Yes, if is only important the error and not its signal. This is frequent is some aplications

Erik Cuevas thanks for your response, have you used this kind of fitness function before? i ask this because i have a doubt with the fitness of the individual, as for my work i just need to change two "genes" per chromosome and i don't know if it's properly to just take this genes to work the rest of the algorithm in base of them or use all the vector data.

If you have some information or pages where i can resolve this kind of questions i'd be very grateful.

Thanks for your attention

Fitness (or misfit), in any kind of fitting algorithm, is the distance between current model and reality. You may try very different kinds of distances, your is one possible. I'm not sure what do you mean by abs in your post. Is it maxi | ideali-expi | or rather sumi | ideali-expi | ? In fact it is not very interesting, both should work, as well as Euclidean distance. Genetic Algorithms do not require fitness to be a smooth function of its arguments. So the choice is yours. What is the most preferable in your case: smallest average distance, smallest maximal deviation or smallest mean square root deviation (MSD)? But don't be surprised when those choices converge at different speed and produce different results.

By the way: in genetic algorithms we have not only mutations (as you prefer) but mostly crossovers of chromosomes. Mutations only make GA a version of Monte Carlo search.