An interactive reference-point-based method for incorporating user preferences in multi-objective structural optimization problems

Dênis E. C. Vargas

doi:10.1016/j.asoc.2024.112106

2024-11-01

PRE

Dênis E. C. Vargas *

Afonso Celso de Castro Lemonge

Hélio J. C. Barbosa

Heder S. Bernardino

DOI：10.1016/j.asoc.2024.112106

原文链接

原文求助

摘要

En 中文

Although multi-objective evolutionary algorithms (MOEAs) are often used to solve multi-objective structural optimization problems (MOSOPs), the incorporation of decision-maker (DM) preferences into MOEAs to search for regions of interest from the Pareto front has received limited attention in these problems. Recently, three MOEAs have shown promising results in solving MOSOPs incorporating DM's preferences based on a priori reference points. However, providing this information can be challenging as the DM may not know an initial reference point for the MOSOP. Therefore, this study analyzes three interactive multi-objective optimization (IMO) algorithms, called I-NSGA-II, I-GDE3, and I-GDE3+APM, in which the reference points are progressively incorporated during the search process. These IMOs are applied to benchmark MOSOPs and evaluated using well-known performance indicators. The results demonstrate that I-NSGA-II outperforms IGDE3 and I-GDE3+APM, which have similar overall performances. Additionally, the IMOs considered here outperformed their respective a priori approaches in MOSOPs where the reference points are located at the central sector of the Pareto front. Also, we identified the performance indicators in which the approaches considered in this work performed better than their corresponding a priori ones. Furthermore, the research findings suggest that these IMOs can generate good results without defining a priori reference points, making them a promising approach for solving MOSOPs and enabling DMs to make better and more reliable decisions.

Keyword:

Multi-objective structural optimization

Interactive preference information

Reference point

Differential evolution

Adaptive penalty method