A parabolic operator for parameter optimization problems

Parameter optimization has been a prime target for evolutionary algorithms for a number of years. Genetic algorithms, evolution strategies, and evolutionary programming have dealt with a variety of nonlinear programming problems. There is a growing evidence that evolutionary algorithms are well suited for optimization of real valued multi-modal difficult functions of many variables. Despite this success story, there are still many open, interesting questions. One of them deals with a relationship between the recombination operators and the landscape of the problem; it seems that different problems "require" different operators. We propose a new multi-parent crossover operator: a parabolic crossover, which works very well for certain types of landscapes. Additionally, this operator maintains an interesting balance between its exploratory and exploitative capabilities and has potential for further generalizations.