A survey on expensive optimization problems using differential evolution

doi:10.1016/j.asoc.2025.112727

Return

A survey on expensive optimization problems using differential evolution

2025-02-01

PRE

Ren, Chongle

Meng, Zhenyu *

DOI：10.1016/j.asoc.2025.112727

Original

Original request for help

Save

Abstract

En 中文

Many real-world problems that require substantial execution time and computational resources for evaluating candidate solutions can be termed Expensive Optimization Problems (EOPs). The major challenge in solving EOPs lies in that evaluating the candidate solutions comes with a high and sometimes even prohibitive cost, thereby presenting notable obstacles for current optimization methods. Differential Evolution (DE), as a branch of evolutionary algorithms, has been widely used to tackle various optimization problems due to its fast convergence speed and powerful search capability. However, the performance of DE may deteriorate rapidly when solving EOPs where the number of function evaluations is limited. Recently, considerable efforts have been dedicated to improving DE-based algorithms for solving EOPs. However, there is a lack of a systematic survey on modifications to DE-based algorithms for tackling EOPs. By collecting and analyzing current DE- based algorithms, this survey presents a comprehensive overview of modifications on DE-based algorithms. The paper divides these modifications into three categories: framework improvement, surrogate-assisted approximation, and parallel and distributed implementation. According to the type of EOPs, existing DE- based algorithms are also classified and analyzed. Lastly, we present current challenges and future directions of employing DE-based algorithms for EOPs. By providing both novice and experienced researchers with a fresh perspective on utilizing DE-based algorithms for tackling EOPs, this survey aims to assist researchers in designing more efficient algorithms for EOPs.

Keywords:

Differential evolution

Expensive optimization problem

Hybrid techniques

Parallel and distributed techniques

Surrogate model