An active-set evolution strategy for optimization with known constraints

Abstract

We propose an evolutionary approach to constrained optimization where the objective function is considered a black box, but the constraint functions are assumed to be known. The approach can be considered a stochastic active-set method. It labels constraints as either active or inactive and projects candidate solutions onto the subspace of the feasible region that is implied by rendering active inequality constraints equalities. We implement the approach in a (1 + 1)-ES and evaluate its performance using a commonly used set of test problems.