Using a genetic algorithm to optimize problems with feasibility constraints

G.E. Liepins et al. (1990) have shown that genetic algorithm optimization of certain combinatorial optimization problems can be more effective when the genetic algorithm evaluates "repaired" versions of the chromosomes. In this sense "repairing" a chromosome means to take an illegal chromosome and force it to be legal through some repair function, Liepens does not however, advocate returning this "repaired" version of the chromosome to the population. Other researchers also advocate "repairing" illegal chromosomes, but they claim that returning the "repaired" version of the chromosome to the population "improves the convergence rates and solution quality". In this paper we show that a combination of these strategies supports another performance enhancing technique, a variety of Lamarckian evolution in which, with a small probability, original chromosomes are replaced in the population by their "repaired" counterparts. Using this strategy, replacing original chromosomes with a 5% probability, on three different combinatorial optimization problems, performance was better than either never replacing or always replacing.<
>