Performance analysis of permutation cross—over genetic operators
L. Djerid, Mane-Claude Portmann, Pierre Villon · Journal of Decision System · 1996
We are interested here in the utilization of Genetic Algorithms (GA) as approximation methods for combinatorial optimization. They are stochastic methods using genetic operators on a population. For their design, two points must be worked: the general scheme of the algorithm with parameter adjustment and the design of chromosome contents and genetic operators. The first point poses globally no specific problem. The latter one involves difficulties when binary genes inside the chromosomes are replaced by more general information such as permutations, in this case, the performance of the genetic operators must be analyzed with respect to the considered specific problems and their specific criteria. As performance of mutation operators are analogous to those of neighborhood operators used in well known local search methods, we will focus here only on cross—over operator (COO) performances. The aims of this paper are first to design permutation cross—over performance indicators which express the probable trend of associated criterion variations, second to list literature proposed permutation cross—over operators and to propose some new ones with particular properties, third to present a numerical experiment which compares the different cross—over operators using the defined performance indicators.