want to do optimization with fmincon but without having exact function

Hi,

I want to do an optimization for constrain problem and use fmincon.

The problem is that, I don't have function that be defined by variables like f(x,y)=x^2+y^2

I have FEM model which is contain a file with thousands of line in bdf format, which is as input for MSC.NASTRAN

I cloud read my design variable from that file and store them in parameters, as well as write new value on the specific place. Then I will call MSC.Nastran to do computation and then generate f06 file, which the value of objective and constrains are stored in that and read these value and store them in parameters too.

I want to use fmincon to do optimization for my problem but it gives me error which i think this is due to calculation of gradient of function, because for 10 variables problem it do 11 calculation and then error appear.

Thanks so much in advanced.

[mohammad haghnejat]

This is the error that appear:

Error using textscan
Invalid file identifier. Use fopen to generate a valid file identifier.

Error in diflecti (line 4)
C2 = textscan(fidInBDF,'%*s %*s %*s %s %*[^\n]'); % Scan second column of the bdf file

Error in barrier

Error in fmincon (line 796)
    [X,FVAL,EXITFLAG,OUTPUT,LAMBDA,GRAD,HESSIAN] = barrier(funfcn,X,A,B,Aeq,Beq,l,u,confcn,options.HessFcn, ...

Error in main (line 25)
[x,y]=fmincon('weigh',x0,[],[],[],[],[],[],'diflecti')

[Dorsa Haji Ghaffari]

Hi Mohammad,

I'm trying to optimize a system without having a formulated function that can define my system. I'm using fmincon, but cannot get it to work without providing a function. I get my objective function values real time from experiments and I want to minimize the objective function value.

I see that you were doing a similar thing in the past. Is there any documentation on working with fmincon when there's no formulated function?

Thank you so much!

[Walter Roberson]

Dorsa Haji Ghaffari:

What you have defined so far would be satisfied with an objective function that returned -inf no matter what the input was.

Given a set of times and any associated measurements so far and a list of trial model parameters, then how would you compute the current objective function value?

As time progresses, are you hoping to refine the parameters (the uncertaintly gets lower as you have more data), or are you hoping to update the parameters (the parameters are changing in time) ?

[mohammad haghnejat]

Hi Dorsa,

I did.

To do optimization, any optimization algorithm needs value for a cost function. It can be obtained from a mathematical equation in a code, or from other techniques.

If you want to use other simulation tools, you need to do these steps:

This is a procedure for func.m

For instance, the function z = Func(X)

X is input from fmincon

1-Your code needs to be able to build a file that assigns X in a way that software can use it.

To assign X to the input, fopen, fseek, and fwrite maybe useful

2-Run the software with the system command,

3-When the calculation is done, an output that can be read in the matlab must be produced.

fopen and textscan maybe useful in this section

I hope you get the idea.

If I can be of assistance, please do not hesitate to contact me

Answer 1

[Walter Roberson]

Your function diflecti failed when it tried to fopen() a file for reading. You should be using something similar to

BDF_filename = 'BDF_FILE_NAME_GOES_HERE';
[fidInBDF, msg] = fopen(BDF_filename);
if fidInBDF < 0
  error('Failed to open BDF file "%s" because "%s"', BDF_filename, msg);
end

Answer 2

[mohammad haghnejat]

thanks so much, I think I fix my problem.

Thank so much,