Global Optimization for a Simulink Model

Hi,

The problem

I need to optimize 10 paramters of a simulink model within some lower/upper bounds so that the mean error between real and simulated ouput is minimized. The problem has multiple optima.

Question

How could I apply the a genetic algorithm or a surrogate optimization as part of the Simulink Design Optimization application? if not, how to convert a Simulink design optimization problem into an objective function that I can pass to the above mentioned functions?

Thank you for your help,

Miguel

Answer 1

[Alan Weiss]

You probably want to minimize the mean squared error. For an example, see lsqnonlin with a Simulink® Model. You can change the solver from lsqnonlin to ga or surrogateopt easily, by reformulating the objective function as in Nonlinear Data-Fitting.

One word of caution: you probably cannot optimize a Simulink model in parallel by setting the UseParallel option. Please try the optimization in serial.

Alan Weiss

MATLAB mathematical toolbox documentation

[Miguel Esteras]

Thanks Alan, that worked as described in the examples.

What would be the problem of optimizing in parallel in this case?

[Alan Weiss]

To call Simulink in parallel, the underlying mechanism should be parsim. However, ga and surrogateopt do not use parsim. Instead, they use parfor or some other mechanism. I believe that Simulink Design Optimization does use the correct mechanism, but I do not know that product so cannot give you advice relating to it.

Alan Weiss

MATLAB mathematical toolbox documentation

[Miguel Esteras]

Right,... for what I have been told by a local Simulink support staff, Simulink Design Optimization might not be able to use global optimization algorithms such as ga or surrogated out of the box.