1

I'm wondering whether a DoE approach could somehow be used as kind of an optimization algorithm? One of my current tasks is to find a set of five parameter which max a sixth one (see here for more: https://scicomp.stackexchange.com/questions/41237/which-optimization-algorithm-to-max-a-single-parameter-by-searching-a-landscape). A greedy approach would simply be to vary all parameters to fully scout the entire parameter landscape. But this is quite inefficient. And this reminded me actually of DoE, where basically the same aspect is addressed: Avoid observing all parameters against all others. Instead, select some useful parameter combinations to avoid a thorough and costly investigation.

img

Does it make sense, to find a maximum in a control engineering setting, to apply DoE to check parameter settings?

Ben
  • 3,443
  • 2
    It seems to make sense to me. I'd suggest looking into response surface designs, like the 2-factor Central Composite Design that you have illustrations of. Be aware, though, that the typical analysis of a CCD is to use 2nd order polynomials to model curvature, and if you have multiple local optima instead of a single global optimum, that might be problematic. Also, if runs are too expensive for a full 6-factor CCD, you can also consider optimal designs, such as an I-Optimal design. – MichiganWater Apr 11 '22 at 07:19
  • 1
    Bayesian optimization is worth a look – John Madden Nov 19 '22 at 22:05

1 Answers1

1

You seem to be looking for response surface designs. Look through some of the posts on this site.

A few other relevant posts:

kjetil b halvorsen
  • 77,844