Comparison of selection schemes for evolutionary constrained optimization

Evolutionary algorithms simulate the process of evolution in order to evolve solutions to optimization problems. An interesting domain of application is to solve numerical constrained optimization problems. We introduce a simple constrained optimization problem with scalable dimension, adjustable complexity, and a known optimal solution. A set of evolutionary algorithms, all using different selection schemes, is applied to this problem. The performance of the evolutionary algorithms differs strongly. Selection schemes that only use a limited number of offspring as parents for the next generation consistently outperform the schemes that accept all offspring as parents and adjust their fertility based on (relative) fitness during the experiments.