An Adaptive Penalty Method for Genetic Algorithms in Constrained Optimization Problems

Abstract

From the experiments performed here one can see that the original APM found more often the best solutions. Among the new variants, the variants 3 and 5 were the most successful. As the performance of variant 4 was slightly inferior to that of variants 3 and 5, one may suspect that the inclusion of the damping procedure was not advantageous. A more important conclusion may be that there were not large discrepancies among the results obtained by the original APM and the four new variants, which provide computational savings, and that robust variants can be obtained from the overall family of adaptive penalty methods considered here.