the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
FastLE: A New Load Extrapolation Method for Site-specific Wind Turbines Using the Load Distribution Meta Model
Abstract. To ensure the safety of wind turbines at specific sites, IEC 61400-1 mandates the extrapolation of loads as a key requirement. Given the variability in wind parameters across different turbine sites, particularly in complex terrains, this task demands significant computational resources for simulations. However, the method recommended in the standard fall short of providing comprehensive assessments and rapid iterations necessary for all turbine locations within wind farm optimization designs. This paper presents a rapid load extrapolation method, named FastLE, which is based on a load distribution meta-model and tailored for specific sites. Based on 20 test cases, the blade root out-of-plane bending moment (OOPBM) for a 50-year return period was calculated using both the IEC method and the FastLE method introduced in this paper. Through comparative analysis, the mean APE is only 3.165 %, and the computation time for a single calculation has been reduced from 20 hours to less than 1 second. The results show that the FastLE method can complete load extrapolation calculations for wind turbines in seconds with high accuracy. This makes it suitable for ensuring structural integrity during iterations of wind farm layout optimization or turbine type optimization, thereby reducing the safety risks associated with wind turbines.
- Preprint
(5571 KB) - Metadata XML
- BibTeX
- EndNote
Status: final response (author comments only)
-
EC1: 'Comment on wes-2025-39', Nikolay Dimitrov, 24 Mar 2025
Dear authors, thank you for submitting this interesting paper. I have a few comments which I hope will be complementary to the upcoming reviews.
- The authors mention an impressive dataset of 541 meteorological towers in a specific region. Some more details would be relevant in order to understand if the data are comparable – i.e., how does the terrain differ among the various met mast locations, are the measurement heights the same, are the instruments the same (cup anemometers, sonic anemometers, lidars)?
- The IEC 61400, ed. 4 standard allows several different approaches to extrapolation, including avoiding the extrapolation altogether by introducing a higher safety factor. It will be useful if the authors could study/compare these different extrapolation approaches in the context of their proposed methodology.
- One significant challenge in the “fitting before aggregation” method is that the distribution fitting on a few values is not very robust, and a few outliers or bad fits can distort the aggregated result. It would be good to check the confidence in the aggregated distribution predictions – for example by doing multiple local distribution fits by bootstrapping the block maxima.
- There is a dependency between the shape and scale parameters in a Weibull distribution fit (if you choose a value of one parameter, it will define what is the value of the other parameter that best represents the data set). Therefore, fitting separate meta models for the scale and shape parameters of the Weibull distribution may limit the accuracy of the results. In the current manuscript, it doesn’t get clear if the authors fit one single MLP model with two outputs, or two separate models? Please discuss.
Citation: https://doi.org/10.5194/wes-2025-39-EC1 - AC1: 'Reply on EC1', Shanshan Guo, 27 Mar 2025
-
RC1: 'Comment on wes-2025-39', Anonymous Referee #1, 19 May 2025
The paper deals with a rapid evaluation of the extreme loads using extrapolation methods. currently the extrapolation requires a significant number of simulations to provide sufficient samples of extreme loads in order to perform the extrapolation procedure. The manuscript uses a machine learning based meta model to accelerate this process while providing extrapolation result with uncertainties comparable to those using aeroelastic simulations. One important question that needs to be clarified is what is the value of the meta model compared to the many load surrogate models that are available in the literature. Afterall the main time saving is coming from the meta load model which is essentially another load surrogate model.
1- when comparing the time saving, how would the authors account the time and efforts needed to produce the data needed to train the meta model. since this would be necessary each time the turbine model or turbine properties have been changes, which is often the case in the design iteration phase. Normally the 50 years return value for extreme load is a design value based on generic wind class or site specific value for certain class of sites, for example typhoon or hurricane affected area. It is usually not needed to perform load extrapolation for each of the wind turbine in a wind farm. Once it has been identified which turbine in the wind farm has the highest extreme loads, one needs only to perform the load extrapolation for the worst case. It is rather unlikely that optimization for extreme loads will be performed for every single turbine. Moreover, it is not clear from the beginning of the design, whether fatigue or extreme load will be the design driver. Therefore, the usefulness and time saving should be considered with these points in mind.
2- In page three, line 84, the word inflow angle is mentioned. In this case, it is referred to the yaw angle between the rotor plane with the incoming wind, that is, the yaw misalignment angle, if the reviewer understands it correctly. Inflow angle is used in the aerodynamics mainly for the angle of the velocity triangle at the airfoil, between the tangential velocity caused by the rotation of the rotor and the incoming wind velocity. The use of the word inflow angle can cause some confusion as this is not used normally in this context.
3- Page 4. which is the shear model used and how is the shear value defined, please elaborate.
4- Figure1, the distribution of the air density looks bi-modal, when sampling the distribution, did the authors take the empirical distribution or the fitted bi-modal distribution
5- Table 1 why is the inflow angle changes from -0.78 to 13.464 degrees (there is no need to go beyond the first digit for this angle, the turbine yaw controller is not that precise) , what about the variation in the negative angle. the loading on the wind turbine is not symmetrical around the yaw angle, negative and positive yaw angles can produce very different loads.
6- Table 2 change RMP to RPM
7- page 7 what is the definition of In plane and out of plane bending moment here. It looks like the authors is using the flapwise bending moment and not the out of plane bending moment of the blade. Once the blade starts pitching after reaching the rated wind speed, the the OOP bending moment and flapwise bending moment are no longer the same.
8- Equation 6, this equation assumes that the 10 minutes wind speeds are independent, which is clearly not the case.
9- page 9, line 171, the authors divided the data into three categories, high wind speed range above 10 m/s , low wind speed range below 10 m/s and full wind speed range, which wind speed would be full wind speed range have?
10- Figure 6 why are the log-normal performed so poorly in QQ plot
11- Table 3, there is not need to have numbers with 9 digits after the decimal point, there are a lot of uncertainties
11- Figure 10, how ar ehte importance of the hyperparameters determined?
12- page 9 line 176, so if the low wind speeds contribute so little to the tail of the distribution, then why simulate them at all.
13- instead of local distribution, maybe it is better to refer them as local maxima, or local peaks distribution.
14- Table 5, the simulation time is 600seconds, what about the transient at the beginning of the simulation, are they removed ?
Citation: https://doi.org/10.5194/wes-2025-39-RC1 - AC2: 'Reply on RC1', Shanshan Guo, 28 May 2025
-
RC2: 'Comment on wes-2025-39', Anonymous Referee #2, 13 Aug 2025
The manuscript presents FastLE, a meta-model–based method for rapid extrapolation of extreme wind turbine loads, aiming to reduce the computational burden of IEC 61400-1 site-specific load analysis. The work addresses a relevant and underexplored problem, and the results indicate high agreement with the IEC reference method for simulated cases.
However, the following issues should be addressed before this work can be considered for publication:
Scientific comments
- The authors mention they use “site-specific” wind parameters, but the wind speed is obtained from the turbine specification, and air density and turbulence intensity are from the literature. In this case, the statistical characteristics of the selected probability distribution fortunately match the measurement. But what if you choose another site? Why not to use the measured statistics directly, e.g. sampling from probabilities in discrete intervals, instead of fitting to a distribution and resampling?
- Figure 5b is not explained. “P-value” is defined neither in the text nor the caption. Does it represent the polulation parameter? Why increasing the block size gives larger P-values if population parameter=0 for independence?
- While the Weibull distribution emerges as the most common optimal fit (52%), the fact that Normal (34%) and Gumbel (14%) distributions perform better for a substantial proportion of wind speeds suggests that a wind-speed-dependent or mixed fitting approach could improve accuracy. Including a sensitivity study of distribution choice on extrapolated loads would be helpful.
- Please clarify how the hyperparameter importance is obtained.
- Did you consider normalization of MLP input and output parameters? Due to the difference in the order of magnitude between the two outputs, the MLP possibly gives more importance to the paras_1(scale) over paras_0 (shape).
- A figure of the proposed framework (similar to Figure 4) would be helpful to understand the added contribution of this paper compared to the IEC-proposed method.
- Page 14, Line 244, the error is in percentage or true value? Formulate what it meant by “error”?
- Page 18, Line 276, the time required to generate the training data for FastLE should be taken into account.
Technical comments
- Reference needed for “It is widely recognized that lower wind speeds contribute minimally to the tails of long-term load distributions.”
- Add the references “To effectively speed up the load extrapolation, this study references certain literature to introduce wind parameters into the Meta model for load components.”
- Some abbreviations are used without defining.
- The manuscript still needs to be carefully proofread.
- Figures could be made more self-contained by including parameter definitions and clearer legends.
Citation: https://doi.org/10.5194/wes-2025-39-RC2 - AC3: 'Reply on RC2', Shanshan Guo, 20 Aug 2025
Viewed
HTML | XML | Total | BibTeX | EndNote | |
---|---|---|---|---|---|
250 | 83 | 21 | 354 | 16 | 41 |
- HTML: 250
- PDF: 83
- XML: 21
- Total: 354
- BibTeX: 16
- EndNote: 41
Viewed (geographical distribution)
Country | # | Views | % |
---|
Total: | 0 |
HTML: | 0 |
PDF: | 0 |
XML: | 0 |
- 1