Reconsidering constraint release for active-set evolution strategies
Dirk V. Arnold · Proceedings of the Genetic and Evolutionary Computation Conference · 2017
We consider the problem of solving constrained numerical optimization problems where the objective function is a black box, but the constraint functions are known explicitly. A recently proposed active-set approach implemented in an evolution strategy that interleaves the evolution of the active set with the search for better candidate solutions is able to solve unimodal problems from a commonly used test function set with relatively small numbers of objective function evaluations. We observe that the algorithm may under some conditions exhibit long phases of stagnation and propose a novel policy for considering constraints for release from the active set. The algorithm using the revised policy is seen to be able to avoid the stagnation observed in runs of the original strategy.