Increasing parallelism of evolutionary algorithms by Nash games in design inverse flow problems

Abstract

PurposeThe purpose of this paper is to compare the efficiency of four different algorithmic parallelization methods for inverse shape design flow problems.Design/methodology/approachThe included algorithms are: a parallelized differential evolution algorithm; island‐model differential evolution with multiple subpopulations; Nash differential evolution with geometry decomposition using competitive Nash games; and the new Global Nash Game Coalition Algorithm (GNGCA) which combines domain and geometry decomposition into a “distributed one‐shot” method. The methods are compared using selected academic reconstruction problems using a different number of simultaneous processes.FindingsThe results demonstrate that the geometry decomposition approach can be used to improve algorithmic convergence. Additional improvements were achieved using the novel distributed one‐shot method.Originality/valueThis paper is a part of series of articles involving the GNGCA method. Further tests implemented for more complex problems are needed to study the efficiency of the approaches in more realistic cases.