Predictive and stochastic reduced-order  modeling of wind turbine wake dynamics

Andersen, Søren Juhl; Murcia Leon, Juan Pablo

doi:https://doi.org/10.5194/wes-7-2117-2022

Articles | Volume 7, issue 5

https://doi.org/10.5194/wes-7-2117-2022

© Author(s) 2022. This work is distributed under
the Creative Commons Attribution 4.0 License.

https://doi.org/10.5194/wes-7-2117-2022

© Author(s) 2022. This work is distributed under
the Creative Commons Attribution 4.0 License.

Articles | Volume 7, issue 5

Research article

|

26 Oct 2022

Research article |

| 26 Oct 2022

Predictive and stochastic reduced-order modeling of wind turbine wake dynamics

Søren Juhl Andersen and Juan Pablo Murcia Leon

Download

Final revised paper (published on 26 Oct 2022)
Supplement to the final revised paper
Preprint (discussion started on 08 Jun 2022)
Supplement to the preprint

Interactive discussion

Status: closed

RC1:
'Comment on wes-2022-45', Anonymous Referee #1, 03 Jul 2022

The authors present a predictive and stochastic reduced-order model based on proper orthogonal decomposition for wind turbine wake flow application. The authors did an excellent job explaining the methodology, which I expect will be used by others in the field. The paper should definitely be published. I just have a few minor comments and requests for clarifications.

1. The introduction section does not provide recent state of art in using POD and/or ROM for wind energy applications. Also, some of the motivations of the current work are already resolved by other studies. For example” Cluster-based probabilistic structure dynamical model of wind turbine wake”, and “Clustering sparse sensor placement identification and deep learning-based forecasting for wind turbine wakes”. There have been many more studies that should be mentioned here.

2. Can you provide details on the time resolution considered in evaluating the POD? How did the authors avoid the correlated features in POD calculation?

3. Figure 7: It is helpful to add the energy profile of the POD eigenvalues to the figure to visualize the trend of both.

4. Fig 8: Can you explain why cases CT=0.422 and CT=0.578 showed less correlated Eigenvectors than the other cases?

5. Fig 13: Can you provide details on the energy content in these 50 modes?

6. It is very practical of the authors to explain the limitations of the current work. Please explain your insight on how to develop the current approach to capture the non-Gaussian trend and if we have yaw/ tilt mechanisms.

Citation: https://doi.org/10.5194/wes-2022-45-RC1
- AC1: 'Reply on RC1', Søren Juhl Andersen, 17 Aug 2022
  
  Please see response in attached file.
  
  Citation: https://doi.org/10.5194/wes-2022-45-AC1
RC2:
'Comment on wes-2022-45', Anonymous Referee #2, 23 Jul 2022

The article provides a model derived from LES to predict turbulent wind turbine wake dynamics, applied at two downstream locations within a long row. The model is built in a reduced parameter space denoting that the thrust coefficient indirectly encapsulates the effects of numerous other parameters of the flow, such as atmospheric turbulence intensity, spacing of the farm, etc. Proper orthogonal decomposition is employed to build the predictive and stochastic model.

Reasoning for why the model is relevant to the community and the general description of the methods are well described in the article. I believe the article is of importance and can benefit future turbulence modeling and recommend the article for publication with revisions based on the points outlined.

Specific comments:

Figure 2a – I believe the y axis (z/z0) should be nondimensional.

It is unclear how many snapshots are used for the POD, is it all 131,072 at 10Hz?

Why are 50 modes used for reconstruction? How much energy is associated with those 50 modes?

Why do the authors present the variance of the model time series in lieu of the eigenvalues? It is noted that the decreasing trends are similar but do the trends between the cases show similar results as well? For the near and far-downstream turbines?

Is there any reasoning for the skew of the PDFs for the first few modes of the farther downstream turbine and in turn, why this is not captured in the model?

Specific corrections:

Line 20: captures –> capture

Line 22: particular –> particularly

Line 62: of model by –> of the model by

Line 229: arise, particular between –> arise, in particular between

Line 248: values affects –> values affect

Citation: https://doi.org/10.5194/wes-2022-45-RC2
- AC2: 'Reply on RC2', Søren Juhl Andersen, 17 Aug 2022
  
  Please see response in attached file.
  
  Citation: https://doi.org/10.5194/wes-2022-45-AC2

Peer review completion

AR: Author's response | RR: Referee report | ED: Editor decision | EF: Editorial file upload

AR by Søren Juhl Andersen on behalf of the Authors (17 Aug 2022) Author's response Author's tracked changes Manuscript

ED: Referee Nomination & Report Request started (26 Aug 2022) by Rebecca Barthelmie

RR by Anonymous Referee #1 (27 Aug 2022)

RR by Anonymous Referee #2 (27 Aug 2022)

ED: Publish as is (09 Sep 2022) by Rebecca Barthelmie

ED: Publish as is (13 Sep 2022) by Sandrine Aubrun (Chief editor)

AR by Søren Juhl Andersen on behalf of the Authors (23 Sep 2022) Manuscript

Download

Article (5442 KB)
Full-text XML

Short summary

Simulating the turbulent flow inside large wind farms is inherently complex and computationally expensive. A new and fast model is developed based on data from high-fidelity simulations. The model captures the flow dynamics with correct statistics for a wide range of flow conditions. The model framework provides physical insights and presents a generalization of high-fidelity simulation results beyond the case-specific scenarios, which has significant potential for future turbulence modeling.