Overlapping Cooperative Co-Evolution for Overlapping Large-Scale Global Optimization Problems
Marcin M. Komarnicki, Michal W. Przewozniczek, Renato Tinós, Xiaodong Li · Proceedings of the Genetic and Evolutionary Computation Conference · 2024
One of the main approaches for solving Large-Scale Global Optimization (LSGO) problems is embedding a decomposition strategy into a Cooperative Co-Evolution (CC) framework. Decomposing an LSGO problem into smaller subproblems and optimizing them separately using a CC framework was shown to be effective when a considered problem is partially separable. Components in CC frameworks are usually disjoint. Thus, the existence of the perfect decomposition of such problems allows of the optimization of independent components. However, for overlapping problems, the perfect, unique decomposition does not exist due to the existence of shared variables. Despite this, each variable is usually assigned to a single component, and the assignment does not change during a whole framework run. In this paper, we propose a new CC framework that allows multiple assignments of shared variables. Allocating computational resources to each of its components is influenced by other components that share variables with it. According to experimental results, our proposed method outperforms the state-of-the-art LSGO-dedicated optimization methods, including other CC frameworks, when overlapping LSGO problems are considered.