Dear Mohammad Bazmara, In Genetic Algorithm the individuals are coded as integers but in Evolutionary Strategy the individuals are coded as vectors of real numbers.

they have their own advantage and disadvantages. if finding optimal solution is more important than GA is suitable but it is bit slow, however if “Good enough” solution

acceptable than other one should be prefer as it provides faster results.

Dear Mohammad Bazmara, In Genetic Algorithm the individuals are coded as integers but in Evolutionary Strategy the individuals are coded as vectors of real numbers.

Don't forget to compare against ramdom search, you might find surprising results...

Thanks a lot for your answers

As a rule of thumb, use inherently continuous optimisation approaches (e.g. PSO, DE, ES, EP, etc.) to solve continuous optimisation problems and inherently discrete methods (e.g. GA, TS, PR, etc.) when you want to solve discrete problems.

But as already said, it depends on the problem at hand.

Tnx dear mahamed

You are most welcome!

@Mahamed Omran: There is no definitive answer. Both strategies can be used for solving optimization problems. I have worked with the following meta-heuristics: GA (we developed the "Epidemic GA", producing better results than standard GA), SA (alternative scheme for better result: re-annealing, or SA with thermal cycles), ACO (Ant Colony Optimization), GEO (Generalize Extreme Optimization: based on self-organized criticality), MPCO (Multi-Particle Collision Algorithm), DE (Differential Evolution), PSO (we developed PSO-t: Particle Swarm Optimization with turbulence), Firefly (we developed the "firefly with predation"). We wrote a book (in Portuguese) dealing with all these optimization schemes for solving an inverse problem: properties identification in radiative transfer.

From the results, the conclusion is:

1. GA vs DE (standard genetic algorithm vs an evolutionary strategy): for the cited inverse problem, the DE obtained better result.

2. A hybrid strategy (meta-heuristic + deterministic method) produces better results. The deterministic methods that we used: quasi-Newton, Conjugate Gradient, Simples, Levenberg–Marquardt (the inverse problem cited is parameter identification problem, for function determination L-M method is not suggested).

Hi Haroldo,

I did say at the end of my message "But as already said, it depends on the problem at hand.". What I said is a rule of thumb not something proven.

It depends on the problem,data and representation, both of them have some merit and also some lack

Depend on the data that used for the expermints

If you solve continous Problems EVO seems apprpriate, while for discrete Problems genetic approaches might be more suitable 

ES is quite suitable for continuous domain problems and GA suitable for discrete domain

There is a theorem for this, called ‘No Free Lunch' theorem, stating that the average of the performance of algorithms on all the problems are actually identical. It's not about which optimization algorithm is better, it is about what kind of problems you want to solve.

There is not an algorithm that can competitively solve all problems. I suggest to use both and evaluate their performances