An empirical comparison of two crossover operators in real-coded genetic algorithms for constrained numerical optimization problems

This paper presents an empirical analysis of two well-known crossover operators in real-coded genetic algorithms: Blend Crossover (BLX-a) and Simulated Binary Crossover (SBX), for constrained numerical optimization problems. The aim of the study is to analyze the ability of each operator to generate feasible solutions and also suggest suitable variation operator parameter values for such purpose. A performance measure is proposed to evaluate the capacity of each operator to find feasible offspring. A set of fourteen benchmark problems is used in the experiments. The results show that in both crossover operators the exploration ability must be enhanced so as to get better results.