the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 17 Feb 2023
Potential of Dynamic Wind Farm Control by Axial Induction in the Case of Wind Gusts
Florian Bürgel
Robert Scholz
Christian Kirches
Sebastian Stiller
Abstract. Wind turbines organized in wind farms will be one of the main electric power sources of the future. Each wind turbine causes a wake with a reduced wind speed. This wake influences the power of downstream turbines. Therefore, there is a strong interaction between the individual wind turbines in a wind farm. This interaction is an opportunity for optimal control to maximize the total power and decrease the load (i.e., tower activity and pitch activity) of a wind farm. We use the already known axial-induction-based control but investigate its potential in the case of a wind gust using mathematical optimization. This case is particularly interesting because a wind gust requires a dynamic control reaction and the consideration of the time delay with which downstream turbines are affected. In particular, this enables to reduce the tower load of downstream wind turbines by dynamic axial-induction-based control of an upstream turbine.
- Preprint
(4746 KB) - Metadata XML
- BibTeX
- EndNote
Florian Bürgel et al.
Status: final response (author comments only)
-
RC1: 'Comment on wes-2023-2', Anonymous Referee #1, 04 Mar 2023
The presented paper aims to explore the use of a dynamic wake model for wind turbine load mitigation during a gust event. The topic is interesting and relevant to the current research, as the dynamics between wind turbines have not been fully explored and exploited yet.
Unfortunately, the quality of the presented work is not fit for publication in my eyes: The presented manuscript has significant shortcomings in the areas methodology, linguistic quality and visual communication. All three are explained in further detail below. I hope that the authors are not discouraged by my comments and that they can serve to improve the presented work and future publications.
Methodology
The main shortcoming of the methodology is the simulation tool: The authors choose to run their optimization in the WinFaST environment, a code which is rooted in the 2014 published FLORIDyn code by P. Gebraad. These kinds of parametric, dynamic models have the advantage of a fast computational time at the cost of fidelity. Recent publications have explored how turbine loads can accurately be modeled in similar, efficient frameworks [1], but this does have strong limitations. With this in mind, the model WinFaST is only cited as a footnote, and no validation study is linked, which would confirm WinFaST’s suitability to predict loads. This would have been fine if a validation study would have been conducted, with a proven tool, but this is not the case. State-of-the-art tools are not even mentioned in the text: Open-source tools such as FAST.Farm [2] are public to download, have the capability to run on a regular PC and have been thoroughly validated for load prediction. Another, similar tool is HAWC2Farm [3].
The model choice further becomes questionable when reading about its limitations to model wind gusts (page 14), the central element of the optimization challenge. The implementation of a wind field model, similar to the one employed in [4], could have possibly improved the setup.
The topic of wind gust modeling is also almost neglected in the manuscript. The presented work would have profited from a discussion around the topic and a motivation for the chosen options. The relevance of wind turbine design and the impact of gusts is a complex topic [5]. And since the title of the presented work is hinting towards an investigation of the topic, it should be treated as such.Many of the points above could have been deemed acceptable if the paper would have introduced a novel approach to the optimization problem and its solution. But the authors do not go to the lengths of understanding and exploiting the characteristics of their optimization problem and choosing an adequate optimizer. Instead, the optimizer choice discussion is reduced to a footnote and the application of a standard Matlab function.
In summary, the authors chose to research an interesting and complex research question but (i) fail to adequately motivate the choice of simulation tool and its legitimacy, (ii) they miss to express knowledge about the researched flow phenomena and justify the modeling of it and (iii) they do not conduct a thorough investigation of optimization algorithms and strategies.
As last remark, I would have liked to see a more in-depth review of the existing literature around wind turbine load mitigation and recent efforts, see [6] and more recently [7].
Linguistic quality
The linguistic quality of the paper is not fit for publication. The text reads as a loose collection of paragraphs, repeats itself frequently and is hard to follow. In my opinion, the paper needs to be rewritten in a much more concise way and without the use of colloquial language.
Visual Communication
The presented figures are not polished and ready for publication. Flaws include, but are not exclusive to,
- lack of proper titles (e.g. GenPwr is not a title, this should have been “Generator Power (MW)” or similar),
- inconsistent axis limits (if two plots of the same quantity of interest are next to each other, they need to have the same y-axis limits to be comparable),
- arguably no added content (Fig. 4 features almost exclusively straight lines but 8 lines of caption).
The authors also miss to communicate complex topics visually. One figure, which would have added a valuable overview is an input/output diagram of the optimalization setup.
[1] Efficient Loads Surrogates for Waked Turbines in an Array,
K. Shaler et al.
https://iopscience.iop.org/article/10.1088/1742-6596/2265/3/032095/meta[2] FAST.Farm User’s Guide and Theory Manual
J. Jonkman, K. Shaler
https://www.nrel.gov/docs/fy21osti/78485.pdf[3] LES verification of HAWC2Farm aeroelastic wind farm simulations with wake steering and load analysis
J. Liew et al.
https://iopscience.iop.org/article/10.1088/1742-6596/2265/2/022069/pdf[4] A Meandering-Capturing Wake Model Coupled to Rotor-Based Flow-Sensing for Operational Wind Farm Flow Prediction
M. Lejeune et al.
https://www.frontiersin.org/articles/10.3389/fenrg.2022.884068/full[5] Extreme gusts and their role in wind turbine design
R. Bos
https://research.tudelft.nl/en/publications/extreme-gusts-and-their-role-in-wind-turbine-design[6] Wind Turbine Control for Load Reduction
E. Bossany
https://onlinelibrary.wiley.com/doi/epdf/10.1002/we.95[7] Wind farm control for wake-loss compensation, thrust balancing and load-limiting of turbines
J.G. Sliva et al.
https://www.sciencedirect.com/science/article/pii/S0960148122017657Citation: https://doi.org/10.5194/wes-2023-2-RC1 -
RC2: 'Comment on wes-2023-2', Anonymous Referee #2, 12 Mar 2023
The manuscript proposes a dynamic control strategy targeted at load alleviations and power optimization during gust events. Though the topic treated is of interest, the manuscript suffers from some major shortcomings that make it unfit for publication.
METHODOLOGY
General comments
The optimization of the transient response of the controller is performed using WinFaST, a Matlab implementation of FloriDyn. Though this tool indeed provides a dynamic description of the flow, it suffers from an array of limitations that are not appropriately addressed.
To begin with, the absence of proper validation study of WinFaST in the context of load modelling is concerning as the topic of load mitigation is core to this paper. If such a validation study does actually exist, it should be clearly cited.
Further, the authors themselves acknowledge the incapacity of the initial WinFaST implementation (L295) to model wind gusts. A work around is presented but not validated.
Together, these two elements make the choice of WinFaST questionable in the context treated. I would recommend the authors to consider higher-fidelity tools such as HAWC2Farm or FAST-Farm that provide accurate yet computationally affordable environments for wind farm simulations. Additionally, to being publicly available, these tools have been extensively validated and are therefore better candidates for modeling wind farm transients.
This first comment regarding the choice of the wind farm simulation environment also reflects a more general lack of literature study. The overview of state-of-the art wake models is incomplete and not up-to date. It fails mention many recent studies in the field of (dynamic) wake modeling. These notably include the latest developments of the FLoriDyn framework despite it being at the center of the optimization strategy presented. The topic of gust avoidance is not covered in enough details neither while, on the other hand, a paragraph is devoted to wind farm layout optimization even though it is not directly related to the subject treated.
Specific comments
L235 Stating that you have turbulence is arguable as the TI is very low (ie: 1%)
L160 The effect of the control inputs of WinFast is loosely presented.
L570 The section conclusion needs to be improved and to notably feature a more quantitative analysis.
STYLE AND FORMAT
The writing style is not satisfactory and needs to be improved: the sentences are confused and the text generally lacks continuity. The sections are structured as series of successive unrelated paragraphs thereby making the paper hard to follow. The authors also introduce some unnecessarily complicated notations. Specifying from the beginning that the study will consider a pair of wind turbines would alleviate the need for vector notation and overall facilitate the comprehension of the paper. Likewise mentioning the yaw angle is not helpful as no yaw optimization is performed.
The quality of the figures needs to be improved. Figures require informative titles: PowerRefCtrl is not an appropriate figure title. Similarly, axis should be scaled consistently and their labels should be indicated on the figure rather than in the legend. Fig 1 is a screenshot and is not acceptable as such, it should be properly formatted and exported. The Appendix do not bring additional value: the tables are bulky and generally hard to read. I would suggest accompanying these tables with recapitulative diagrams for better readability.
Citation: https://doi.org/10.5194/wes-2023-2-RC2 -
AC1: 'Comment on wes-2023-2', Florian Bürgel, 15 Apr 2023
Both referee comments (RC1 and RC2) are critical but fair and constructive, which we explicitly appreciate. We realize that it is most purposeful to replace the wake model as well as the wind gust model for future work on this interesting topic.
METHODOLOGYGeneral comments
Our model selection was based on a tradeoff between computational speed and accuracy. Finally, it was too strong in terms of speed, i.e., as suggested by both referees, we have to use a higher-fidelity tool.
We welcome the advice to update and expand the related work to include wind gust modeling (RC1), load mitigation (RC1), state-of-the-art wake models (RC2) and gust avoidance (RC2).
Specific comments
Our approach to the optimization problem would benefit from the exploitation of physical characteristics of the problem---on this point we agree with RC1. However, in a rudimentary way we have already taken characteristics into account: First, the boundary knots of the spline interpolation are fixed by previous optimization (for low and high wind speed) (L240 and L280). Second, the initial guess of the intermediate knots was done on suggestions with results of (slow) global optimization (L315). As the used SQP is not a global optimization method, an adequate choice of initual guess is required---we should have mentioned that. An analog approach with a higher-fidelity simulation tool should be purposeful, whereas a new, sophisticated optimizer (at best in real time) for the problem is a really difficult challenge.
The problem of low turbulence intensity (1%), that was mentioned in RC2 (specific comment to L235), can be eliminated with a model replacement too. As mentioned in Footnote 4 the choice was influenced by model restrictions: if a wind gust is from 11 to 12 m/s, it does not make sense to chose a turbulence intensity of 10%.
The specific comment in RC2 to L160 probably does not address the yaw reference but the power reference and fatigue reference. Both are deliberately without details as the exact realizations are part of WinFaST, see lines 146--147 and Footnote 5, and therefore it is unfortunately not public. We probably should have made this clearer.
Of course, we can add a more quantitative analysis as suggested in RC2 with regard to L570.
LANGUAGE
We are sorry that the manuscript was hard to follow (RC1 and RC2). We need to find a better order to avoid repetitions (RC1) and ensure continuity (RC2). In that context we would like to thank for the advice to remove unnecessary like wind farm layout optimization (RC2), yaw optimization (RC2), and complicated notation (RC2).
Note that we can not cite WinFaST in the recerence list instead of the footnote (see RC1) as unpublished works should not be cited in the reference list (according to the guidelines of the journal).
FIGURESThe mentioned improvements of both referees for the figures are highly welcome (informative titles, axis limits in RC1 and RC2; replace Fig. 1 in RC2). In principle, we advocate axis labels (RC2) but see it critically due to the space required.
We would like to thank for the idea in RC1 to create an input/output diagram to keep the overview. This would work out the connections between the scenarios (influcence of 1 and 2 on 3) better than the criticized Fig. 4.
We also appreciate the idea in RC2 to add diagrams to the tables in the appendix to make them more comprehensible.
Citation: https://doi.org/10.5194/wes-2023-2-AC1
Florian Bürgel et al.
Data sets
Dataset for Article: Potential of Dynamic Wind Farm Control by Axial Induction in the Case of Wind Gusts Florian Bürgel, Robert Scholz, Christian Kirches, Sebastian Stiller https://doi.org/10.5281/zenodo.7520545
Florian Bürgel et al.
Viewed
| HTML | XML | Total | BibTeX | EndNote | |
|---|---|---|---|---|---|
| 613 | 111 | 18 | 742 | 8 | 7 |
- HTML: 613
- PDF: 111
- XML: 18
- Total: 742
- BibTeX: 8
- EndNote: 7
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1