Comparação entre técnicas de tratamento de restrições em algoritmos evolutivos

Abstract

Constrained optimization problems are commonly found in many knowledge areas. Several approaches are available to reach reliable and feasible solutions for these problems. In this context, handling the problem imply in finding not only an efficient algorithm but also an adequate method to reliably handle infeasible solutions. The present work investigates some well-known approaches to handle constraints in evolutionary algorithms, measuring its performances in two different searching algorithms, differential evolution and particle swarm optimization. The well-known CEC 2006 suite of benchmark functions was the set of problems used as case studies. The work highlights the remarkable characteristics of each handling technique as the sensitiveness to the parameter configuration of the algorithms, the stableness in delivering feasible solutions and the amount of best overall solutions.