A systematic validation of CFD flow modelling on commercial wind

PO. ID
108
A systematic validation of CFD flow modelling
on commercial wind farms sites
Jean-François Corbett, Ulrik Horn, James Bleeg, Rich Whiting
DNV GL – Energy, Renewables Advisory
Poster number PO.108
Results
Abstract
We present here what we believe to be the most comprehensive
validation of the commercial application of CFD on wind farms
seen to date, with 115 very diverse sites and 2500 mast pairs.
Previously published validations have generally rested on very small
sample sizes and narrow categories of site types1.
The validation we present here is of a different character: all multi-mast
sites on which we performed CFD in 2012 and 2013 were systematically
included, without any manual selection or filtering criteria.
Averaged across all sites, the error on MWS was one fifth lower with CFD
than with the linear model (4% vs. 5%).
On the 70% of sites where CFD was more accurate than the linear model, it
reduced error on MWS by 2% on average. In the context where CFD is
executed in addition to the linear model, the estimated accuracy benefit on
MWS is therefore 70% × 2% = 1.4%. For a typical energy analysis, the
corresponding benefit on energy uncertainty would be approximately 3%.
Purpose
However, CFD also places higher demands on the expertise of the user,
both in terms of setting up and running simulations, but also of
interpreting the vast quantities of information produced.
DNV GL aims to help the financial community make informed judgments
regarding the bankability of the latest modelling techniques. To this end
we have consistently published growing validations as our CFD service
has matured and ventured into new types of geographies and
challenging flows2,3.
CFD error on MWS
CFD can reduce flow modelling errors, leading to better wind farm
designs, better project returns, and ultimately lowering the cost of wind
energy.
15%
Equal error
•In-house software powered by STAR-CCM+, a state-of-the-art CFD solver
•Reynolds-Averaged Navier-Stokes (RANS) equations with k-ε closure
•Stability modelled on a subset of the sites: transport equation for potential
temperature, shallow Boussinesq approximation, source term in the turbulent
kinetic energy equation.
•Forestry: source terms for turbulence production and aerodynamic drag4.
•For purpose of comparison, linear model calculations were also performed.
Choice of sites
•All sites on which we ran CFD and a linear model in 2012 and 2013, and which
were equipped with at least two meteorological masts.
•Exclusion criteria: only one site mast; linear model results not available.
•No exceptions, no filtering, no cherry-picking, no tuning.
•Total of 115 sites with 2500 mast pairs, located across a range of continents,
representing a broad variety of terrain types and atmospheric conditions.
Error metric
•Relative error on long-term, hub-height mean wind speed (MWS) resulting from
cross-predictions between masts, averaged for all mast pairs on a given site.
•Advantage: MWS is readily available from our existing commercial processes,
making possible its systematic collection in a busy commercial environment.
•Drawbacks: directional information is discarded; includes noise resulting from
long-term correlation and vertical extrapolation errors, and from the inclusion of
many mast pairs where no microscale model can be expected to perform well.
This was deemed an acceptable trade-off given the vast amount of data
produced.
0%
0%
5%
10%
15%
Linear model error on MWS
20%
25%
40%
30%
Frequency
Simulations
5%
115 sites
The present validation focuses on the setup and execution of
simulations, in an approach that mimics a blind test. When CFD is
deployed in a formal energy analysis, engineering judgment can extract
additional value by making manual adjustments and producing an
intelligent initiation strategy based on the model’s performance on
different parts of a site. However, in order to ensure absolute objectivity
in this study, only the raw model outputs were considered.
Methods
10%
Linear model
CFD
20%
10%
0%
0% - 2%
2% - 4%
4% - 6%
6% - 8%
8% - 10%
10% - 12%
> 12%
Error on MWS
Conclusions
CFD has oft been cited as the remedy to the weaknesses of linear models,
but these claims have been met with industry scepticism in the face of the
scarce validation published to date. With an increasing commercial application
of CFD at DNV GL and elsewhere, large-scale validation is now possible.
The results of this extensive, two-year validation project show that the CFD
analysis process and engineering expertise developed at DNV GL add
significant value to energy production assessments. This benefit is bound to
increase as our CFD methods continue to mature. Moreover, engineering
judgment can provide additional value on a site-specific basis.
Though this study focused on multi-mast sites, the outcome gives confidence
that CFD can also reduce uncertainty on single-mast sites where model
performance is not as easily assessed.
References
1. M. Brower, J. Vidal, P. Beaucage, “Evaluation of four numerical wind flow models”, EWEA Resource Assessment Workshop 2013,
Dublin, Ireland.
2. J.F. Corbett et al., “CFD can consistently improve wind speed predictions and reduce uncertainty in complex terrain”, EWEA 2012,
Copenhagen, Denmark.
3. J. Bleeg, D. Digraskar, J. Woodcock, J. F. Corbett, "Modeling stable thermal stratification and its impact on wind flow over topography",
Wind Energy 2014. DOI:10.1002/we.1692. (print publication pending)
4. J.C.L. da Costa, F.A. Castro, J.M.L.M. Palma, P. Stuart , “Computer simulations of atmospheric flows over real forests for wind energy
resource evaluation”, J. Wind Eng. Ind. Aerodyn., 94 (2006), pp 603-620.
EWEA 2014, Barcelona, Spain: Europe’s Premier Wind Energy Event

Download Report