Re: Optimal Power Flow with fmincon

I don't know anything about OPF, but I have used fmincon a fair amount. Depending on what my objective function looked like, I sometimes had to change some of the default optimization parameters. In particular, some of my objective functions were a little noisy, so I had to increase DiffMinChange so that I could get a well-behaved (pseudo-)gradient.

The other thought that I have is that if the different variables of the objective function are scaled very differently you might have problems. The TypicalX parameter *might* help with that, but I'm not sure.

--Arthur

On 2008-02-19 12:08:23 -0500, "Guoliang Zhang" <zhangl@xxxxxxxxxxxxx> said:
Sorry i'm not an expert on this function but i would suggest compare your
simulation result againts from other enviornments to see if it was because
of the misconfiguration or is it an algorithm problem.

You should be able to get an evaluation copy of the power system toolbox,
which is in Simulink format. And there are other tools with free evaluation
like www.powerworld.com (origined from UIUC).

go illini

GZ


"Tom " <tdd214dontspam@xxxxxxxxxxxxxxxxxx> wrote in message
news:fpetik$4qk$1@xxxxxxxxxxxxxxxxxxxxx
I have a very general question, for those of you who might
have experience with similar projects.

I'm running an AC optimal power flow (OPF) with fmincon.
It's a simple 3-bus system. I'm getting some strange
results, even after playing with convergence criteria for a
very long time. There are generators that I can tell
immediately are inefficiently dispatched in many of my
runs. The generators are at the same bus, and have
identical increasing cost functions, but are dispatched
differently! You can prove pretty easily that the only
efficient dispatch for such generators is an identical
dispatch, so these results must be inefficient. If I run a
bunch of different optimizations with different initial
guesses, it seems that I can usually get some of them to
converge to the correct dispatch.

The gist is that I know that, in most of my runs, the OPF
does not converge on the correct answer, and in fact
converges on an answer that would appear to be easy for an
optimization program to improve on. Shouldn't there be a
gradient somewhere in fmincon that would tell it that it
could easily improve the objective function by increasing
one generator and decreasing the other?

Any general advice? (I'm not trying to have you
troubleshoot my specific model... Just want to hear from
anyone who has modeled similar systems, and perhaps figured
out how to make them run well.) Have you had luck with
fmincon for similar models before? Does another
optimization routine work better? It seems like my model
is pretty standard and should be pretty easy to solve.

I was going to start writing a routine to solve a given
dispatch by running a number of optimizations for different
randomized initial guesses, and picking the best
covergence. I'm not an optimization expert, but seemed
like there might be some optimization routines that take
that kind of approach without me having to re-create the
wheel. Is it pretty standard to have to investigate a
bunch of initial guesses in order to solve a simple 3-bus
model, or is this an indication that my optimization
routine isn't well suited to my model? I'd love any input
along these lines.

I am aware there is a power toolbox for Matlab, but I need
a lot of flexibility in what I am modeling, so I've been
thinking that a free-standing model would be better for me
(and I'm a poor researcher not looking to spend money I
don't have, unless I have to). But I'm also interested in
any comments on how flexible the power package is, or if
anyone knows what routines it uses to run optimizations (if
it even does OPFs). Does it use fmincon? Does anyone have
a reason to believe that it would be significantly better
for my application?

Thanks,
Tom


.

Relevant Pages

  • Using fmincon to optimise vectors?
    ... I have an optimization problem which contains two unknown variable ... have taken sum(objective function) to be my new objective function f. ... Is there a way to use fmincon correctly to solve my problem or do I ...
    (comp.soft-sys.matlab)
  • Re: Fmincon number of obj-function evaluations
    ... seemingly caused fmincon to still compute the gradient by finite ... explained in the optimization toolkit's handbook (in fact, ... even mentioned in the section about fmincon). ... associated with the objective function evaluation. ...
    (comp.soft-sys.matlab)
  • Multivariable Optimization for Robot arm responce
    ... Here comes optimization. ... Question about fmincon from optimisation toolbox ... % Compute the objective function value at x ...
    (comp.soft-sys.matlab)
  • Re: singular matrix?
    ... I get an error when performing an optimization with fmincon. ... Matrix is close to singular or badly scaled. ... on the objective function. ...
    (comp.soft-sys.matlab)
  • Re: Scaling in fmincon()
    ... Yes, yes, I did change the tolerance values accordingly. ... optimization variables are of similar magnitude, ... imply that Matlab fmincon() does not do these. ... I'll form two objectives, ...
    (comp.soft-sys.matlab)