Problem with Genetic Algorithm using the global optimization toolbox

Hi,

I am implementing an optimization problem using GA in Matlab. I use the Matlab Global optimization toolkit. It gives fine results when I use it without non linear constraints.

However when I add even a trivial nonlinear constraint (such as c = -1, in the i.e. in reality this constraint is always satisfied.), the GA does not proceed over 1 generation.

Below is an example of output from the Solver, which I manually terminated:

1 504 -64.5333 0 0

As you see, the solver has computed the fitness function 504 times and is still stuck in the 1 st generation.

My nonlinear function is as follows:

***************************************

function [c, ceq] = nonlcon(ae)

c = -1; ceq = [];

end

***************************************

As you can see, the nonlinear constraint is always satisfied. I must emphasis that the solver gives the expected result when I run without using the nonlinear constraint function option.

Vinay

Why is this so? I

[vinay]

Update: The solver did proceed to the second generation, but after 40,000 calls to the fitness function! The constraint is very trivial. I do not understand why it evaluated the fitness so many times.

Answer 1

[Geert]

Can you give some more details about your genetic algorithm? How many variables, crossover/mutation operator, ... Maybe a code snippet or something?

Since your nonlinear contraint is of the form

function [c, ceq] = nonlcon(ae)

c = -1; ceq = [];

end

I assume that you optimize over a singly variable (i.e. length(ae) = 1)?

Please provide us with more details, if you want us to solve your problem...

[vinay]

No of Variables : 3

Pop Size: 8

Crossover: CrossoverIntermediate

Mutation: MutationUniform

Fitness Function: I calculate the Peak Gain of Antenna Array Distribution.

No Linear constraints.

The solver works well when I do not include the nonlinear constraint.

Answer 2

[Alan Weiss]

The inclusion of a nonlinear constraint function causes ga to perform a very different algorithm. See Nonlinear Constraint Solver Algorithm.

Alan Weiss

MATLAB mathematical toolbox documentation

[vinay]

Dear Alan,

I also tried Pattern Search. I face the same problem. Without including the non linear constraints solver I get expected answer. However on including the constraints the PS solver doesnt proceed over 1 poll. I does 2-3 Iterations. (Number of Variables in Problem is N = 3, and I use GPS2N solver)

I read that both GA and Pattern Search use the same Nonlinear Constraints Solver Algorithm.