WIND RESOURCE ASSESSMENT BESTPRACTICES

Calibration of a New Actuator Disk CFDWake Model

Picowind Validation Project (France)

Document EWEA 2015 –Paris –17-20 November 2015Version : Final

WHITE PAPER - PUBLIC REPORT

Karim Fahssis1, Tristan Clarenc1, Guillaume Terris2, Benoit Buffard2, Clement Belgodere2,and Philippe Alexandre2

1aZiUgo - Marseille - France2La Compagnie du Vent –Montpellier - France

17-20 November 2015

Wind Resource Assessment Best Practices - Actuator Disk

IntroductionAccurate wake modeling is important to reduce projectuncertainties before financial closure, especially whendeveloping offshore wind farms. Wake modeling canalso be used to optimize sector management duringproject operation. Due to hardware limitations andhuge simulation durations, empirical wake models suchas the ones included in traditional CFD (Computa-tional Fluid Dynamics) software are based on linearequations. Simplifying the physics of turbine to tur-bine interactions can lead to high level of modelingerrors for both waked wind speed and turbulence levelestimates.But now, cloud computing offers unlimited, scalableand elastic calculation power through ZephyCLOUDservice and advantages of unstructured meshing tech-niques included in ZephyCFD software can be lever-aged to define new approaches taking into considera-tion wake modeling within the CFD simulations.The PICOWIND research project is funded by LaCompagnie du Vent to deeply evaluate and improvethe power production of a 22MW wind farm in oper-ation and to benchmark traditional wake models andcalibrate the new ZephyCFD wake model. La Picoteriewind farm (11 turbines Gamesa ; G90 2 MW ; 78m athub height) is located in Charly-sur-Marne (France)and is instrumented with one nacelle-mounted LiDARand three ground-based vertical profilers (two SoDARunits -Sound Detection and Ranging- and one LiDARunit -Light Detection and Ranging). Remote sensorsmeasurements of free stream wind speeds along withboth upwind and downwind vertical wind profiles arecollected in order to first calibrate a new CFD (Com-putational Fluid Dynamics) actuator disk wake modeland then validate it against available measurementsof the PicoWind project. The proposed wake modelleverages the advantages of unstructured meshing tech-niques by running a first wind flow model to take intoconsideration the horizontal flow deviation at each tur-bine locations before generating an actuator disk meshfor the final simulation.

Site DescriptionLa Picoterie operating wind farm data are used tobenchmark traditional wake models and calibrate newwake models in the frame of the PICOWIND researchproject.

Figure 1: La Picoterie OperatingWind Farm

Figure 2: Wind Rose during StudyPeriod

Benchmarking Empirical WakeModels

Objectives• Comparing measured wake against simulated

wake using two empirical wake models (PARKand Fast EVM) included in a traditional CFDsoftware.

• For freestream wind directional sectors, calculatethe alpha ratio at six different heights betweenwaked wind speeds and freestream wind speedsmeasured by two sodars and simulated.

MethodologyMeasured Wake :

• Measured wind data from East Sodar and WestSodar are considered;

• Data sets are filtered and only freestream winddata are kept from both Sodars for six differentfreestream wind directional sectors;

• Alpha ratio (downstream Sodar wind speed / up-stream Sodar wind speed) is calculated for sixdifferent heights.

Simulated Wake :

• For six different freestream directional sectors,Freestream Sodar data at 80m height are usedas reference wind data in the traditional CFDsoftware;

• Two empirical wake models are considered :PARK and Fast EVM wake models;

• For six different freestream directional sectors,Waked wind speeds are calculated for six differ-ent heights;

• For six different freestream directional sectors,Alpha ratio is calculated for six different heights.

EWEA 2015 –Paris –17-20 November 20151

Wind Resource Assessment Best Practices - Actuator Disk

ResultsThe table below(in appendix as well) compares thesimulated alpha ratio against the calculated ratio forsix different freestream wind sectors, six differentheights and two different empirical models of a tra-ditional CFD software.Alpha = downstream Sodar wind speed / up-stream Sodar wind speed

Table 1: Simulated Alpha RatioAgainst the Calculated Ratio forSix Different Freestream Wind

Sectors

Figure 3: Simulated Alpha RatioAgainst the Calculated Ratio forSix Different Freestream Wind

Sectors(graph)

Benchmarking Conclusions• The simulated wake is higher than the measured

wake for both the empirical models.

• As expected, the measured wake is higher at hubheight.

• Simulated wake is not higher at hub height forone empirical model

• Such results show that a new approach for wakemodeling is required in order to reduce wakemodeling uncertainties in complex sites.

Toward ZephyCFD actuator diskwake model

Objectives• Define a workflow allowing automatic and opti-

mized cloud computations through actuator diskCFD wake model.

• Define the validation methodology to be fol-lowed.

Methodology• Free of wake calculations are run for 36 ten-

degrees wind sectors based on standard Zephy-CFD calculation process, and resulting wind di-rections are extracted at each hub location.

• 36 actuator disk meshes are generated, takinginto account the exact orientation for each of therotors (based on the previously evaluated winddirections).

• Flow is automatically initialized, remappingfrom previously evaluated free of wake calcula-tions results, considering a high wind conditionthrust coefficient.

• 10 consecutive CFD runs are processed ; betweeneach process, the thurst coefficient is varied toevaluate the wakes for 10 different bins of windspeed

• All the wind speed bins and directions sectors arestatistically processed with the measurements toevaluate the wake effects at each hub.

• Total of 38 meshes, 432 CFD computations, 37remapping process ; can be considered as an ex-pensive numerical process and cannot beprocessed easily by any standard hard-ware.

ResultsFree of wake calculations:

• Standard ZephyCLOUD service is used ; a 6.5millions of cells mesh is generated on the cloud,and 36 wind directions are then evaluated.

• The mesh accuracy in the viscinity of the projectis equal to 25.2 meters

Figure 4: The 36 calculations arerun simultaneously on 36 virtualmachines ; each machine has 36

cores.The complete duration for the

whole process is 1h32.

EWEA 2015 –Paris –17-20 November 20152

Wind Resource Assessment Best Practices - Actuator Disk

Actuator Disk Meshes:• Position, size, and rotor orientation are con-

troled.

• Refinement in the rotor volume and downwindeach rotor should be investigated and bench-marked thanks to the soldar measurements.

• Mesh generation should be distributed on virtualinstances to make possible simulteneous meshgenerations and optimized computation dura-tions.

Figure 5: Refined Rotors inActuator Disk Mesh and Their

Wake Shapes

Actuator Disk Computations：

• The number of iterations is highly reduced, dueto the initialization strategy, using remapped re-sults from free of wake calculations ; the initialthrust coefficient calculation converges in a fewhundred of iterations.

• A few tens of iterations allows to evaluate eachnew thrust coefficient configuration, so that theglobal process can be evaluated in a few hours.

Conclusion• The modeling strategy for optimized wake cal-

culations through Actuator Disk model has beendefined.

• Thanks to ZephyCLOUD scalability, the wholeprocess is expected to last a few hours with highrefinement criterions.

• The measurements from PICOWIND researchproject will allow to benchmark and to calibratethe mesh refinements in both rotor volume andwake zone.

EWEA 2015 –Paris –17-20 November 20153

Wind Resource Assessment Best Practices - Actuator Disk

Appendix

Table: Simulated Alpha Ratio Against the Calculated Ratio for Six DifferentFreestream Wind Sectors

Figure: vSimulated Alpha Ratio Against the Calculated Ratio for Six DifferentFreestream Wind Sectors(graph)

EWEA 2015 –Paris –17-20 November 20154