How to map various properties using genetic algorithm?

Guys, I have a list of various steel section sizes and their corresponding properties (such as area,moment of inertia,radius etc) in an excel file. I am currently working on Genetic Algorithm. My GA is supposed to give me a section out of these section sizes . I want to input the section size as design variable. So when GA picks any section size it should also pick its corresponding properties.These properties will be used in defining constraints. As I do not know which section size it is going to pick, how do I input the file into GA and how to map the corresponding properties? Basically how should my design variable be inputted? Any suggestions and help is highly appreciated.

[Star Strider]

I have no idea what you’re talking about. What is a ‘section size’, and how does it relate mathematically to the other properties?

[Ace_ventura]

Hi Star Strider. I am designing an optimum steel building. 'Section size' means the size I will be using for my columns and beams ( I section with various flange and web dimensions). So , I have a list of various section sizes like ISMB 225,ISJB 350 etc (Indian Standard sections).Each section size has different cross sectional area (A), different moment of inertia (I) and so on and so forth.

I want to input my section sizes as design variable in genetic algorithm. Such that whenever a particular section size is selected, the other properties such as area and 'I' could also be accessed for that particular section. Since, GA chooses randomly,I would not be knowing which section size it has selected. I need these values to write the constraint equations for my algorithm. I hope ,I am more clear now. Do you have any suggestions?

[Ace_ventura]

following excel file might give you some idea

Answer 1

[Star Strider]

I’m still not certain that I understand ‘section size’.

The genetic algorithm (‘GA’) is an optimisation routine, similar in use to least-squares curve-fitting. It is likely not appropriate for your problem.

A technique that would be appropriate is a k-th nearest neighbour (‘knn’) classifier. Those are easy to write, and reasonably efficient in MATLAB. You could also use the pdist2 function. The idea is that you would enter whatever criteria you wanted, and the knn (or pdist2) routine would then find the closest match, based on whatever distance metric you selected. (The Euclidean distance is the most frequently used.)

I would likely use one of those rather than a GA if I wanted to find the closest members of a set to the criteria I selected. Once you found the indices of the closest matches, you could then easily find the other parameters.

----------

Interesting avatar image! Skyrim?

[Ace_ventura]

Thanks .. But I have to apply GA only. Yeah Skyrim..Hic sunt dracones!:) Did u see the excel file? That might give you the idea of section size.

[Ace_ventura]

Star Strider, if you could tell me a way to locate the position from which my value is selected ;it would be a big help. Forget sections, if I have a list of certain items say football,basketball,video games etc ,say 45 items, and my GA when runs chooses a certain item,Is there a way by which I can locate what serial number has it chosen?

[Star Strider]

Dovakin —

In order to apply GA to your problem, you have to have a cost function of some sort to minimise. If you absolutely have to use GA, then use pdist2 in your cost function to minimise the distance between the parameters you select and the parameters in your database. That’s how I would do it.

You could return the item number that represented the least distant item in your data if you told your objective function to do that.

I still believe that using a GA is a really awkward way to solve this problem!

Answer 2

[Alan Weiss]

If you are solving a mixed-integer optimization problem using GA, then you can follow the example on choosing among Discrete Non-Integer Variable Constraints.

Alan Weiss

MATLAB mathematical toolbox documentation