A New Breeding Crossover Approach for Evolutionary Algorithms

In a previous work, we introduced a population-based, bio-inspired algorithm. The proposed algorithm is inspired by the biological animal life cycle, consisting of the birth, growth, reproduction, and death stages. Our algorithm was initially based on the canonical Genetic Algorithm (GA), where all the individuals have a genotype (chromosome). One difference to highlight in our algorithm is that both the crossing and the mutation are executed through independent processes that randomly affect the population. This paper focuses on breeding, whereas in earlier versions of the algorithm, we used the traditional GA one-point crossover. In this paper, we propose a different alternative to the classical approach, where part of the genetic information is directly copied to each of the offspring in the crossover operator, where this type of crossover may not perform well in continuous optimization problems. In this proposal, we use the parent’s genetic information for each gene, using those values as lower and upper bounds of a range, where a random value within that range determines the new value for that gene index of the offspring. This is similar to what is used by algorithms such as Differential Evolution, where we consider our proposal as a variation of existing proposals. We expect this new operator to allow the offspring to continue exploring new search spaces with the birth of individuals. In this paper, we use the benchmark functions introduced in the Competition on Evolutionary Computation for the 2017 edition (CEC-2017) to compare the traditional one-point crossover and our proposed strategy. Experimental results indicate that our proposed operator may be a good alternative for the canonical crossover.