A traditional penalty function formulation for treatment of nonlinear constrained optimization problems in genetic search has been shown to be extremely sensitive to user-specified schedules of selecting penalty parameters. The sensitivity ofsuch an approach is manifested in a biasing of the search towards suboptimal desings, and a general increase in the number of function evaluations required to obtain a converged design. The present paper describes alternative methods for handling constraints that derive from the fact that the structure of both feasible and infeasible designs is present in the population of designs at any generation of search. A preconditioning of the infeasible designs prior to the genetic transformations, by an "expression" operation that is conceptually analogous to the theory of dominant and recessive genes in genetics, is shown to be highly effective in evolving feasible solutions, and with savings of computational resource. The paper describes two alternative implementations of this approach and a comparison of numerical efficiency vis-a-vis the penalty function based approach.
展开▼