Hybrid evolutionary multi-objective optimisation using outranking-based ordinal classification methods

A large number of real-world problems require optimising several objective functions at the same time, which are generally in conflict. Many of these problems have been addressed through multi-objective evolutionary algorithms. In this paper, we propose a new hybrid evolutionary algorithm whose main...

Full description

Saved in:
Bibliographic Details
Published in:Swarm and evolutionary computation 2020-05, Vol.54, p.100652, Article 100652
Main Authors: Cruz-Reyes, Laura, Fernandez, Eduardo, Sanchez-Solis, J. Patricia, Coello Coello, Carlos A., Gomez, Claudia
Format: Article
Language:English
Subjects:
Evolutionary multi-objective optimisation
Multi-criteria ordinal classification
Preference incorporation
Citations: Items that this one cites
Items that cite this one
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:A large number of real-world problems require optimising several objective functions at the same time, which are generally in conflict. Many of these problems have been addressed through multi-objective evolutionary algorithms. In this paper, we propose a new hybrid evolutionary algorithm whose main feature is the incorporation of the Decision Maker’s (DM’s) preferences through multi-criteria ordinal classification methods in early stages of the optimisation process, being progressively updated. This increases the selective pressure towards the privileged zone of the Pareto front more in agreement with the DM’s preferences. An extensive experimental research was conducted to answer three main questions: i) to what extent the proposal improves the convergence towards the region of interest for the DM; ii) to what extent the proposal becomes more relevant as the number of objectives increases, and iii) to what extent the effectiveness of the hybrid algorithm depends on the particular multi-criteria method used to assign solutions to ordered classes. The issues used to evaluate our proposal and answer the questions were seven scalable test problems from the DTLZ test suite and some instances of project portfolio optimisation problems, with three and eight objectives. Compared to MOEA/D and MOEA/D-DE, the results showed that the proposed strategy obtains a better convergence towards the region of interest for the DM and also performs better characterisation of that zone on a wide range of objective functions.
ISSN:2210-6502
DOI:10.1016/j.swevo.2020.100652