Self-Adaptive Ensemble-based Differential Evolution with Enhanced Population Sizing

Abstract

Differential Evolution (DE) a branch of Evolutionary Algorithm (EA) is recently prominently sough after in teerms of global optimization purpose. The pinpointed issues for this work sheds light on its impact towards population diversity and dimension size. In retrospect of balance incorporation among exploration and exploitation capability, DE emphasizes on preference of control parameters. The algorithm scheme, known as self-adaptive ensemble-based differential evolution with enhanced population sizing (SAEDE- EP), is compared to self-adaptive ensemble-based DE (SAEDE). The SAEDE-EP algorithm is proposed to minimize user setting and exhausting trial-and-error procedure for appropriately configuring scale factor, crossover rate, mutation strategy, and population size. In SAEDE-EP, an ensemble is used to trigger the change of diversity in population when the best solutions between generations have been stagnant for a certain period. The performance is appraised based on 26 benchmark functions comprising on 8 low dimensions and 18 high dimensions. Experimental results indicate that SAEDE-EP achieves maximum success rate with better efficiency in 24 out of the 26 benchmark functions.