Downscaling those predictions with a CFD model, we can obtain a horizontal grid resolutions of around 10 meters which allows to have several grid points over the rotor area of a turbine. To be able to make meaningful wind predictions in a reasonable time a statistical approach is developed which allows to simulated only few weather patterns per year and to develop an annual energy production estimate from them.
In this work suitable methods for direct dynamical downscaling will be developed and implemented into the WindSim software. New methods like mesoscale modelling and machine learning will be used and evaluated. The final goal is to have an improved forecast of the wind speed and turbulence production in a wind farm.
Two novel meso-microscale coupling methodologies have been developed. The first approach consists in utilizing the average values of the mesoscale fields by wind directional sector. The second instead, extracts weather patterns by utilizing a fully automated clustering methodology.
Both methodologies were validated with data obtained for commercial projects. Advantages and limitations depending on the site conditions were studied. Ways to tackle the current limitations of the methodologies were identified.
This development allows WindSim to modernize its current meso-microscale coupling code. Furthermore, it enables potentially new services for customers that require to downscale mesoscale datasets.
Currently the developed methods are being adapted for their application in other areas of wind energy like wind energy forecasting, wind turbine wake modelling and dynamic line rating.
Wind energy is a volatile energy source which is not always available when it is needed. Therefore, it is very important to have an accurate wind power forecast such that other energy sources can be used to balance the energy feed into the electrical grid.
WindSim AS has developed forecasting systems based on Artificial Neural Networks (ANN) and based on Computational Fluid Dynamics (CFD). The ANN solutions are very efficient when the focus is only on predicting the wind speed at the turbine positions. But the power output of a wind farm depends also on the turbulence inside the wind farm and the wake effects between the turbines.
To model the turbulence and wake effects inside a wind farm the best physical description is a dynamical downscaling of numerical weather prediction models. Those models are typically run on a horizontal grid of several kilometers which is too coarse to describe the detailed flow field inside a wind farm. Downscaling those predictions with a CFD model we can obtain horizontal resolutions of around 10 meters. To be able to make meaningful predictions the downscaling should be run as so called transient simulations which makes it possible to calculate the time dependent behavior of the wind field and thereby the turbulence.
The purpose of this project is to establish an efficient meso-microscale coupling, answering the following questions:
- Can the TKE profiles from the mesoscale be used or is there a better formulation how to calculate the TKE inlet profiles?
- Can the turbulence prediction be improved by using more advanced turbulence models and can the wake prediction be improved by better models?
- Is Large?Eddy Simulation (LES) superior to Reynolds averaged Navier-Stokes (RANS) simulation or can transient RANS simulation work well for industrial application?
- What is the economic benefit of a meso-microscale coupling with regard to a better wind power forecast compared to the ANN methods when we consider the imbalance costs?