I am trying to optimize the parameters of a 2 by 2 controller via fminsearch but I am getting the result same as the initial points that are inserted to initiate the optimization process.

I already check the programs and found no errors on the programs.Can someone please give me some guidance?Is it possible that an optimization routine would not work on certain problem? The attachment above are the programs and the main program is "Taguchi_fminsearchp.m".

Answer 1

[Alan Weiss]

Your objective function is the result of a simulation. You might need to take larger steps in your optimmization routine to avoid the function looking flat at nearby points; see Optimizing a simulation or ODE.

Unfortunately, fminsearch does not allow for tuning options. I suppose that you could artificially scale the problem by taking fun(1e3*x) so that small changes in the position get magnified. Or you could use an Optimization Toolbox solver, using larger-than-default finite difference steps, or a Global Optimization Toolbox solver such as patternsearch.

Good luck,

Alan Weiss

MATLAB mathematical toolbox documentation