An adaptive many-objective evolutionary algorithm based on decomposition with two archives and an entropy trigger

Cao, Li and Wang, Maocai and Vasile, Massimiliano and Dai, Guangming and Wu, Huanqin (2023) An adaptive many-objective evolutionary algorithm based on decomposition with two archives and an entropy trigger. Engineering Optimization. ISSN 0305-215X (https://doi.org/10.1080/0305215X.2023.2283038)

[thumbnail of Cao-etal-EO-2023-An-adaptive-many-objective-evolutionary-algorithm-based-on-decomposition] Text. Filename: Cao-etal-EO-2023-An-adaptive-many-objective-evolutionary-algorithm-based-on-decomposition.pdf
Accepted Author Manuscript
Restricted to Repository staff only until 11 December 2024.
License: Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 logo
Download (3MB) | Request a copy

Abstract

This article proposes two novel mechanisms to improve the performance of many-objective evolutionary algorithms based on Chebyshev scalarization. One mechanism improves the efficiency and effectiveness of the adaptation of the descent directions in criteria space, while the other ensures that extreme solutions are preserved. Weight adaptation via WS-transformation has shown promising results, but its performance is dependent on the choice of the start of the adaptation process. In order to overcome this limitation, in this article an efficient entropy-based trigger is proposed with fast calculation of the entropy that scales favourably with the number of dimensions. The novel entropy-based method is complemented by a dual-archiving mechanism that preserves extreme solutions. The dual-archiving strategy mitigates the possibility to discard those critical individuals whose loss affects the whole evolutionary process. The new algorithm proposed in this article (called aMOEA/D-2A-ET) is compared against a set of state-of-the-art MOEAs and shown to have competitive performance.

CORE (COnnecting REpositories)