You might find some useful information in Darrell Whitley's "A Genetic Algorithm Tutorial"

http://www.google.com/url?sa=t&rct=j&q=&esrc=s&source=web&cd=2&ved=0CCUQFjAB&url=http%3A%2F%2Fibug.doc.ic.ac.uk%2Fmedia%2Fuploads%2Fdocuments%2Fcourses%2FGeneticAlgorithm-tutorial.pdf&ei=4v2qU-KXE8yryASS8YKQBA&usg=AFQjCNFzdY5pQPiRLoceikeDKLMuBXuQcw&bvm=bv.69620078,d.aWw

@Michael Wendl: Thank you, this literature tip was really helpful!

Well, I just can recommend you based on my experience, I had non-linear inverse problem with bunch of unknown parameters (around 100), i had tried a lot to solve this problem using GA, unfortunately i couldn't, as GA exactly with MATLAB function just looks for find the right solution mathematically not physically, but there is two differences, i had inverse problem with more unknown parameters, so my problem is worth, but i strongly recommend you to focus on object function, hopefully you can find the best results,

As another method, i used Levenberg Marquardt algorithm, it works terrific,

good luck

Ismael,

Dear Mr. Mahak,

GA is ,Metaheuristic algorithm. any it is base on random selection and natural progress.

I have solved many examples using Advance optimization algorithms, and I can suggest you that this will work on error and trial based selection of parameters. Because GA is black box type arrangement and it take input and give out put without knowing actual analysis of our problem.

Thank You

Dear Michael, Thank You very much for this tutorial. I will try to understand.

Thank you Ismael & Ghanshyam for your suggestion. These are really helpful.

Dear Ismael,

Did you use this (Levenberg Marquardt algorithm) in Matlab ?

As I saw, it is given in the options of Least Square Fit Method or you used it by another way.

Please suggest.

Thank You very much

Mahak:

Please find this paper

"A Memo on How to Use the Levenberg-Marquardt Algorithm for Refining Camera Calibration Parameters"

Although It is not related to earth science, it has very useful code that you can easily  implement that in MATLAB, it is approriat code for non-linear problems.

If you couldnt find this paper just email me, i will send that to you,

I've never used  defult functions of LM in MATLAB, usually i have been trying to write my codes to solve the specific equations, as i know better whats going on there.

Sorry for delay in response,

Cheers,

Ismael,

The most critical thing is the ranges and step intervals of the parameters that you are examining using a GA (which is NOT Marquardt-Levenberg - this is a linearized inversion method). This will determine to total time taken, so it is better to start with a small number of steps in each parameter spanning a reasonable range for each. For example, 35 parameters and 8 steps in each will imply 35^8 =2x10^12 evaluations of your function for each generation for a simple Monte-Carlo method (which GA is a refinement of). If there are some parameters you have a good estimate of, fix them initially or give them a small range. In my experience, the other GA parameters are not as critical, See Goldberg (1989) for more details.