Are Advanced Wind Flow Models More Accurate? A Test of Four Models Philippe Beaucage, PhD Senior Research Scientist Michael C. Brower, PhD Chief Technical Officer Brazil Wind Power Conference 2012
Albany New York, USA Barcelona Spain Curitiba Brazil Bangalore India AWS Truepower in Brief Partnering with industry and government to advance renewable energy worldwide Consulting in atmospheric science, meteorology, environment, and engineering Involved in design and assessment of >60,000 MW of wind projects Established in 1983 105 staff in three main offices Project roles in over 80 countries Opening office in Curitiba (Fall 2012) Partners: Camargo Schubert (Curitiba), Aires Renovables (Buenos Aires)
Why Do Wind Flow Modeling? Extrapolates from a few wind resource measurements to an entire wind farm Allows optimization of the plant layout Doing it well is essential for accurate energy production estimates
Physics Based Numerical Models Solve some subset of the primitive equations governing atmospheric conditions Typically use finite element methods, sometimes in combination with spectral methods A wide variety of models have been developed reflecting Available computer power The developer s understanding of the physical processes most relevant to wind energy
The Primitive Equations Mass conservation: Momentum conservation (Navier-Stokes): Energy conservation: Equation of state: d dt u t dt c p u 0 P RT 1 2 3 4 1 1 u u g 2 u P Ff dq dp 5 N S Terms 1: Advection 2: Gravity force 3: Coriolis force 4: Pressure force 5: Viscous stress 6: Friction force 6
Key Challenges in Numerical Modeling Spatial resolution must be ~50 m or better over a domain ~25 km or larger Must simulate a wide range of wind conditions for accurate energy production estimation Should simulate all relevant physical processes but which are relevant? More advanced models require much more computer time than simpler models. But are they worth the cost?
Jackson Hunt* Fast, linearized, steady state N S solver (e.g., WASP) Assumes terrain is a small perturbation on a constant background wind field Infers the regional wind climate based on point measurements Reverses process to extrapolate to other points Includes obstacle, surface roughness modules *Jackson, P.S. and J.C.R. Hunt (1975). "Turbulent Wind Flow over Low Hill". Quart. J. R. Met. Soc., vol. 101, pp. 929 955.
WAsP Example: Orographic Influence on Wind Speed WAsP (Risoe)
CFD Models Most are Reynolds averaged Navier Stokes (RANS) solvers (e.g., Meteodyn WT, WindSim) Non linear; can simulate re circulations, flow separations, other effects of steep terrain Include turbulence parameterization Usually assume constant, homogeneous boundary conditions Do not simulate thermal gradients or stability (buoyancy)
CFD Example: Recirculation Behind a Ridge Source: WindSim, Vector AS
Numerical Weather Prediction (NWP) Models Full time varying 3D physical model of the atmosphere. Examples: WRF, MASS, KAMM, ARPS Solve all the Primitive Equations, including energy balance, surface exchanges, phase transitions, with turbulence parameterization Require 100 1000x as much computer time as linear models; usually done on a high performance computing cluster Can be coupled with linear models to improve resolution, reduce runtime Current research: Coupled large eddy simulations to resolve the larger scales of turbulence
NWP Example: Island Blocking and Channeling
NWP Modeling: Gravity Wave
Research Outline Objective: Determine how much accuracy improves (if any) with more advanced models Other research* has produced mixed results for Jackson Hunt and CFD models. NWP based models rarely studied. AWS Truepower assessed 4 leading models at 4 sites with a total of 26 masts Round robin approach: Use one mast to predict the mean wind speed at the other masts Calculate errors (predicted minus observed mean speed) Repeat with other masts Compile root mean square error (RMSE) for all models and cases *For summary, see Wind Resource Assessment: A Practical Guide to Developing a Wind Project (Wiley, 2012), chapter 13.
Models Jackson Hunt: WASP (Risoe National Laboratory) RANS CFD: Meteodyn WT (Meteodyn) Coupled NWP Linear: SiteWind (AWS Truepower) Coupled NWP LES: ARPS (Oklahoma University) Case Studies Case Terrain Land Cover No. Masts 1 Simple Grass and trees 8 2 Moderate Grass and shrubs 6 3 Complex Coastal Mostly forest 3 4 Complex Forest 9
Results By Case Root Mean Square Error (m/s) Model 1 2 3 4 All WASP 0.26 0.34 1.15 0.74 0.62 Meteodyn WT (neutral stability) 0.50 0.46 1.07 0.95 0.76 Meteodyn WT (mixed stability) 0.41 0.55 1.09 0.97 0.76 SiteWind 0.10 0.39 0.56 0.67 0.48 NWP/LES 0.28 0.49 0.57 0.49 0.45
Conclusions No single model performed best at all sites On the whole, NWP based models performed better than CFD and WASP type models by a substantial margin Overall, CFD did not perform better than WASP!! Supports other research reporting mixed findings This shows that wind flow in complex terrain cannot be solved by the same methods used to model air flow over wings and through jet engines The atmosphere is never in equilibrium, and you cannot ignore the effects of vertical and horizontal temperature gradients