An investigation of genetic operators for continuous parameter space
Tovi Grossman, Yuval Davidor · University of North Texas Digital Library (University of North Texas) · 1994
The success of a genetic optimization algorithm in continuous parameter space depends on the recombination (crossover) operators that it uses. In this paper we consider a wide spectrum of such operators within a unified framework and study their relative importance in the search process. We consider four basic types recombination operators which cover the relevant exploration potential of a continuous space: Interpolation, Extrapolation, Exchange and Mutation. Each of these basic types may have several variants. We characterize the various operators and their variants by their spatial sampling properties and examine their contributions to the search by applying different mixtures of the operators in several benchmark problems. The results suggest that the optimal mixture of operators may depend on the problem. But, in general, all basic types are needed for efficient optimization.