the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 02 May 2024
Evaluating the potential of wake steering co-design for wind farm layout optimization through a tailored genetic algorithm
Abstract. Wake steering represents a viable solution to mitigate the wake effect within a wind farm. New research that consider the effect of the control strategy within the layout optimization are emerging, adopting a co-design approach. This study estimates the potential of this technique within the layout optimization for a wide range of realistic conditions. To capture the benefits of such method, a genetic algorithm tailored to the layout optimization problem has been developed in this work, hence appointed as layout-optimization genetic algorithm (LO-GA). The crossover phase is designed to recognize and exploit the differences and the similarities between parent layouts whereas the randomness of the mutation is limited to improve the exploration of the design space. New relations have been introduced to calculate the geometric yaw angles based on the reciprocal positions between the turbines. For a base case of 16 turbines located at Hollandse Kust Noord site, a gain in the annual energy production between 0.3 % and 0.4 % is obtained when the co-design approach is adopted. This increases up to 0.6 % for larger farms, saturating after 25 turbines. The benefit of the co-design decreases if the power density of the wind farm is lower than 15 Wm-2 or if the wind resource is highly unidirectional. On the other hand, in case wake steering is not applied during the operation of the farm, a decrease in the AEP up to 0.6 % is registered for a layout optimized with the co-design method. To prevent the risk related to future decisions on the control strategy, a multi-objective co-design approach is proposed. This is based on the simultaneous optimization of the layout for a scenario in which wake steering is applied versus a case where wake steering is not adopted during the operation of the farm. We have concluded that the solutions obtained with this method ensure an AEP gain higher than 0.3 % for a 16-turbines farm while limiting the loss below 0.1 % in case wake steering is not applied.
- Preprint
(1172 KB) - Metadata XML
- BibTeX
- EndNote
Status: closed
-
RC1: 'Comment on wes-2024-50', Anonymous Referee #1, 13 May 2024
The paper describes a new way to tackle a combined wind farm layout and wake steering optimization problem by applying a genetic algorithm. With some clever methods to handle mutation and crossover, it is able to produce better results than with a basic implementation available in PyGAD, which has trouble converging. Moreover, the paper builds on the work of Stanley et al. (2023) to avoid a nested optimization and does so by proposing a new formulation for the geometric yaw relation. The AEP gains are in the order of 0.4%, which is significant for a commercial wind farm and worth pursuing. The authors explain the method very thoroughly and included a couple of helpful drawings to help the reader understand at key points in the story. It is also very valuable that the results are presented in a statistical fashion to show the random nature of the algorithms.
Nevertheless, I would like to make some suggestions to improve the paper. First, it would be good to explain (qualitatively) how the geometric yaw expressions depend on the assumed wake shape and possibly the wind turbine type. This could explain why the expression from Stanley et al. (2023) does not translate well to this study. The tuning parameter γmax could also benefit from some introduction, especially since it can be linked to physical actuation limits and may be a first step towards a multi-objective optimization with structural loads.
Furthermore, with 36 wind direction bins (i.e., 0, 10, …, 350°), there may be a danger that the optimizer is gaming the cost function by orienting the turbines along the edges of the wind direction bins (i.e., along 5, 15, …, 355°). Since the AEP increases are only < 0.5%, this could play a role. Have the authors checked the resulting layouts for such artefacts and whether they impact the result significantly?
The discussion section would benefit from some reflection on the 15-W/m² saturation point. Particularly whether it is a universal limit or case-specific – would one expect the same with different wind farm plot shapes, or is this simply driven by the turbine spacing and how much a wake can actually be deflected from the downwind turbine?
Some other minor points that may be considered for a next revision:
- There are a couple mix-ups between US vs. British English (e.g., recognize vs. recognise, favorable vs. favourable).
- For one who wants to reproduce the work, it would be helpful to add a reference to the HKN site conditions shown in Fig. 8 (they are public).
- The shaded areas in Fig. 10 are very light and difficult to read from screen.
- It is very difficult for a reader who quickly browses through the paper to understand what is meant with “Std of the probability of occurrence” (Fig. 16). I would suggest renaming the axis label to “wind direction variability” and/or spent a line in the caption to explain.
- It would be helpful to state what happens with the area of the wind farm plot under different power densities and higher numbers of turbines in Section 6.4.1 and 6.4.2, respectively. Now it is not immediately clear whether the wind farm plot increases in size with an increasing number of turbines.
- I feel that renaming the sections 6.4.2 and 6.4.3 to simply "Number of turbines" and "Wind direction variability" woud reflect the contents better. A browsing reader could also connect those better to the axis labels of Fig. 15 and 16.
Overall, nice work and I hope to see the final paper soon!
Citation: https://doi.org/10.5194/wes-2024-50-RC1 - AC1: 'Reply on RC1', Matteo Baricchio, 04 Sep 2024
-
RC2: 'Comment on wes-2024-50', Anonymous Referee #2, 05 Jul 2024
Thank you for this contribution! In general, writing is well done and figures are clear. References to literature is thorough.
The innovation of optimizing for the coupled condition that wake steering is or isn't applied to the co-designed farm is a valuable contribution.
Figure 17: I understand the pareto front for the multi-objective points, and for the co-design points, I understand they improve with wake steering, but lead to losses without. The sequential points are more mysterious to me, there are some that appear to lose AEP with and without wake steering applied?
The algorithm for the geometric yaw of Stanley is also implemented and publicly availble within FLORIS: (see for example here: https://github.com/NREL/floris/blob/main/floris/optimization/yaw_optimization/yaw_optimizer_geometric.pyand here: https://nrel.github.io/floris/examples/examples_control_optimization/006_compare_yaw_optimizers.html)
This could be interesting to point out. Figure 15 also gives the impression that these improvements can be fairly critical, and perhaps one of the improved algorithms from this paper could be submitted as a pull request back to FLORIS.
Citation: https://doi.org/10.5194/wes-2024-50-RC2 - AC2: 'Reply on RC2', Matteo Baricchio, 04 Sep 2024
Status: closed
-
RC1: 'Comment on wes-2024-50', Anonymous Referee #1, 13 May 2024
The paper describes a new way to tackle a combined wind farm layout and wake steering optimization problem by applying a genetic algorithm. With some clever methods to handle mutation and crossover, it is able to produce better results than with a basic implementation available in PyGAD, which has trouble converging. Moreover, the paper builds on the work of Stanley et al. (2023) to avoid a nested optimization and does so by proposing a new formulation for the geometric yaw relation. The AEP gains are in the order of 0.4%, which is significant for a commercial wind farm and worth pursuing. The authors explain the method very thoroughly and included a couple of helpful drawings to help the reader understand at key points in the story. It is also very valuable that the results are presented in a statistical fashion to show the random nature of the algorithms.
Nevertheless, I would like to make some suggestions to improve the paper. First, it would be good to explain (qualitatively) how the geometric yaw expressions depend on the assumed wake shape and possibly the wind turbine type. This could explain why the expression from Stanley et al. (2023) does not translate well to this study. The tuning parameter γmax could also benefit from some introduction, especially since it can be linked to physical actuation limits and may be a first step towards a multi-objective optimization with structural loads.
Furthermore, with 36 wind direction bins (i.e., 0, 10, …, 350°), there may be a danger that the optimizer is gaming the cost function by orienting the turbines along the edges of the wind direction bins (i.e., along 5, 15, …, 355°). Since the AEP increases are only < 0.5%, this could play a role. Have the authors checked the resulting layouts for such artefacts and whether they impact the result significantly?
The discussion section would benefit from some reflection on the 15-W/m² saturation point. Particularly whether it is a universal limit or case-specific – would one expect the same with different wind farm plot shapes, or is this simply driven by the turbine spacing and how much a wake can actually be deflected from the downwind turbine?
Some other minor points that may be considered for a next revision:
- There are a couple mix-ups between US vs. British English (e.g., recognize vs. recognise, favorable vs. favourable).
- For one who wants to reproduce the work, it would be helpful to add a reference to the HKN site conditions shown in Fig. 8 (they are public).
- The shaded areas in Fig. 10 are very light and difficult to read from screen.
- It is very difficult for a reader who quickly browses through the paper to understand what is meant with “Std of the probability of occurrence” (Fig. 16). I would suggest renaming the axis label to “wind direction variability” and/or spent a line in the caption to explain.
- It would be helpful to state what happens with the area of the wind farm plot under different power densities and higher numbers of turbines in Section 6.4.1 and 6.4.2, respectively. Now it is not immediately clear whether the wind farm plot increases in size with an increasing number of turbines.
- I feel that renaming the sections 6.4.2 and 6.4.3 to simply "Number of turbines" and "Wind direction variability" woud reflect the contents better. A browsing reader could also connect those better to the axis labels of Fig. 15 and 16.
Overall, nice work and I hope to see the final paper soon!
Citation: https://doi.org/10.5194/wes-2024-50-RC1 - AC1: 'Reply on RC1', Matteo Baricchio, 04 Sep 2024
-
RC2: 'Comment on wes-2024-50', Anonymous Referee #2, 05 Jul 2024
Thank you for this contribution! In general, writing is well done and figures are clear. References to literature is thorough.
The innovation of optimizing for the coupled condition that wake steering is or isn't applied to the co-designed farm is a valuable contribution.
Figure 17: I understand the pareto front for the multi-objective points, and for the co-design points, I understand they improve with wake steering, but lead to losses without. The sequential points are more mysterious to me, there are some that appear to lose AEP with and without wake steering applied?
The algorithm for the geometric yaw of Stanley is also implemented and publicly availble within FLORIS: (see for example here: https://github.com/NREL/floris/blob/main/floris/optimization/yaw_optimization/yaw_optimizer_geometric.pyand here: https://nrel.github.io/floris/examples/examples_control_optimization/006_compare_yaw_optimizers.html)
This could be interesting to point out. Figure 15 also gives the impression that these improvements can be fairly critical, and perhaps one of the improved algorithms from this paper could be submitted as a pull request back to FLORIS.
Citation: https://doi.org/10.5194/wes-2024-50-RC2 - AC2: 'Reply on RC2', Matteo Baricchio, 04 Sep 2024
Viewed
| HTML | XML | Total | BibTeX | EndNote | |
|---|---|---|---|---|---|
| 356 | 88 | 129 | 573 | 17 | 17 |
- HTML: 356
- PDF: 88
- XML: 129
- Total: 573
- BibTeX: 17
- EndNote: 17
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1