Back to the programme printer.gif Print

Delegates are invited to meet and discuss with the poster presenters in this topic directly after the session 'How does the wind blow behind wind turbines and in wind farms?' taking place on Tuesday, 11 March 2014 at 16:30-18:00. The meet-the-authors will take place in the poster area.

Lasse Svenningsen EMD International, Denmark
Lasse Svenningsen (1) F P Morten Lybech Thøgersen (1)
(1) EMD International, Aalborg, Denmark

Printer friendly version: printer.gif Print

Presenter's biography

Biographies are supplied directly by presenters at EWEA 2014 and are published here unedited

Lasse holds a PhD in Geophysics from the University of Aarhus in Denmark.
In his current position Lasse has worked for EMD International for 5 years as scientific researcher and wind energy consultant. The focus of his work is development and integration of improved calculation methods for the software tool WindPRO in particular CFD, Meso-scale and Aero-elastic models and their use in energy yield assessments and site suitability analyses. Another focus area is power curve correction methods.
In his previous employment Lasse worked in the wind and site competence centre of a wind turbine manufacturer.


Quantitative evaluation of an automated, HPC cluster-based CFD solution


Computational Fluid Dynamics (CFD) models are becoming increasingly important in wind energy site assessments. The main reason is that the CFD models account for more of the non-linear flow effects in complex terrain compared to the simple and fast linearized flow models. The drawback of CFD modelling is the notorious challenges in setting up appropriate computational meshes and getting the numerical solution to converge stably, but also large computational requirements. In this study we evaluate the performance of a new approach to CFD modelling, which is fully automated and runs on a dedicated HPC cluster.


In this work we evaluate the model WAsP-CFD, developed by Risø-DTU; it integrates the non-linear flow solver Ellipsys3D also developed at Risø-DTU over many years and for multiple purposes (Sørensen, 1995, and Michelsen 1992, 1994).

The WAsP-CFD set-up is fully automated, and requires only user interaction as the simpler linearized flow models like WAsP. We therefore find it important to evaluate WAsP-CFD on a diverse range of test cases to quantify if WAsP-CFD results constitute an improvement when compared to the linearized flow model WAsP.

A number of test cases were selected for the evaluation of WAsP-CFD and the tests were split in two main groups. First group “classical cases” is based on the literature, either theoretical studies or data from frequently cited measuring campaigns. Second group “reference cases” cover real wind power sites in complex terrain with wind measurements suitable for model evaluation.

The basic concept of WAsP-CFD is to combine an automated CFD solution for flow perturbations with the general modelling concept and stability model of WAsP. The basis of the WAsP model (Troen and Petersen, 1989) is a superposition of the following four sub models:
1) a linearized flow model to calculate terrain speed-ups (IBZ),
2) a model to calculate the effect of roughness transitions,
3) a model that predicts the vertical flow perturbations due to non-neutral stability,
4) a model to predict reductions caused by obstacles.

In WAsP-CFD flow perturbations of model components 1) and 2) are replaced by flow perturbations predicted by the non-linear Ellipsys3D solver running on a dedicated HPC cluster. WAsP model components 1+2 assume neutral stability which is also the case for the flow perturbations predicted by Ellipsys3D.

Main body of abstract

A key distinction of the automated WAsP-CFD setup is that all calculation domains have the same fixed size. Only flow results for the central 2x2km of the fixed domains are returned to the user. As a result of this setup, a wind farm site area must be split into a number of 2x2km CFD-tiles which are then submitted as individual calculations. Once submitted to the HPC cluster the automatic processes take over, generating the computational mesh (Sørensen, 1998), establishing inlet conditions, running the Ellipsys3D solver and ensuring convergence of the iterative numerical solution for each of the 36 sectors.

The main focus of this study is to evaluate the performance of the automated WAsP-CFD model, with emphasis on the CFD model components. Model performance has been evaluated by direct evaluation of speed-up factors, and at a higher level using cross-predictions of observed wind climatologies. In a cross-prediction, observations of the wind climate at one position are used to predict the wind climate at another position which also has observations. Model performance is then characterized by the cross prediction error of the predicted mean wind speed. For multiple masts on a site an overall performance measure is the Root-Mean-Square-Error (RMSE) of a set of cross-predictions.

The “classical test cases” selected in this study comprise:
• Speed-up versus terrain slope (theoretical)
• Aveiro-Viseu site, Portugal (original Delta RIX site)
• Askervein hill experiment (1983 experiment)

First classical case “Speed-up versus terrain slope” refers to reproduction of a theoretical study (Yamaguchi, Ishihara, and Fujino, 2002) of speed-up factors for different terrain slopes using synthetic cosine shaped hills with mean slopes from 3° to 39°. Second classical case "Avero-Viseu" is a well-studied site in Portugal, which was originally used to establish the Delta RIX correction method to correct WAsP modelling errors in complex terrain (Bowen and Mortensen, 2004). Third classical case is an often studied dataset from the “Askervein hill” atmospheric experiment conducted in 1982-83 in Scotland (Taylor and Teunissen, 1984). This experiment is probably more cited in the wind energy industry, than any other meteorological experiment. The dataset used here includes a transect across the Askervein hill at 10m a.g.l. and a profile measured at the hill top at 3 to 34m a.g.l.. Concurrent with the transect and profile measurements an up-wind mast measured at identical heights to serve as reference point in speed-up calculations.

Second group of tests, the “reference cases”, are real sites which fulfilled the criteria: complex terrain, multiple masts, acceptable data quality and acceptance by the project owner. A large geographical spread was also prioritized. Eventually, six available wind power projects turned out to fulfil the criteria and these projects were named according to their region:
• Oceania
• Central Europe
• North-West Europe
• East Asia
• Central America
• South America

For the reference cases model performance is decomposed into lateral cross-predictions between observations at same height and in prediction of the vertical profile at each mast. The former is the most direct evaluation of the actual CFD-solution as the WAsP stability model does not influence this performance. The latter on the other hand is strongly influenced by the settings of the WAsP stability model. In all runs default stability settings have been employed. Profile fit could be further optimized by tuning the WAsP/WAsP-CFD stability parameters for each site. However, as the stability model is a post processing and is identical for WAsP and WAsP-CFD such analysis would not contribute to the validation of the actual CFD solution in WAsP-CFD.

All reference cases were run with a 2x2km CFD-tile centred on each mast position. However, for three of the sites all masts could fit within just one 2x2km CFD-tile. Comparing results between the two tile configurations, “each-mast-centred” and “all-in-one-tile” allows an evaluation of how WAsP-CFD performance depends on the CFD-tile configuration.


Results of the classical case “Speed-up versus slope” show that WAsP-CFD speed-ups are consistent with WAsP at terrain inclinations below 10-15°. At large inclinations WAsP predicts larger speed-up than WAsP-CFD. A maximum in speed-up is seen for WAsP-CFD around a slope of 20° where WAsP speed-ups are roughly 15% higher (below, left). Speed-up factors depend on roughness length in a consistent way for WAsP and WAsP-CFD (below, right).

For the classical case “Original Delta RIX site” WAsP shows cross prediction errors up to 38%, strongly correlated with Delta-RIX. Using WAsP-CFD cross-prediction errors are within 9% and do not correlate with Delta-RIX (cf. below).

For the classical case “Askervein hill experiment” WAsP-CFD RMS errors reduce to c. 1/3 compared to WAsP, both along the transect across the hill (below, left) and for the the hill-top vertical profile (below, right). Speed-up factors at the lee side of the hill and closest to the ground of the hill top are better captured by WAsP-CFD.

For the six reference cases cross-prediction of wind climatologies was performed between the masts on each site. For top anemometers, which have the highest data quality, results show a consistent reduction in RMS cross prediction errors (RMSE) of 0.31±0.04m/s for WAsP-CFD relative to WAsP (below). On average the RMSE is more than halved and corresponds to a reduction of prediction errors of 4.0±0.5% on mean wind speed or 6-8% on annual energy production. Using measurements at lower heights WAsP-CFD results also show a consistent but smaller improvement relative to WAsP (not shown).

WAsP-CFD did not improve predictions of observed wind profiles for the 15 masts on the reference sites, using default stability settings. Profile RMSEs show a large scatter both for WAsP and WAsP-CFD, but averages are comparable (below). This suggests that WAsP-CFD performance should not be evaluated solely from profile-fit, and that the stability model may be further improved in WAsP-CFD.

WAsP-CFD performance shows a minor dependence on the 2x2km tile configuration. RMSE improvements relative to WAsP worsened averagely by 0.04m/s for “all-in-on-tile” compared to the “each-mast-centred” configuration. Improvements of the WAsP-CFD “all-in-on-tile” were still significant compared to WAsP (below).

Learning objectives
This study evaluates the performance of the fully automated non-linear flow model WAsP-CFD relative to the linearized flow model WAsP. The analyses reproduce several classical studies from the literature which contribute to a better understanding of the differences in performance between linearized and non-linear flow models.

Finally the results of this study quantify the improved accuracy of using non-linear models instead of linearized models in complex terrain.

Bowen, A.J. and N.G. Mortensen, 2004, WAsP prediction errors due to site orography. Risø-R-995(EN). Risø National Laboratory, Roskilde. 65 pp.
Michelsen, J.A., 1992, Basis3D – a platform for development of multiblock PDE solvers. Technical report AFM, 92-05. Technical University of Denmark, Denmark.
Michelsen, J.A., 1994, Block structured multigrid solution of 2D and 3D elliptic PPDE solvers. Technical report AFM, 94-06. Technical University of Denmark, Denmark.
Sørensen, N.N.,1995, General purpose flow solver applied to flow over hill. Risø-R-827(EN). Risø National Laboratory, Roskilde, Denmark, 154 pp.
Sørensen, N.N., 1998, HypGrid2D a 2-D mesh generator. Risø-R-827(EN). Risø National Laboratory, Roskilde, Denmark.
Taylor, P. A, and Teunissen, H. W., 1984, The Askervein Hill Project: Report on the sept/oct 1983, main field experiment. Technical report, Atmospheric research.
Troen, I. and Petersen, E. L., 1989, European Wind Atlas, ISBN 87-550-1482-8, Risø National Laboratory, Roskilde. 656 pp.