fmincon - infinetismal change in variable values & result far from target

Question

Karthik Vemireddy 2011-12-14

0
链接

此问题的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/23906-fmincon-infinetismal-change-in-variable-values-result-far-from-target

回答： Matt J 2017-10-12

Hi everyone,

I'm trying to identify suitable parameters for a system using the the sum of percent deviations of my actual objectives from their targets as the fitness function F for fmincon.

The solver stops the optimization prematurely as it exceeds the function evaluation limit. Acutally the values of the variables change infinetismally after 50 evals even though the fitness value is far from its target. Change in finite differences (i.e., Perturbations), maxfunevals, max iterations dint help. I am using SQP, even then the solver voilates the constraints for couple of iterations.

Change in other optimset parameters have changed niether the variable values (or change in variable values) nor the result much.

Can someone help me out with this?

Thnx

Rgds, Karthik

1 个评论
显示 -1更早的评论隐藏 -1更早的评论

Steve Grikschat 2011-12-14

Is there a reason you want percent deviations instead of squared deviations (i.e. least-squares)? Just curious, since least-squares has a ready-made interface in lsqnonlin or lsqcurvefit.

It could be any number of things that cause the solver to stall. Changing the finite-difference options (DiffMinChange, DiffMaxChange) can have an impact, positively or negatively, depending on the appropriate values.

Have you tried different start points (x0)?

请先登录，再进行评论。

请先登录，再回答此问题。