Multi-Population-based Parallelization of Ensemble of Differential Evolution Variants for Constrained Real Parameter Optimization

Abstract

Differential evolution (DE) algorithms face performance challenges, which lean on improving solutions quality, speed-up, and exploitation of computational resources. Parallelism represents a suitable paradigm for overcoming the DE challenges. The ensemble of differential evolution variants (EDEV) algorithm is a recent DE algorithm. EDEV constitutes three DE variants (JADE, CoDE, and EPSDE), which may decrease its speedup. In this paper, a multi-population parallel ensemble of differential evolution variants (MPPEDEV) is proposed based on the synchronous master/slave parallel model. The performance of the proposed MPPEDEV is tested using a constrained real parameter problem proposed in CEC 2006. Compared to four state-of-the-art DE algorithms, which are JADE, CoDE, EPSDE, and EDEV, the results show that MPPEDEV outperforms EDEV in terms of execution time and solutions quality, depending on the population size as a control parameter. Furthermore, MPPEDEV and EDEV outperform JADE, CoDE, and EPSDE in terms of solutions' quality.