Regularity model based offspring generation in surrogate-assisted evolutionary algorithms for expensive multi-objective optimization

Evolutionary algorithms face significant challenges when it comes to solving expensive multi-objective optimization problems, which require costly evaluations. One of the most popular approaches to addressing this issue is to use surrogate models, which can replace the expensive real function evalua...

Full description

Saved in:

Bibliographic Details
Published in:	Swarm and evolutionary computation Vol. 86; p. 101506
Main Authors:	Li, Bingdong, Lu, Yongfan, Qian, Hong, Hong, Wenjing, Yang, Peng, Zhou, Aimin
Format:	Journal Article
Language:	English
Published:	Elsevier B.V 01.04.2024
Subjects:	Expensive optimization Multi-objective optimization Pareto set learning Regularity model Surrogate assisted evolutionary algorithm 99-00 Surrogate assisted evolutionary algorithm Expensive optimization Pareto set learning 00-01 Multi-objective optimization Regularity model
ISSN:	2210-6502
Online Access:	Get full text
Tags:	Add Tag No Tags, Be the first to tag this record!

Description
Summary:	Evolutionary algorithms face significant challenges when it comes to solving expensive multi-objective optimization problems, which require costly evaluations. One of the most popular approaches to addressing this issue is to use surrogate models, which can replace the expensive real function evaluations with cheaper approximations. However, in many surrogate-assisted evolutionary algorithms (SAEAs), the process of offspring generation has not received sufficient attention. In this paper, we propose a novel framework for expensive multi-objective optimization called RM-SAEA, which utilizes a regularity model (RM) operator to generate offspring more effectively. The regularity model operator is combined with a general genetic algorithm operator to create a heterogeneous offspring generation module that can better approximate the Pareto front. Moreover, to overcome the data deficiency issue in expensive optimization scenarios, we employ a data augmentation strategy while training the regularity model. Finally, we embed three representative SAEAs into the proposed RM-SAEA to demonstrate its efficacy. Experimental results on several benchmark test suites with up to 10 objectives and real-world applications show that RM-SAEA achieves superior overall performance compared to eight state-of-the-art algorithms. By focusing on more effective offspring generation and addressing data deficiencies, our proposed framework is able to generate better approximations of the Pareto front and improve the optimization process in expensive multi-objective optimization scenarios.
ISSN:	2210-6502
DOI:	10.1016/j.swevo.2024.101506