Setting non-rectangular boundaries for layout optimization

[Rodrig03]

Hello,

Im running layout optimization tests using LayoutOptimizationScipy from floris.tools.optimization.layout_optimization.layout_optimization_scipy

I am experiencing issues setting irregular wind farm boundaries. It appears that a the most outer "x" and "y" coordinates are taken to create a new rectangle boundary, which is then used for the optimization.
For example, when setting the following boundaries = [(0, 0), (0, 1600), (1000, 1600), (1600, 1200), (1600, 0), (0, 0)] I get the results shown in the image.

Is there any way to set irregular boundaries based on "x", "y" coordinates for layout optimization?

[cherifM]

Hi @Rodrig03

yes you are right, switching to scipy-optimizer is faster but its not working properly, its checking boundary-geometry bounding box, not each boundary lines, here the boundary is basically polygone.
I switch back to PyOptSparse interface, and everything works, i use slsqp , and results are perfect. This is my solution which I recommend, I you have to use scipy Optimizer, I did a lot of tests, my conclusions is you have to have really good initial layout, than scipy Optimizer works (bit faster than PyOptSparse), but again, PyOptSparse is in my opinion is more robust until the dev team fix this bug.
hope this help ,

cheers,

cherif

[bayc]

@Rodrig03 Thank you for bringing this to our attention. I will look into this shortly and get back with more info.

[cherifM]

[cherifM]

I did update stock_examples/18_optimize_layout.py to use PyOptSparse
i.e. use LayoutOptimizationPyOptSparse interface instead of scipy :

from floris.tools.optimization.layout_optimization.layout_optimization_pyoptsparse import LayoutOptimizationPyOptSparse

layout_opt = LayoutOptimizationPyOptSparse(fi, boundaries ,  freq=freq ,solver = "SLSQP")
sol = layout_opt.optimize()
print(sol )
layout_opt.plot_layout_opt_results()

with some changes tin the file :

.../floris/tools/optimization/layout_optimization/layout_optimization_pyoptsparse.py

[Rodrig03]

Hello cherifM,

Thank you a lot for looking into this,

The PyOptSparse solution fits the optimized turbine positions into the irregular boundaries. However it seems that you have to have a very good initial layout to use this solution since the turbines are slightly readjusted and not completely repositioned.

Will the option of performing layout optimizations utilizing Scipy be available in a future release?

Also, where can I find the code you updated (stock_examples/18_optimize_layout.py)?
I'm not able to find it in the Git repository.

[cherifM]

Hi @Rodrig03 ,
these are my answers:

1. with pyoptspare you don't have to have good initial condition, this is issue happend only with scipy.
2. I don't know when floris development teams with fix this issue with scipy, im n ot part of this team.
3. the file "stock_examples/18_optimize_layout.py" is the same file as in "example/18_optimize_layout.py", i did change it bit like i showed in my previous message, beside this file "floris/tools/optimization/layout_optimization/layout_optimization_pyoptsparse.py". As I explained in my previous message :
   (line 93 if i m not wrong) needs to be changed like this

        # Update turbine map with turbince locations
        #self.fi.reinitialize(layout_x = self.x, layout_y = self.y)
        self.fi.reinitialize(layout=(self.x, self.y))

and of course you need to install pyoptspare, please let me know if you need further help ,

best regards,
cherif