A Distributed Cooperative Co-evolutionary CMA Evolution Strategy for Global Optimization of Large-Scale Overlapping Problems

Abstract

Distributed cooperative co-evolution (DCC) is an effective way to solve large-scale optimization problems. The cooperative co-evolution methodology can reduce the optimizing complexity by dividing a large-scale problem into small subcomponents, and distributed computation can accelerate the optimizing speed. However, existing DCC algorithms often encounter deficiency in overlapping problems that cannot be ideally divided due to unavoidable overlaps between subcomponents. To address this issue, this paper proposes an algorithm called distributed cooperative co-evolutionary covariance matrix adaptation evolution strategy (DCCMAES) to solve large-scale overlapping optimization problems. First, a new grouping scheme is introduced, where a variable, if needed, can be assigned into multiple subcomponents. Second, to address the conflicts caused by optimizing multiple copies of shared variables, a random orthogonal experiment method is proposed to generate global solutions. Moreover, three cooperation schemes are proposed to coordinate the optimizers in DCCMAES during evolution. The experiments on benchmark functions and a real-world application show that DCCMAES is promising for global optimization of large-scale overlapping problems.