A greedy hypervolume polychotomic scheme for multiobjective combinatorial optimization

The usual goal in multiobjective combinatorial optimization is to find the complete set of nondominated points. However, in general, the nondominated set may be too large to be enumerated under a tight time budget. In these cases, it is preferable to rapidly obtain a concise representation of the no...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Veröffentlicht in:Computers & operations research Jg. 183; S. 107140
Hauptverfasser: Lopes, Gonçalo, Klamroth, Kathrin, Paquete, Luís
Format: Journal Article
Sprache:Englisch
Veröffentlicht: Elsevier Ltd 01.11.2025
Schlagworte:
Greedy hypervolume algorithm
Hypervolume scalarization
Local bounds
Multiobjective combinatorial optimization
Representation of the nondominated set
Multiobjective combinatorial optimization
Hypervolume scalarization
Local bounds
Greedy hypervolume algorithm
Representation of the nondominated set
ISSN:0305-0548
Online-Zugang:Volltext
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
Beschreibung
Zusammenfassung:The usual goal in multiobjective combinatorial optimization is to find the complete set of nondominated points. However, in general, the nondominated set may be too large to be enumerated under a tight time budget. In these cases, it is preferable to rapidly obtain a concise representation of the nondominated set that satisfies a given property of interest. This work describes a generic greedy approach to compute a representation of the nondominated set for multiobjective combinatorial optimization problems that approximately maximizes the dominated hypervolume. The representation is built iteratively by solving a sequence of hypervolume scalarized problems, each of which with respect to k reference points, which is a parameter of our approach. We present a mixed-integer formulation of the hypervolume scalarization function for k reference points as well as a combinatorial branch-and-bound for the m-objective knapsack problem. We empirically analyse the functional relationship between k and its running-time and representation quality. Our results indicate that the branch-and-bound is a much more efficient approach and that increasing k does not directly translate into much better representation quality.
ISSN:0305-0548
DOI:10.1016/j.cor.2025.107140