Tuning of an adaptive unified differential evolution algorithm for global optimization

Abstract

Recently, an adaptive unified differential evolution algorithm for single-objective global optimization was proposed in an internal report. Instead of the multiple mutation strategies in conventional differential evolution algorithms, this algorithm employs a single equation unifying multiple strategies into one expression. It has the virtue of mathematical simplicity and also provides users the flexibility for broader exploration of the space of mutation operators. However, the four control parameters used in the unified mutation expression might slow down the speed of convergence due to an over exploration of the search space. In this paper, we systematically studied the choice of those control parameters in the unified mutation strategy using fourteen unimodal and multimodal functions from the CEC2005 benchmark. Those numerical results suggest that the use of three control parameters in the unified mutation strategy improves the performance of the original algorithm and shows promising performance in comparison to several conventional differential evolution algorithms.