Scoping Wakebench Phase 4
Towards a more actionable wind farm flow model evaluation framework that can drive industry-wide adoption.
Background
The IEA-Wind Task 31 “Wakebench” community has established an international framework for the evaluation of wind farm flow models. This framework is currently documented in the Wind Energy Model Evaluation Protocol (WEMEP), an open-source project that provides general context about verification, validation and uncertainty quantification (VV&UQ) procedures as well as benchmark repositories for the validation of atmospheric flow and wake models being used in energy yield assessment, siting and wind farm design. The evaluation process is based on the combination of validation benchmarks targeted at reproducing relevant phenomena of interest, captured by experiments from public R&D projects, with validation on the intended use of the models, based on blind tests with long-term operational campaigns from industry.
Value Proposition
The next challenge for the Wakebench community is to make this framework more actionable towards the improvement of engineering models used by the wind industry. This is stated in the following value proposition:
Build industry-wide consensus on the adoption of a model evaluation framework and quality-assurance criteria for engineering wind farm flow models used in energy yield assessment (EYA), site suitability and wind farm design.
Objectives
To this end, the following objectives will be pursued:
- Support the development of multi-scale wind farm flow models from mesoscale to near-wake, at the core of next-generation statistical and dynamic engineering wind farm models.
- Complement physics-based models with data science techniques to build cost-effective surrogate models, train model parameters, mitigate biases and quantify uncertainties.
- Maintain the good level of physics-based benchmarks driven by academia in connection to large experimental campaigns.
- Reinforce benchmarks for intended use (siting, EYA, wind farm design) driven by industry in connection to operational campaigns.
- Build industry-wide consensus on quality-acceptance criteria, model evaluation methodologies and metrics for a traceable evaluation of flow models.
- Define pathways to shorten the cycle for advanced models to reach industry tools and standards, notably: IEC-61400–15–1/2 and IEC-61400–12–4.
Structure
For now, we are considering this simple WP structure, following Task 31 practices and organization:
- WP1: Deconstructing high-impact flow phenomena (Pat Moriarty, NREL).
- WP2: Benchmarking wind farm flow models on intended use (Javier Sanz Rodrigo, SGRE).
- WP3: VV&UQ framework adoption (David Maniaci, SNL).
Activities
Consistent with the philosophy established in Task 31, the main body of collaborative work will be based on the organization of model inter-comparison benchmarks. This is where model developers and end-users meet to test their models and reach consensus about model evaluation methodologies. Upon publication of the benchmark results in peer-review articles, the benchmark documentation is integrated in WEMEP with links to open-access repositories of datasets and evaluation scripts.
Benchmarks are typically supported and organized within national or European research projects that want to leverage modelling experts from the Wakebench community. A bottom-up approach is typical, where benchmarking opportunities are presented as they become available from partner projects. During Phase 3, these benchmarks have followed the validation strategy established in the New European Wind Atlas (NEWA) project and the U.S. Department of Energy Atmosphere to Electrons (A2e) program. In addition, industry-driven benchmarks have been organized with the Offshore Wind Accelerator (OWA) and the IEC 61400–12–4 to test engineering models at array efficiency prediction and numerical site calibration respectively. Examples of recent benchmarks include: OWA Wake Modelling Challenge on array efficiency prediction, SWiFT wake evolution and dynamics or the GABLS3 diurnal cycle in flat terrain.
In Phase 4 we will continue exploiting data from NEWA and A2e on multi-scale flow modelling in complex terrain and wind farms respectively, collaborate with Task 29 around the DanAero rotor aerodynamics experiment and with the German X-Wakes project on offshore mesoscale coastal gradients and farm-farm interaction effects. These high-fidelity experiments allow us to improve the physical insight of flow models by addressing phenomena that has high-impact potential in improving the predictive capacity of flow models. WP1 will further the collaboration with the academic community to coordinate benchmarks on phenomena like: global blockage, turbulence intensity from inflow to wake-added, coupling of ABL and wind farm wakes at varying stability levels, offshore mesoscale wind resource gradients and farm-farm interaction, meso-micro coupling in complex terrain, etc.
The science-push benchmarking of WP1 will be complemented with the market-push approach of WP2. Here, the objective is to test flow models in the assessment of quantities of interest that are relevant for energy yield assessment, wind turbine siting and wind farm design. This will be done in collaboration with industry to challenge models in the prediction of long-term statistics from wind resource campaigns and operational wind farms. Therefore, the objective of WP2 benchmarks is to quantify the added value of the advanced models validated in WP1. Examples of this type of benchmarks in Phase 3 were the Offshore Wind Accelerator Wake Modelling Challenge (OWAbench) on the prediction of array efficiency for 5 offshore wind farms and the Alaiz Numerical Site Calibration (AlaizNSC) benchmark to support the activities of the IEC 61400–12–4 standard.
WP3 is devoted to building the VV&UQ framework that constitutes the foundation for an internationally coordinated validation strategy for wind farm flow models. Here is where a protocol for model evaluation is defined (WEMEP) and then applied to atmospheric and wind farm flow models following a building-block approach that connects benchmarks along a hierarchy of increasing complexity. While the building-blocks for atmospheric “wind” models is already in good track with a solid base around the NEWA experiments, Phase 4 will significantly improve the definition of the “wake” modelling building-blocks by connecting DanAero, SWiFT, AWAKEN and X-Wakes experiments.
As the framework builds momentum by adding benchmarks, the focus should be placed on industry-wide adoption. Adoption is realized when the framework is used to validate models that are then integrated in wind energy design tools in compliance with IEC standards, in particular: IEC-61400–15–1/2 on energy yield and site suitability assessment and IEC-61400–12–4 on numerical site calibration. These standards mention the use of flow models but there is lack of a formal model evaluation process that could be used to guarantee quality-assurance from these models. The IEA-Wind offers the best forum to reach consensus between end-users and model developers on model evaluation methodologies and application-specific metrics and criteria to formalize quality-assurance in IEC standards.
While WEMEP provides context for model evaluation best practices, a more actionable process is achieved through software implementation. In previous phases some benchmarks have published the evaluation scripts in open-source python repositories but each one was developed independently of the others. The next objective is to use the lessons learnt to build a community model evaluation library that can be imported as a python package in forthcoming benchmarks to facilitate the process and guarantee a consistent definition of data structures, quantities of interest, metrics and plots to quantify and visualize the benchmark results.
Having a traceable model evaluation process enabled by a community code facilitates transparency and scaling at industry level. The code could be used internally by institutions to leverage anonymized validation results and insights from private data to improve the statistical significance of the framework without the need to provide access to sensitive data.
In addition to open-source code development, we will support the FAIR philosophy of making benchmark data and Task results findable, accessible, interoperable and reusable by publishing datasets in Zenodo and other institutional open-access repositories.
Schedule
A draft of the new Task proposal will be presented at the next IEA-Wind TCP ExCo by mid-May 2021, at the same time we present the final management report of Task 31 that also closes in May. We plan to start the new Task in Q4–2021 or Q1–2022 and use the interim period to fine-tune the proposal as we engage participants and collaborations with other IEA Tasks. The Task will have a duration of four years.
Further Reading
You can sign up to The Wind Vane / Wakebench mail list here.
