Development of Hybrid Real-Coded Genetic Algorithms with Active Constraints

Abstract

This paper presents a new efficient real-coded genetic algorithms for constrained optimization. In general, it is efficient to use the gradient methods for the constrained optimization. However, because actual design problems often include extremely nonlinear or discontinuous characters, the gradient methods do not necessarily work well. Therefore, the present authors developed the Center Neighborhood Crossover (CNX) for real-coded genetic algorithms and applied to truss structure optimization. But, GAs have some fatal problems in handling the constraints. In the constrained optimization using GAs, designers have to translate the constrained problem to a basic unconstrained problem by using some methods such as the penalty function methods. It is, however, difficult to decide appropriate penalty parameters. In the present study, we developed Real-coded Genetic Algorithms with Active Constraints (RGAAC) In RGAAC, the points outside feasible area is pulled back to active constraints by the gradient method. The present method enables an efficient search.