the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 08 Dec 2025
Combining wake steering and active wake mixing on a large-scale wind farm
Abstract. Wind farm flow control mitigates wake effects within a wind farm by adjusting the turbine settings to improve the overall farm performance rather the output of each turbine. Wake steering is an established approach while active wake mixing has recently emerged as a promising solution. This study quantifies the value of a combined strategy, in which each turbine can apply wake steering or the helix. The analysis is performed considering different degrees of uncertainty in wind direction using engineering wake models which enable the simulation of these techniques on large-scale wind farms. A novel optimization algorithm, called Multi-strategy Serial-Refine (MSR), is developed in this study, extending the state-of-the-art method for yaw optimization to multiple control strategies and a generalized objective. A scaled version of an offshore wind farm in the Netherlands is selected as the case study, consisting of 69 IEA 22 MW turbines. The proposed combined strategy yields a 1.98 % increase in annual energy production compared to the baseline scenario, in contrast to the 1.68 % and 1.15 % gains obtained by applying only wake steering or the helix, respectively. This trend persists under wind direction uncertainty. Given the pronounced sensitivity of wake steering to such uncertainty, the combined strategy takes advantage of the superior robustness of the helix under these circumstances.
Competing interests: At least one of the (co-)authors is a member of the editorial board of Wind Energy Science.
Publisher's note: Copernicus Publications remains neutral with regard to jurisdictional claims made in the text, published maps, institutional affiliations, or any other geographical representation in this paper. While Copernicus Publications makes every effort to include appropriate place names, the final responsibility lies with the authors. Views expressed in the text are those of the authors and do not necessarily reflect the views of the publisher.- Preprint
(4883 KB) - Metadata XML
- BibTeX
- EndNote
Status: final response (author comments only)
-
RC1: 'Comment on wes-2025-265', Anonymous Referee #1, 01 Mar 2026
This study uses an engineering wake model tuned from large eddy simulations (LES) to assess the benefit of using both wake steering and helix control on a turbine-by-turbine basis, allowing a turbine to actuate one of the methods at any given time. An optimization algorithm was developed to determine the optimal strategy and its parameters, and this was used to make predictions for a turbine pair and a real-world wind farm layout. Results show that a combined optimization strategy is worth pursuing because the benefits outweigh either strategy on their own.## Comments1. Why is wake steering considered quasi-static? Don't yaw misalignments need to be dynamic as wind conditions change?2. It could be argued that wake steering is still not very mature, as it is rarely applied on real world wind farms.3. Perhaps the title would benefit from noting that the helix method was used, not the more general active wake mixing concept, which would include the pulse method. "Combining wake steering and helix wake mixing on a large-scale wind farm". The title could also benefit from clarification that this is a modeling study using an engineering wake model, not LES or an experimental study.4. It would be helpful to clarify exactly which version of PyWake was used to perform the calculations.5. Can this model account for nonuniformity in wind speed and direction across the farm on the quasi-steady (~10 minute) timescale?6. Please comment on the generality of the results taken at a constant 4% turbulence intensity. How realistic is this? If the gains are nonlinear with respect to TI, perhaps simulating more of the actual distribution is necessary.7. Please comment on how the wind direction uncertainty used for simulation models either natural variability on timescales shorter than the control system reaction time or sensor error. Does it model either source of uncertainty? If so, why are they equivalent? For example, sensor bias may be much more detrimental than noise or other random error. Can or should deficiencies be separated out by category? What is a typical value for a real-world wind farm?8. Tuning of the engineering wake model may be worth including as a major section in the paper rather than an appendix. Perhaps include some information regarding the uncertainties of the values derived from the calibration process as well. LES and OpenFAST are not observations of the absolute truth of course. How might these uncertainties propagate through to the final AEP gain calculations?9. Sharing code and data is much appreciated. Please include in the README the steps that must be taken with the repository to compute the EWM calibration values, compute the AEP at all the discretized points, compute the integrated AEP, and generate the figures and numerical results. The data should be stored in an open format like CSV, HDF5, or Parquet, as Python Pickle files can represent security risks. Lastly, before publication please deposit the compendium in a long term archival service like Figshare or Zenodo and provide the digital object identifier (DOI) in the paper references.## Minor comments1. Line 29: "the pulse.": Is this a typo? Should it read "the pulse technique" or "the pulse method"?2. Line 87: "the helix": Similarly, should this read "the helix method" or "helix mixing"?3. Line 34: "static" should be "quasi-static"?4. Line 376: "as a results": Should be "as a result".5. Line 384: "performing the helix": Should be "operating in helix actuation mode" or similar.6. Section 2.5 could be named "Cases studied".7. Why are the COT values doubly negative in figure 9?8. Line 453: "The helix" used again with no additional nouns like "method" or "approach". "Helix control" could also be used in lieu of "the helix".Citation: https://doi.org/
10.5194/wes-2025-265-RC1 -
RC2: 'Comment on wes-2025-265', Anonymous Referee #2, 29 Apr 2026
The manuscript presents a valuable and as far as I am aware unique analysis of the benefits of applying a combination of wake steering and wake mixing on turbines in a large-scale wind farm. The analysis is executed using a low-fidelity engineering model, which is both the strength and the weakness of this work.
The manuscript is a valuable addition to existing research in the field of WFFC. However, I believe the manuscript in its current form has some significant shortcomings that need to be addressed before publication.
I was torn between recommending minor or major revisions, so please do not interpret my final choice to mean that I do not like the manuscript. I actually really liked the manuscript, which is why I wrote such an extensive review. I believe this manuscript is good work that has the potential to be great work if some of the shortcomings are addressed. I do not doubt the authors’ ability to address my comments though, and when they do, I will gladly recommend this manuscript to be accepted.
First and foremost, I believe that the fidelity of the engineering model substantially puts into question the reliability of the final results. I have the following concerns with the engineering model that I feel need to be addressed or at least mentioned:
- The model is used (far) outside of the space it is tuned for. As far as I can tell, the model is tuned for one specific boundary layer condition, one wind speed, one TI percentage, and only for up to three wind turbines with only one specific spacing. It is specifically NOT tuned for partial wake overlap, deeper arrays, or control actions implemented on downstream (waked) turbines. However, the case study does involve all of these conditions. This daisy chain of extrapolations makes that the uncertainty of the model, which is not quantified in any way in the paper, likely far exceeds the reported AEP gains for the case study wind farm. The fact that the model was tuned without any of these dynamics should explicitly be mentioned throughout the paper.
- I know neutral boundary layer conditions are common in the North Sea, but it is a significant simplification to consider this representative of the year-round conditions in the case study wind farm. The same is true for the wind speed and turbulence levels. I can see how 10 m/s and 3-6% might be average for the North Sea, but “average” is not the same as “representative”. I recommend adding a Weibull distribution of the wind speed measured at the wind farm, accompanied by the TI range for each wind speed bin. This will likely show that the TI is generally higher at low wind speeds, when WFFC in general, and the helix specifically, is most effective. Therefore, by tuning for a relatively low TI, you might overestimate the benefit of the helix, which should be mentioned. You do say you use the Weibull distribution to calculate the AEP, so regardless of this point, I feel like a figure of the distribution should be added to the paper.
- Currently, the flow model validation is limited to two figures tucked away in the appendix. This should be a major part of the paper, as the reader needs to be able to assess the fidelity of the model to put any trust in the reported findings. The results in Section 3.1, using the 2-turbine layout, are perfectly suited to compare the engineering model with LES data. You should be able to produce LES equivalents of Figure 4 at different control settings (using rotor-averaged wind speeds) to compare with the model. This is the “simplest” case that is basically exactly what you tuned your model for, so this provides a floor on the error that should be expected for the wind farm case. I believe this needs to be added to the revised manuscript.
- I understand that LES runs are timely and costly, so although I believe more simulations at different conditions are necessary to sufficiently tune the model, I respect that you have to deal with a limited amount of (computational) resources. However, this should be mentioned as a limitation of the study much more explicitly. It should be made (more) clear to the reader that the case study far exceeds the conditions for which the model was validated, and that the results should be interpreted accordingly.
Other major comments:
- The quality of the written text is fair, not great. Many sentences do not flow naturally, the order of words is sometimes unintuitive, the interpunction is not always logical, and the choice of words is not always fluent (and this is coming from a non-native English speaker). In my opinion, it is good enough, the manuscript is certainly understandable in its current form, but I would personally strive for a higher standard. I therefore recommend having the manuscript reviewed by a professional editor or a native English speaker.
- I would suggest changing the title of the paper to improve expectations. When I first read the title, I assumed 1) that wake steering and wake mixing were going to be applied on turbines simultaneously (about which I was skeptical), and 2) that different wake mixing strategies would be studied. I quickly learned that neither was the case, but I would suggest changing the title so that it better reflects this and better guides the expectations of the reader.
- Why do you use the IEA 22MW turbine? Unless my Googling skills betray me, the case study wind farm HKN is equipped with 11MW turbines. However, you seem to conveniently leave this out of your manuscript. Wouldn’t it make much more sense to use, for example, the DTU 10MW reference turbine (or even the IEA 15MW turbine), as it is much closer in size to the actual wind turbines used? I assume you had a practical reason to use this specific reference turbine, which is understandable, but should be justified. The subsequent scaling of the whole farm adds another layer of uncertainty to the already highly uncertain final conclusions of the paper. Did you also scale the wind speed to correspond to the same point on the turbine wind speed-power curve?
- You do not mention whether controller behavior changes between region II and III. I therefore assume it doesn’t. However, up until now, as well as in your LES cases, the helix approach is only studied in region II. If your study does not limit helix to region II, you should mention this as an additional degree of uncertainty.
- I really like the approach taken to consider the uncertainty in wind direction using probability distribution. I think this might be the most valuable contribution of this paper. However, I do have a few smaller remarks with respect to this approach:
- I assume this uncertainty does not propagate downstream, is that correct? In other words, you only consider unintentional misalignment for turbine power production, not for how it might affect the wake (unintentional wake steering). I understand that including this would be much more complicated, so I do not expect you to add that, but I think it is worth mentioning as this could further affect the performance of WFFC, specifically wake steering.
- Did you consider using a non-zero mean for the probability distribution? I believe that sensor bias is a common issue for wind turbine anemometers, so this could be relevant to study. How might that affect your findings?
- Why did you choose the uncertainty levels of 2.5 and 5 degrees, if literature shows that values of up to 5.25 degrees are observed in the field?
- Finally, and this is my only major objection to the method used, you assume different levels of uncertainty, but at the same time, you also assume perfect knowledge of the level of uncertainty when choosing the control actions. However, in a real wind farm, you do not know the level of uncertainty a priori, and it usually changes over time. How would your results change if you do NOT have perfect knowledge of the uncertainty? I would really like to see a section added to the paper where you use the LUT for 0 degree uncertainty, but then implement it on the case with 5 degree uncertainty, and compare that to the case where you had perfect knowledge (for both the 2T and the HKN case). I feel like this better represents how WFFC is implemented on real wind farms, and requires only a limited additional simulation effort. This would in my opinion better show the effect of wind direction uncertainty on AEP than your current approach.
- Given all the uncertainty regarding the reliability of the AEP gains as addressed in multiple points above, I would seriously consider not mentioning the specific percentages at all. I think it would be better to stick to the more robust RELATIVE conclusions that wake steering works better with no WD uncertainty, helix with WD uncertainty, and a combination of both works best. That is in my opinion still a valuable finding, that is much less likely to be rebuked by future research using a better / more thorough methodology. At the very least, I would remove the percentage claims from the abstract and conclusions sections, or add lengthy caveats mentioning all the layers of uncertainty surrounding these numbers.
- I do not fully understand why wake steering performs as badly as it does in your simulations when WD uncertainty is considered. I would expect a more significant drop than for the helix approach, as is observed, but I hadn’t expected the difference to be this big. This is also supported by section 3.1, which is much more in line with my expectations. I think this surprising result, and discrepancy between a simple 2-turbine setup and a complex 69-turbine wind farm, deserves deeper digging into. If these results are to be believed, the simple 2- or 3-turbine setups commonly used in assessing WFFC are actually terribly bad at predicting the benefit of WFFC strategies. That would be a very significant finding. However, as the analysis of this discrepancy is limited in the current manuscript (and my limited belief in the fidelity of the model), I am not sure if I can fully trust these results.
- I would be very interested to see where the wind farm AEP gains originate from. Would it be possible to add a plot showing how much each turbine within the farm gains/loses power with respect to baseline?
- Related to the above, figure 7b indicates to me that a significant part of the gain of helix is obtained by using it on downstream turbines. As you briefly mention yourself in the discussion, this is still very much uncharted territory and requires synchronization. Your model is also not tuned for waked turbines using the helix approach, not to mention waked turbines where the upstream turbine is also using the helix approach. I therefore feel like I need to know how much of the AEP gain can be explained by waked turbines using the helix. Could you please run a simulation in which only freestream/upstream turbines are allowed to use the helix, and compare the AEP gain with the situation in which all turbines are allowed to use it? I do not think this simulation needs to be added in detail to the manuscript, but I do think it is important to quantify what portion of the AEP gain comes from downstream turbines using the helix approach.
Minor comments:
Line 2: rather THAN the output
Line 2: an established WIND FARM FLOW CONTROL approach
Line 4: you should define/explain the helix here, or at the very least specify that this is the wake mixing strategy that you are studying in this paper.
Line 19: I would say wake steering is THE most effective solution, or at the very least the most established solution.
Line 30: “These concepts have …”, I think you mean the helix here, as all the citations refer to helix studies. If so, it should be singular (“This concept has …”)
Introduction: I know this is a personal preference and I wouldn’t say it’s wrong, but I don’t love the use of present perfect tense throughout the introduction. You are at least consistent, but I would personally prefer past tense or even present tense (“Citation A showed that …” instead of “Citation A has shown that …”).
Line 43: It is fair to say that the analysis limits itself to two-turbine arrays, but in your study, you also only do LES for three-turbine arrays. I would therefore argue that your findings have similar limitations.
Line 80: “… in reaching the global optimum.” Please add a source for this claim.
In general: Why did you choose to specifically study only the helix approach and not DIC? I have no issue with the choice, but it couldn’t hurt to justify it.
Line 90-94: Make the contributions listed here full sentences please. Also, I wouldn’t split them up into two lists, the “additional contributions” are in my opinion just as much contributions as the main ones.
Line 97-99: Use “Section” or “Sect.”, but be consistent.
Line 106 “The effect of wake steering …”: add “on the upstream/waked turbine”.
Line 130 / Appendix A: In my opinion, the process of tuning the model is fine to be moved to an appendix. However, the results of the tuning, i.e., how well your model matches the higher fidelity LES data, is very important in assessing the reliability of the results, and should be moved to the main part of the manuscript.
Table A1: “inflow”, not “infow”
Section 2.5: I would move this section forward, as I felt like I already needed this information to understand some of what was mentioned in previous sections.
Figure 3&4: consider trimming the y-axis, at least for the first two rows, to show a more relevant/zoomed representation of the findings.
Figure 7a: what is the downstream distance between the controlled turbine and the nearest downstream turbine in the 225 degree-sector where helix is implemented? How much does this downstream turbine benefit from the shown turbine using the helix? Does that correspond to the LES results?
Line 381-383: You only mention one of the many dimensions in which you are extrapolating LES simulations here. This section should include a much more extensive summary of ways in which you are extrapolating results.
Line 451 “with a specific focus on the helix technique”: a specific focus implies that you also looked at other wake mixing techniques, which you didn’t.
Table A2 (1): For tuning the turbine model, did you use LES simulations or standalone OpenFAST simulations? If you did the former, you could use these simulations to compare optimal values found with the pyWake model, at least in the case of a 2-turbine setup, to validate the match.
Table A2 (2): As part of the model verification, please provide a comparison of turbine powers between the 3-turbine LES cases and the exact same setup in pyWake. As mentioned above, this would function as a floor on the modeling error and helps the reader assess the reliability of results in the more complex case.
Appendix B: It seems to me that the current algorithm uses unnecessary repetitions of the same simulations. Perhaps that doesn’t matter since the simulations are cheap, but if that is the case, then why do you not take more iteration steps than the 3 mentioned in Table 1 to assure convergence? Furthermore, I am surprised to see that smaller values are found to be optimal for the helix method. How does this compare with the results from the LES simulations?
Citation: https://doi.org/10.5194/wes-2025-265-RC2
Viewed
| HTML | XML | Total | BibTeX | EndNote | |
|---|---|---|---|---|---|
| 457 | 279 | 29 | 765 | 39 | 33 |
- HTML: 457
- PDF: 279
- XML: 29
- Total: 765
- BibTeX: 39
- EndNote: 33
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1