|
Yang wrote:
> I'm using meshless method to model localization problem and I want to
> discretize a 2D problem domain using a set of randomly distributed
> nodes, is there anyone who knows how to generate the random nodes using
> Matlab? I would appreciate your help very much!
Generating the random nodes without some kind of feedback from the fit is
unlikely to find a good minimum unless the problem is pretty flat.
I suppose you could take all of the generated nodes 4 at a time, and use them
to predict the minima of a quartic fit through those points (quartic being the
maximum order with exact solutions), and test the values at all of the
predicted points... (you might have to use fewer points or a lower order or to
predict x and y separately in order to make the orders work out.) Not quite
sure how else you could hope to minimize the function except by a lot of
trials without feedback... and if you are going to do that, then using a mesh
and gradient techniques would seem more effective...
Have you considered possibilities such as using a simplex approach in which
you add gaussian noise to the generated coordinates (or at least to each new
node) ?
|
Feed for this Thread