Control of Crossed Genes Ratio for Directed Mating in Evolutionary Constrained Multi-Objective Optimization
Minami Miyakawa, Keiki Takadama, Hiroyuki Satō · 2015
As an evolutionary approach to solve constrained multi-objective optimization problems (CMOPs), an algorithm using the two-stage non-dominated sorting and the directed mating (TNSDM) has been proposed. To generate offspring, the directed mating utilizes useful infeasible solutions having better objective values than feasible solutions in the population. The directed mating achieves higher search performance than the conventional mating which avoids using infeasible solutions in several CMOPs. However, since the directed mating uses infeasible solutions, generated offspring tend to be infeasible compared with the conventional mating. To further improve the effectiveness of the directed mating by improving the feasibility of generated offspring, in this work we propose a method to control the crossed genes ratio in the directed mating. In this method, we control the amount of genes copied from infeasible parents to offspring in the directed mating. Experimental results using m-objective k-knapsack problem with 2-4 objectives show the contribution of the directed mating for the search performance is further improved by controlling crossed genes ratio.