Assessment of Mean Flow Across a Wind Farm Edge
The OWA Rødsand 2 Scaning Lidar experiment is analyzed to characterize the formation (and exit from) an internal wind farm boundary layer
Status
This benchmark was cancelled due to the poor quality of the validation dataset. We tried to produce ensemble-averaged data for the West sector, binning with the same settings used in the OWA Challenge, but there were not enough samples to obtain smooth mean wind fields.
Background
The Offshore Wind Accelerator (OWA) scanning lidar experiment (Adams, 2015) is analyzed to characterize the flow incoming to and leaving the Rødsand 2 wind farm. This detailed analysis complements current activities within the OWA Wake Modelling Challenge towards the assessment of array efficiency prediction from wake models using operational data from 6 offshore wind farms.
Scope and Objectives
The objective of this more detailed OWA benchmark is to understand how array efficiency prediction depends on appropriate modeling of the interaction between the incoming atmospheric boundary layer (ABL) and the wind farm boundary layer (WFBL) in large wind farm clusters.
Two long-range scanning lidar systems measure the transition between the incoming (relatively) homogeneous flow from the West to a wind farm canopy flow resulting from an internal boundary layer generated by wake expansion and interaction from the first rows of wind turbines. Conversely, from the East, the systems capture the initial recovery to free-stream from deep-array conditions leaving the Rødsand 2 — Nysted wind farm cluster. The analysis focuses on generating data for validation on these two sectors for three stability classes based on the same categorization of wind conditions introduced in the Anholt benchmark.
Both systems operate independently of each other targeting two non-overlapping areas along a 12-km long transect in the W-E direction. Therefore, strict validation can only be done based on the line-of-sight (radial) wind speeds that these systems capture. However, to help interpreting the flow behavior, a reconstruction of the horizontal wind speed is also performed by projecting to the local wind direction. More details about the experiment set-up and post-processing are provided next followed by a description of the outputs required from benchmark participants to conduct the validation study.
Test Case: Rødsand 2 — Nysted Wind Farm Cluster
Figure 1 shows the location of the Rødsand 2 — Nysted cluster in the Baltic Sea and details about their respective wind turbine model. The layout of both wind farms and instruments from the OWA experiment are shown in Figure 2. The available SCADA data spans a period from 1 February 2013 to 30 June 2014. Figure 3 shows the data availability from the sensors during this period.
Input Data
Mesoscale simulations using the NEWA WRF set-up are produced to generate background wind conditions for wake models that are free of (microscale) site effects. Neighbouring wind farms have also been simulated using the Fitch wind farm parameterization available in WRF. The innermost nest in the simulations has a horizontal resolution of 3 km. Following the same approach that was adopted in the EERA-DTOC project (Schepers et al, 2015), two sets of input data are generated and provided in NetCDF format:
- Control: Free of wake effects, to characterize the background wind resource.
- Wakes: Includes neighbor wind farms but not the target wind farm, to characterize inflow conditions for microscale wake modeling.
More details about the WRF configuration and the contents of each input data file are provided in the Anholt benchmark.
For consistency, we provide the same generic power and thrust curves that were used in the EERA-DTOC project.
Input and output data is facilitated through shared folders in b2drop.
Reference Wind Conditions
Consistent with the array-efficiency prediction benchmarks, bin-averaged validation data is defined in terms of reference wind conditions at the Rødsand 2 centroid simulated by a mesoscale model. While we count with mast measurements at the site, they are disturbed by the presence of the wind farm and, for consistency with the other wind farm cases, we preferred to keep using the same data to characterize inflow conditions. These virtual mast data is generated by horizontally averaging WRF simulations within a 30-km wide 10x10 squared grid around the centroid. Then, reference wind speed and direction are obtained by interpolation at hub-height (68.5 m) and stability is defined by the z/L parameter, where z = 10 m and L is the surface-layer Obukhov length computed by WRF. Bin-averages are obtained for the W and E based on 30º sector width and 9 ± 1 m/s wind speed bin for three stability classes:
- Unstable (u): -0.2 < z/L < -0.02
- Neutral (n): -0.02 < z/L < 0.02
- Stable (s): 0.02 < z/L <0.2
As observed in Figure 1, the 30-km box from which the reference wind conditions are obtained includes both wind farms. Since we are interested in simulating wake effects explicitly through a microscale model, we should use reference data in control conditions, i.e. without adding Nysted in the mesoscale simulation (i.e. Rodsand2_Control_Lav30km_ref.nc).
Validation Data
Scanning lidar data is averaged into hourly intervals for two reasons: 1) to bring all temporal variability within the scanning pattern to a common timeframe and 2) to make the validation data consistent with the mesoscale integration time which filters out variability of shorter duration. This integration time is longer than the flow-trough time required for the flow to travel the longitudinal span of the wind farm cluster (around 20 km) at the reference wind speed of 9 m/s. This ensures that all turbines and sensors are under the same reference wind conditions regardless of the time-lag between them and the reference site.
As in all the OWA benchmarks, the validation will only include nominal conditions when all the wind turbines operate near the generic power curve. A flags file is provided indicating which hourly data will be retained in the validation.
Radial velocities from the lidar systems falling within each reference bin are selected to generate a validation dataset confined in a control volume that is 12 km long, 2 km wide and 1 km high (Figure 2). Hence, the validation data consists on a cloud of points corresponding to the scanning patterns of the lidar systems, with higher density at low elevation angles within the wind farm than above.
To facilitate the interpretation of the validation results, a reconstruction of the horizontal wind speed is performed by projecting both the observed and simulated radial wind speeds to a local wind direction. Then, a 2D longitudinal transect is obtained by subdividing the control volume into a Cartesian grid and then volume-averaging all the data falling within each cell. Note that we average across the lateral dimension to focus on the longitudinal and vertical variability of the wind speed. The grid is stretched in the vertical direction to accommodate the decreasing number of samples with height.
Contour plots of speed-up ratios with respect to the reference wind speed will show how the vertical structure of the incoming ABL is modified into an internal wind farm boundary layer (WFBL), when the flow comes from the West, and how the flow recovers from an equilibrium WFBL when the flow comes from the East.
Simulation Strategy
In a single wind farm approach we shall simulate the target wind farm (Rødsand 2) at microscale with no explicit modeling of the shore or neighbor wind farms (Nysted), since we trust that mesoscale simulations already include these effects. Then, you can assume horizontal homogeneity in the inflow conditions by using the reference virtual mast to define the inflow conditions for all the turbines (Rodsand2_Wakes_Lav30km_ref.nc). Alternatively, you can use heterogeneous wind conditions at each wind turbine (Rodsand2_Wakes_WindTurbines.nc, i.e. hub-height interpolated data).
In the case of the Rødsand 2 — Nysted cluster, the wind farms are separated by only 3.5 km and, therefore, it makes sense to simulate both at microscale as a cluster. In this case, you shall use input data in control conditions, i.e.: Rodsand2_Control_Lav30km_ref.nc assuming homogeneous inflow and Rodsand2_Control_WindTurbines.nc and Nysted_Control_WindTurbines.nc in heterogeneous inflow conditions.
Note that WRF 3D fields are also available upon request.
The single wind farm approach is the default in the multi-site assessment. Therefore, we suggest benchmark participants to also run the cluster approach for the East sector to quantify the difference between the two methodologies.
Output Data
Participants can submit their results as ensemble-averaged quantities or as time-series. Only formatted data following the benchmark guide should be submitted. You are encouraged to use the model evaluation script to test compatibility and self-assess your results by comparing with other model results that may be available. Please attach documentation that will help interpret your simulations and your self-assessment.
Turbine power prediction output data should be provided in the same format described in the Anholt benchmark (as part of the multi-site assessment).
In addition you should submit the 3D velocity components at the coordinates of the scanning lidar data cloud. A xyz file is provided for you to interpolate the data from your simulations. This file is equivalent to the one generated for the scanning lidar data so both can be post-processed in the same way to generate the longitudinal transect contour plots of speed-ups.
An upload-only b2drop folder is available for you to submit your output data.
Schedule
- 30 September 2019: final deadline for submission of results for all sites.
- 4 October 2019: Blind test results.
- End of October: Rodsand 2 validation results.
- 26–28 November 2019: Presentation at WindEurope Offshore.
Acknowledgements
This benchmark is organized and funded by the Offshore Wind Accelerator (OWA) Partners with support from the IEA Task 31 “Wakebench” Phase 3. The observational data is kindly provided by OWA participants as per Table 1.
References
Adams N (2015) OWA Wakes Measurement Campaign. Analysis of Measured Data and Comparison with Models. FNC 39560–40785R report, 479 pp, confidential
Schepers G., et al. (2015) EERA DTOC calculation of scenarios. Deliverable D5.12 of the EERA-DTOC project, June 2015
Sanz Rodrigo J., Gancarski P. (2019) OWA-Anholt Array Efficiency Benchmark. https://github.com/CENER-EPR/OWAbench
