PTGen: allow more optimization levels

4 posts / 0 new
Last post
solarjoe
PTGen: allow more optimization levels

Hello,

I have a feature request and hope that you can help me.

We came across the problem that SAM's optimization
wizard PTgen returns rather non-continuous values
if the input parameters are changed.

This is due to the Optimization Levels for
Tower Height
Receiver Height
Receiver H/D ratio

Since the discretization is in three dimensions,
the optimizer is likely to find a local minima
if the optimization levels are to coarse.

From what I read in the DELSOL3_Optimize.txt
after adding
save_files = 1.
to the SAM_Tower_DELSOL.inc, the optimization is
just straight forward crawling through all
combinations.
So, I guess there are just three nested for loops
running across the ranges.

Well, here is my feature request:
The current maximum for the optimization levels
is 15, which Ì consider as coarse.
Would it be possible to allow arbitrary steps?

We are investigating effects in detail, e.g.
the impact of a heliostat's image mirror
on the LCOE. There we saw that the local minima of
the optimizer lead to non-consistent results.

So we would be very interessted in using a finer
descretization, even if this means the optimizer
will run for an hour.

Kind regards,
Joe

solarjoe

Hello,

I thought about a SamUL workaround, but this is rather
inconvenient way to do it:

One could define a finer discretization and then break it
into computable pieces of 10 ( or the current "max. optimization levels")
and do the fine optimization piecewise by storing
the results and picking the best.

Well, this is not impossible to do, but I would really prefer
to find another solution.

If the source code of PTgen would be available, I could do the
changes myself and provide a patch.
Or also just modify my personal version of the PTgen.exe .

Kind regards,
Joe

mwagner

Hi Joe,

The maximum number of optimization levels is set by DELSOL3 which is contained within the PTGen wrapper. Unfortunately, DELSOL3 is difficult to modify without introducing unintended errors, so we have left the optimization algorithms intact. Consequently, we don't distribute the source code for DELSOL3 or PTGen since it is very easy to make a change that invalidates the model outputs. As a side note, we have been developing a replacement tool for DELSOL3 that will offer significantly more flexibility in optimization, but it is not yet available.

You've described the best way to approach refining optimization, and this is the method that we recommend as best practice in any case.

Best,

Mike Wagner

solarjoe

Hello Mike,

thanks for your response.

Ok,then its understandable that it is difficult to do.
I thought PTgen ist just using DelSol3 for lot of different
standard runs to find the optimum.

Do you have any SamUL snippets or functions that implement a finer
discretization?

And can you already provide some information about the new
tool that will replace DelSol3 in SAM (papers, presentations...) ?

Kind regards,
Joe

Theme by Danetsoft and Danang Probo Sayekti inspired by Maksimer