2754. "Enhancing Convergence and Diversity in Evolutionary Multiobjective Optimization via Dynamic Reference Point Adaptation"
Invited abstract in session TC-51: Multiobjective Decision Making, stream Multiobjective and vector optimization.
Tuesday, 12:30-14:00Room: Parkinson B22
Authors (first author is the speaker)
| 1. | Antonio Borrego Ortega
|
| Economía Aplicada (Matemáticas), Universidad de Málaga | |
| 2. | Rubén Saborido
|
| University of Málaga | |
| 3. | Mariano Luque
|
| Applied Economics (Mathematics), Universidad de Malaga |
Abstract
Multiobjective optimization problems require algorithms that balance convergence and diversity in approximating the Pareto optimal front. The performance of these algorithms heavily depends on the front’s shape. This paper presents a framework designed to enhance the later stages of Evolutionary Multiobjective Optimization (EMO) algorithms. We divide a Pareto front approximation into subregions using a distributed set of reference points in the objective space. For each subregion, weight vectors are computed and dynamically adapted based on the front's geometry. This adaptation improves both convergence and diversity. Additionally, we explore how to optimally determine the number of reference points to ensure a uniform distribution of solutions. A series of computational experiments were conducted using the WFG benchmark suite, comparing the performance of our approach integrated into NSGA-III with the original NSGA-III. The comparison, based on the hypervolume indicator, shows the advantages of our method. This study emphasizes the potential of such mechanisms in EMO algorithms, particularly for bi-objective problems, and sets the stage for future extensions to many-objective optimization.
Keywords
- Programming, Multi-Objective
- Algorithms
- Mathematical Programming
Status: accepted
Back to the list of papers