Multistage covariance matrix adaptation with differential evolution for constrained optimization

Abstract

Single Objective minimizations often involve simultaneous satisfaction of a number of conditions, known as constraints. MCMADE proposes a two-stage algorithm having an initial CMA or Covariance Matrix Adaptation phase and a subsequent Differential Evolution strategy in the second phase. The two phases are synchronized using a stagnate parameter. To handle the constraints, a simple penalty function, without any penalty parameter has been employed which adds the margin of violations to the fitness value of each particle in the landscape. MCMADE has been tested on the problem set specified by the CEC 2010 benchmark. © 2012 Springer-Verlag.