You can do a sensitivity study to fix the parameters. I use to do that.

Regards !

That's not a trivial task. That's way many people resort to self-adaptive techniques to do that. I feel that this is the way to go, otherwise you may need to change the parameter setting even when you change the test instance to obtain the best results (but this is not acceptable in comparisons)

You may like to see our paper of 2009 and also all you can find about adaptive parameter control.

http://ieeexplore.ieee.org/xpls/icp.jsp?arnumber=4633338

Article Limitations of Existing Mutation Rate Heuristics and How a R...

Hello sir

I have used GA in my papers.

You just see how did I use it.

GA is a powerfull method provided you know how to use it effectively.

Best regards

Dr.Indrajit Mandal

GA is very intuitive and generally powerful. You must adapt efficiently the code for representation of each individual to your problem (only simple GA have binary code), as well as the three operators (selection, crossover and mutation). For example, each crossover or mutation must produce always feasible solutions, the children must have always the crossed characteristics of their parents, and mutation must provide the consideration always of random genes in the search space. Also, the selection must select effectively and randomly the best individuals. If you can model code and operators correctly, generally you do not need to worry much with the parameters you use. Generally the parameters just depend from the dimension of the search space and can be similar to any conventional GA. However, an exception can occur with the mutation rate that may need to be adapted to the particular problem, in order to avoid that an excessive lost of gene diversity occurs along the generations.

You probably want to check out Gusz Eiben's overview paper on exactly this topic.

http://www.cs.vu.nl/~gusz/papers/2011-Eiben-Smit-Tuning-for-configuring-and-analysis.pdf

Article Parameter tuning for configuring and analyzing evolutionary algorithms

You can use design of experiments (DOE) to find the appropriate parameter combination for GA. See the paper- http://www.sciencedirect.com/science/article/pii/S0305048305000174 

There's a big fat book about it:

Lobo, F. G., Lima, C. F., & Michalewicz, Z. (2007). Parameter setting in evolutionary algorithms (Vol. 54). Berlin: Springer.

There's a very classical paper from Grefenstette that got the ball rolling

http://ieeexplore.ieee.org/xpl/login.jsp?tp=&arnumber=4075583&url=http%3A%2F%2Fieeexplore.ieee.org%2Fxpls%2Fabs_all.jsp%3Farnumber%3D4075583

And of course our very own paper that uses statistical experimenta design, as proposed by our colleague

http://ieeexplore.ieee.org/xpl/login.jsp?tp=&arnumber=1009128&url=http%3A%2F%2Fieeexplore.ieee.org%2Fxpls%2Fabs_all.jsp%3Farnumber%3D1009128

settings of parameters of GA always depends on the task you want to solve. Simple GA use population size from 20 to 100 individuals. Mutation rate depends on how multtimodal is the optimization problem, you can start with 5% then go up to 30-40% more is not usual. max step of local mutation, is very sensitive setting I use 1/100 of search space size. Selection method which works very well in general is tournament selection. But when is the task very hard you need to think about parallel genetic algorithms, but there is more settings to be set.

Please see Improvements in Genetic Algorithms. Vasconcelos et al, 2001, which shows how to link the mutation rate and the crossover rate with the genetic diversity (mean fitness/maximum fitness)

http://www.cpdee.ufmg.br/~jramirez/publicacoes/p2001-1.pdf

The value of operating parameters of GA can be chosen by conducting pilot trials

DOE can be used to set the GA parameters

Please refer to the following articles which use Design of experiments to set GA parameters:

Rojas, I., González, J., Pomares, H., Merelo, J. J., Castillo, P. A., & Romero, G. (2002). Statistical analysis of the main parameters involved in the design of a genetic algorithm. Systems, Man, and Cybernetics, Part C: Applications and Reviews, IEEE Transactions on, 32(1), 31-37.

Tsai, J. T., Liu, T. K., & Chou, J. H. (2004). Hybrid Taguchi-genetic algorithm for global numerical optimization. Evolutionary Computation, IEEE Transactions on, 8(4), 365-377.

Reeves, C. R., & Wright, C. C. (1999, January). Genetic algorithms and the design of experiments. In Evolutionary Algorithms (pp. 207-226). Springer New York.

Han, S. A. N. G. H. E. O. N. (2002). Design of experiments to identify optimal parameters of genetic algorithm. Scientiae Mathematicae Japonicae, 55(3; ISSU 213), 539-546.

http://data.eefocus.com/08-03/4155_1206094940/File/1206980806.pdf

http://www.jams.or.jp/scm/contents/Vol-5-4/5-34.pdf

Article Hybrid Taguchi-Genetic Algorithm for Global Numerical Optimization