Help for details on the use of the YawOptimizationSR class (Serial-Refine).

[CaiZhiming-HHU]

Hi there,
I would like to ask for some help with some of the internal details of the SR algorithm usage process.

I know that using SR algorithm to calculate the optimal yaw angle combination of wind farm involves the use of YawOptimizationSR and YawOptimization class, but I have some questions when the wind farm is optimized for multiple times.

For example, now I set Ny_passes=[3, 4], which means that the yaw angles of each turbine in the wind farm will be optimized twice, first for the Serial pass and then for the Refine pass. Secondly, I set the upper and lower boundary of yaw optimization for each turbine to 15° and -15° respectively. In the Serial pass, FLORIS will traverse the three angles of -15°, 0° and 15° for each wind turbine in the wind farm in the order of upstream and downstream, and record the whole field yaw angle combination when the maximum power appears.

But my confusion arises in the Refine pass, where I don't know how to get the yaw angle subset of each wind turbine in the Refine pass. I looked up the python code of YawOptimizationSR and YawOptimization class, there are 17 functions in total. I admit that my python programming skills are not great, so I have trouble reading this code.

I also read the article "Serial-Refine Method for Fast Wake Steering Yaw Optimization", which gave me a lot of inspiration, as shown in the figure below, when Refine pass, The actual yaw angles traversed by each wind turbine is a set of more refined yaw angles superimposed on top of the optimal yaw angle found in the Serial pass, but I am curious about how this set of "more refined yaw angles" relates to the parameters in the Ny_passes array.

Please use the parameters I have set: Ny_passes=[3, 4], the yaw boundary of each turbine is [-15°, 15°], I would be grateful if you could explain in detail how the "more refined yaw angles" are obtained for the Refine pass of the turbine!

@kflemin @andrewning

[Bartdoekemeijer]

Hi,

We have not clearly separated a single "Refine" step in the code. Actually, you can have as many refine steps as you'd like! For example, you can do Ny_passes = [3, 4] but you could also do Ny_passes = [3, 4, 4]. Then the explored spaces are:

1. First step, it would explore [-15.0, 0.0, +15.0]. Let's say 0 gives the highest power. Then,
2. Second step it would explore [-7.5, -3.75, 0.0, 3.75, 7.5]. If the optimal is 3.75, Then
3. Third step would explore [1.875, 2.8125, 3.75, 4.6875, 5.625].

Don't pin me down on the exact values, but you get the gist.

In the FLORIS code, the refinement explored space is generated for a single turbine and for a single findex here:

floris/floris/optimization/yaw_optimization/yaw_optimizer_sr.py

Line 211 in fd0e599

yaw_angles_subset = np.linspace(yaw_lb, yaw_ub, Ny + 1)[ids]

And collected in a single variable evaluation_grid for a single turbine and all findices here:

floris/floris/optimization/yaw_optimization/yaw_optimizer_sr.py

Line 213 in fd0e599

evaluation_grid[:, iw, turbid] = yaw_angles_subset

Put a breakpoint in your code near those lines and try to inspect what values are evaluated. Try it first for a single findex (so a single wind direction/speed/turbulence) to really help that understanding. Hopefully that will clarify the code better for you.

Hope this helps!
Bart

[CaiZhiming-HHU]

Thank you for your prompt reply, which helps me a lot!