One can choose an evolutionary algorithm based on the problem on hand if the problem can be represented as graph then Ant colony optimization algorithm would be an ideal choice.If the the problem is continuous then PSO would perform better provided adequate care is taken to prevent the the algorithm from converging into local optimas. For combinatorial optimization genetic algorithms are a better optimization .Genetic algorithms don't get trapped in  local optimas but the solutions obtained are more coarse when compared to that of a PSO.

Have a look at the book: Experimental research in evolutionary computation. Also do some googling on the "no free lunch theorem" in the context of optimisation.

Dear Fourie.

I know the no free lunch theorem but I want to know when we have an optimization problem what criterion makes you choose the algorithm you use,  factors like familiarity with it or its ease of implementation or .....?

For example both PSO and ES are good algorithms for continuous parameter optimization in high dimensional search space , how to pick one of them for a continuous parameter optimization problem.

Dear Hossein Abedi 

Please check this out

http://osp.mans.edu.eg/elbeltagi/AI%20Comparison%205EAs.pdf

The paper presents a brief description of each algorithm. Benchmark comparisons among the algorithms are presented for both continuous and discrete optimization problems, in terms of processing time, convergence speed, and quality of the results. Based on this comparative analysis, the performance of EAs is discussed along with some guidelines for determining the best operators for each algorithm.

The PSO method was generally found to perform better than other algorithms in terms of success rate and solution quality, while being second best in terms of processing time.

Thanks Mohammad,

Nice Paper.

In the context of CEC competition, my option will be to test 2-3 algorithms (which show the most potential), and based on the general results (average), choose the best suited approach, which you can then improve with different techniques. I know that it is a trial and error approach but I do not think that there is other methodology, as the CEC competition has a wide range of problem types.

You do not need to choose just "one". You can have a set of metaheuristics solving the problem cooperatively. In such set you may have different methods or several copies of one of them with different parameters. We have doing this since a long time ago.

 Using memory and fuzzy rules in a co-operative multi-thread strategy for optimization

David Pelta, Alejandro Sancho-Royo, Carlos Cruz and Jose L. Verdegay

in Information Sciences, Volume 176, Issue 13, 3, Pages 1849-1868 2006

A centralised cooperative strategy for continuous optimization: the influence of cooperation in performance and behaviour

Antonio D. Masegosa, David A. Pelta, José L. Verdegay

Information Sciences, Volume 219, Pages 73–92, January 2013,

Good luck

David

There is no scientific reason! based on the journal you want to submit the paper as well as the topic of your work you will choose it. Don't forget to claim that's the best way to solve the problem. 

Hi Hossein,

I agree the friends' comments, especially Elena and David. You can choose some of the famous metaheuristics and evaluate them in terms of some criteria like solution quality, CPU time, etc. then decide to select the best suited method. 

Be advise that parameter tuning for metaheuristics plays more than 50% of success of an algorithm in my opinion. So compare metaheuristics when they are tuned. Taguchi method can help you. 

Based roughly on the principle of correspondence:either the esoteric hermetic version or for instance a version known to Stephen Wolfram as "the principle if computational equivalence".And to bring matters to a more down-to-earth level: the hardware's ability to handle the maximal computational complexity indicated by the operations of the algorithm.

mathworld.wolfram.com/PrincipleofComputationalEquivalence.html