Matlab GA toolbox coupling with external simulation code.

I have the following problem when using GA tool for optimization. My objective function is the result of simulating each population in an external code. My question is: how can I handle the evaluation of the objective function for each population, running the simulation code for each individual.

[Ameer Hamza]

Are you able to call the external simulation code from MATLAB, pass it some parameters, and get the output in return?

[Yro]

Thanks for your reply.

Yes, I can execute the code from Matlab, which for each individual in the population I have to execute the code, so I want how I can handle the evaluation of the objective function in Matlab.

Answer 1

[Star Strider]

See if the approach in Parameter Estimation for a System of Differential Equations will do what you want. It uses the ga function to optimise the parameters for differential equations integrated using ode45.

[Yro]

Thanks for your answer, I will consult it.

[Star Strider]

My pleasure!

[Yro]

Hi, any idea of how I can evaluate in an external code each individual of the population.

Thanks.

[Star Strider]

The ga function does that for you in each iteration. The result is the fitness value.

[Yro]

The problem is that each population I need to evaluate in the external simulation code and it returns the value of the objective function. Any idea how you could control each evaluation in Matlab?

Thanks in advance.

[Star Strider]

My pleasure!

I am not certain that I understand what you are asking here.

The ga function evaluates the fitness function for each ‘individual’ in the population each time. It then compares the fitness of the individuals and selects for those with the highest fitness (lowest scalar value of the fitness function). The code I referred to you that adjusts the parameters of the differential equations does exactly what you describe, so you can use it for guidance.

Note that the ‘ftns’ fitness function first calls the objective function ‘kinetics’ that uses the time vector and parameters (that define each ‘individual’ in the population each time) to integrate the differential equation system and return the desired output (here defined as ‘C’). That output is then compared to the concentrations in the ‘c’ matrix and the fitness evaluated for each individual by calculating the norm of the difference.

The ‘external simulation code’ in this example is the ‘kinetics’ function that integrates the differential equations and returns the desired output. That should be analogous to the sort of situation you are describing.

[Yro]

Thank you very much for your explanation. I am attaching a diagram of the problem I want to solve. In each iteration, for each individual in the population, I have to execute the external code (ERANOS code, for example with the system function) to evaluate and return to Matlab the result obtained for the objective function. The results using ga for an arbitrary function is the value its best value in each generation. So I have no information or how you evaluate each solution.

Thanks.

[Star Strider]

The diagram simply outlines the essence of the genetic algorithm.

You can likely substitute ‘ERANOS’ for ‘kinetics’ in my code, obviously with the appropriate changes. (I have no idea what ‘ERANOS’ is, so I cannot help you with it.)

Answer 2

[Ameer Hamza]

You can write a MATLAB function wrapper around the call to the external simulator. Something like this

function f = objFun(x)
% call your external simulation here, pass value of x 
% and return the objective function value y.
y = call_external_simulation(x);
end