A self-selecting framework for control parameters and mutation strategies in differential evoluation

This study aims to identify optimal parameter configurations in Differential Evolution (DE) by considering population diversity and dimension size. The goal is to select control parameters that balance exploitation rate and algorithm performance. The proposed approach introduces two adaptation stages - microadaptation and macroadaptation - to influence offspring generation and maintain population diversity. An adaptive population sizing scheme called Self-Adaptive Ensemble-based DE with Enhanced Population Sizing (SAEDE-EP) further enhances diversity. Microadaptation and macroadaptation are triggered based on individual fitness and solution stagnation. SAEDE-EP was evaluated on 26 benchmark functions, the CEC 2019 challenge suite, and 57 constrained optimization benchmarks from CEC 2020. Findings showed SAEDE-EP achieved efficient optimization across unimodal, multimodal, hybrid, and composition problems without exhaustive parameter tuning. Comparative analysis indicated that SAEDE-EP performed well in single-objective unconstrained optimization problems, demonstrating faster computation times within the CEC evaluation limit for optimal solutions than jDE100, despite the latter achieving a lower ratio in DC02 and DC03, requiring approximately 5 and 15 times more function evaluations than the specified CEC limit. Given these promising findings, the proposed techniques pave the way for more capable and autonomous DE optimization. Future work should explore alternative adaptation schemes, analyze parameter impacts, and incorporate additional mutation strategies to further advance DE.