Bug report: calling reinitialize_flow_field leads to incorrect turbine model information in FLORIS 2.4

[pjireland]

Bug report: calling reinitialize_flow_field leads to incorrect turbine model information in FLORIS 2.4

When calling FlorisInterface.reinitialize_flow_field with layout_array not None, turbine-model-specific settings (e.g., hub height, rotor diameter, power-thrust curves) are overwritten with settings from the first turbine. This leads to problems if I want to simulate turbines of different models on the same farm when changing the farm layout.

Note that I haven't explored FLORIS 3.x to see if this issue persists, but given the significant restructuring in FLORIS 3.x, I would assume it does not. However, I am focused on using FLORIS 2.4 in my work, since FLORIS 3.x does not have support for wind direction heterogeneity, which is an important feature I need for simulations of large farms (see #441).

How to reproduce

First set up an example wind farm (based on the AEP example in FLORIS 2.4):

from floris.tools.floris_interface import FlorisInterface

# Use `example_input.json` file from FLORIS 2.4
# Taken from here:
# https://github.com/NREL/floris/blob/ef4934ec7feb7afd2615772d364a1eaa28db93e9/examples/example_input.json
fi = FlorisInterface("example_input.json")

# Set wind farm to N_row x N_row grid with constant spacing
# (2 x 2 grid, 5 D spacing)
D = fi.floris.farm.turbines[0].rotor_diameter
N_row = 2
spc = 5
layout_x = []
layout_y = []
for i in range(N_row):
    for k in range(N_row):
        layout_x.append(i * spc * D)
        layout_y.append(k * spc * D)
N_turb = len(layout_x)

fi.reinitialize_flow_field(
    layout_array=(layout_x, layout_y), wind_direction=[270.0], wind_speed=[8.0]
)
fi.calculate_wake()

Then modify the rotor diameter of the last turbine.

print(fi.floris.farm.turbines[-1].rotor_diameter) # 126.0
fi.floris.farm.turbines[-1].rotor_diameter = 100
print(fi.floris.farm.turbines[-1].rotor_diameter) # 100

Finally, reinitialize the flow field, changing just the layout and notice the the rotor diameter of the last turbine is now incorrectly equal to that of the first turbine:

fi.reinitialize_flow_field(
    layout_array=(
        [0, 0, 320, 320],
        [0, 320, 0, 320]
    )
)
print(fi.floris.farm.turbines[-1].rotor_diameter) # 126.0

Floris version

2.4

System Information

• OS: Ubuntu 20.04.1 LTS (GNU/Linux 5.10.16.3-microsoft-standard-WSL2 x86_64)
• Python version: Python 3.9.15
• Library versions
  • matplotlib: 3.6.2
  • numpy: 1.23.5
  • numexpr: 2.8.3
  • scipy: 1.8.1
  • pandas: 1.5.2

[pjireland]

The root of the issue seems to be these lines in floris.tools.floris_interface.FlorisInterface.reinitialize_flow_field:

                turbine_map = TurbineMap(
                    layout_array[0],
                    layout_array[1],
                    [
                        copy.deepcopy(self.floris.farm.turbines[0])
                        for ii in range(len(layout_array[0]))
                    ],
                )

where we see that the first turbine's information is used to overwrite those of the other turbines.

[rafmudaf]

Thanks for documenting this nicely @pjireland. Unfortunately, I don't recall why we used the first turbine for all in the reinitialize routine in v2. Have you checked whether it's valid to do this:

# change this
copy.deepcopy(self.floris.farm.turbines[0])

# to this
copy.deepcopy(self.floris.farm.turbines[ii])

If the indexes in the layout array do not correspond to the correct indexes in the floris.farm.turbines array, then this won't work.

[pjireland]

@rafmudaf Thanks for the feedback. It looks like that works if you already have 4 turbines in self.floris.farm.turbines. But it will cause problems in the example I provided (and I would assume many of the examples in the codebase, which are similar) because there are not yet 4 turbines when reinitialize_flow_field is called.

Perhaps something like this could work?

assert len(self.floris.farm.turbines) in [1, len(layout_array[0])]
...
                turbine_map = TurbineMap(
                    layout_array[0],
                    layout_array[1],
                    [
                        copy.deepcopy(self.floris.farm.turbines[0])
                        if len(self.floris.farm.turbines) == 1
                        else copy.deepcopy(self.floris.farm.turbines[ii])
                        for ii in range(len(layout_array[0])) 
                    ],
                )

[rafmudaf]

Without having tested it, that looks right to me.

Regarding incorporating a fix, we haven't considered how to maintain FLORIS v2 in the case of users being tied to functionality that isn't available in v3. However, this and #684 seem like straightforward changes and easy enough to include as patches. I'll discuss it with the NREL team and let you know what we decide.

[rafmudaf]

Hey @pjireland if you're interested to submit a pull request with this change and any corresponding tests showing that it fixes the bug, I'm happy to review, merge, and complete the release process for a bug fix.

After some discussion, we've described the policy regarding v2 updates in #715, for reference.

[pjireland]

Hey @pjireland if you're interested to submit a pull request with this change and any corresponding tests showing that it fixes the bug, I'm happy to review, merge, and complete the release process for a bug fix.

After some discussion, we've described the policy regarding v2 updates in #715, for reference.

Thanks for the feedback, @rafmudaf. For now, we have a workaround that will allow us to continue v2.4, but I'll submit a pull request if this capability becomes necessary.