Different levels of fidelity of Floris

[FMenhorn]

Hi,

I am very new to Floris and want to thank you, first of all, for the great documentation of your code. It was quite easy to set up the code and get some first examples running.

I am working in the field of Uncertainty Quantification, specifically for Multilevel/Multifidelity methods and would like to test Floris (or Wind Energy in general) as an application. Is it possible in Floris to create different levels of fidelity with different computational cost? In other words, are there different levels of accuracy of the solution with are easily steerable, e.g., with the input file, which result in different computation times--- usually higher accuracy equals higher computational cost? In the area of multifidelity this is often achieved by changing the discretization of the grid or using different solvers with different accuracy. I played a bit around with different number of grid points/turbines/solver but could not find an obvious choice.

Thanks a lot for your help!

[paulf81]

Hi, and thank you very much for the kind words about FLORIS. I'm very glad to hear you had an easy time getting started!

So to your question, it's a nice one and probably off the top of my head I can only give an incomplete answer. I think your first idea is also mine, changing the number of grid points per turbine. Some of the models allow you to go all the down to 1 point per turbine, which does speed things up but should involve some variation in output.

Another place where there can be a clear trade in time/accuracy can be how many wind speeds and wind direction are included in an AEP calculation. Start with say 0 - 25 m/s in 1 m/s steps and 0 - 360 in 1 deg steps, you can save a lot of time by going to larger steps and eventually things will start to vary. On this topic you can find some literature looking at a good compromise between speed and precision independent of FLORIS specific.

I wanted to say your question reminds me of this paper:
https://iopscience.iop.org/article/10.1088/1742-6596/1037/4/042012/meta

Which might also give you some ideas of model parameter variations that could be considered

Tagging @jaredthomas68 in case has some other thoughts

[FMenhorn]

Sorry for the belated reply due to Easter break.

Thanks Paul! This indeed increased the computation time. I just swapped it around and sweep over all wind directions for a single wind speed but the increase in computation time is similar. (I do the change because I want to apply uncertainty on the wind speed in the study)

However, unfortunately, I don't get a big difference with respect to the power production when using the difference grids, turbine_grid_points=1 or turbine_grid_points=3. I guess the single grid point is already good enough in the approximation quality and the improvement turbine_grid_points=3 provides is very small. Hence, I am unsure if I can use it in my study. Thank you very much for your help though!

[Bartdoekemeijer]

Hi @FMenhorn. Check out #499 for a convergence (or lack thereof) study of the number of turbine grid points.

[paulf81]

Thanks @Bartdoekemeijer for a helpful link!

[FMenhorn]

Yes, thanks @Bartdoekemeijer and @paulf81 ! This indeed helped me. (I had to modify the code a bit to not raise a ValueError for grid points > 3) Together with the other suggestions I was able to create three levels of fidelity:

• coarse: 1 turbine grid points + calculate_no_wake for the computation of the turbine power
• medium: 1 turbine grid points + calculate_wake for the computation of the turbine power
• fine: 40 turbine grid points + calculate_wake for the computation of the turbine power

I get a nice increase in runtime and convergence of the solution.

Thanks all for your help!

[FMenhorn]

Now, I have an additional question. I want to control the yaw angles and then visualize them. I can set the yaw angles as parameter for calculate_no_wake/calculate_wake. For the visualization using calculate_horizontal_plane, I have to set the same yaw angles as parameter in that function call as well, correct? It doesn't automatically know the yaw angle, I believe. Is there also a way to create a gif like you have on your page on the top left of https://nrel.github.io/floris/intro.html ?

Thanks again!

[Bartdoekemeijer]

@FMenhorn could you open a separate discussion for this question? I think more people will find it useful and it'll be easier to find that way. Thanks!

[rafmudaf]

@FMenhorn I've found the script I used to create that gif. Please point me to the new discussion and I'll post it there. Thanks!