Parameter Determination Using Nelder Mead Method

Hello everyone, I am working on my project now and I am developing a model for my work. I need to optimize the model by determining its parameters. I am required to use the Nelder Mead Algorithm in MATLAB to do this. Please I need help. I am totally clueless in the use of MATLAB or the said Algorithm. Anyone to help can please contact me and I will follow up from there. Thank you very much.

Answer 1

[Star Strider]

The fminsearch function implements that (or so it claims in algorithm in the Description section of the documentation for it).

[Victor Emah]

Consider me a layman. I have my sets of data and I have my model equation with three to four unknowns. I want to find these unknowns.

[Star Strider]

The mention of fminsearch in my first Answer post is a hyperlink to the online documentation for it. That explains how to do the optimisation, and has a link to another algorithm, fminbnd, at the end of the page, if you need its functionality. If you have the Optimization Toolbox, you have a number of other options.

Search the documentation for ‘Nelder Mead’ to look for other functions that may use it. That’s what I did. I stopped after finding fminsearch because it’s a core MATLAB function everyone has access to.

If you’re doing curve-fitting, fminsearch requires the extra step of creating a sum-of-squares cost function to use as the objective function for fminsearch. See the documentation for Curve Fitting via Optimization for details.

[Zhengru Yang]

Dear Star Strider,

Thanks for your explanation. I'm facing a similar question as Victor. I have a model that can generate a curve with 3-4 tunable variables. Now, I need to use this model to history match an objective curve. Does it make sense to use fminsearch or say Nelder Mead/Downhill Simplex Algorithm to help me optimize the model to get a match? I know people typically use the least-squares function to minimize deviations during history matching. But I'm still confused with the logic between the history matching and the optimization of variables using Downhill Simplex Algorithm,

Could you please kindly give me some hint regarding this? Thanks a lot!

[Star Strider]

Zhengru Yang — The fminsearch function uses the derivative-free Nelder-Mead algorithm, and for that reason is less likely to get stuck in a local minimum that gradient-descent algorithms.. It might be able to do what you want, however a better choice is likely the Global Optimization Toolbox ga (genetic algorithm) function, largely because fminsearch is limited in the number of parameters it can successfully optimise. It may require the ga function several runs with different random initial populations to find the best parameter set, however it is more likely to be successful than other approaches.

[Zhengru Yang]

Thank you!

[Star Strider]

My pleasure!