CONSTRAINT SCALING IN THE MESH ADAPTIVE DIRECT SEARCH ALGORITHM

摘要

In an optimization problem, multiplying an inequality constraint by a positive scalar has no effect on the domain. However, such a transformation might have an effect in practice. A common strategy in constrained optimization is to aggregate the sum of all constraint violation in a single real-valued function. Multiplying a constraint by a scalar impacts that function. The present work proposes a dynamic methodology to select weights for each constraint in the Mesh Adaptive Direct Search (MADS) algorithm with the progressive barrier.