Algorithm comparison by automatically configurable stochastic local search frameworks: a case study using flow-shop scheduling problems - Inria - Institut national de recherche en sciences et technologies du numérique Accéder directement au contenu
Communication Dans Un Congrès Année : 2014

Algorithm comparison by automatically configurable stochastic local search frameworks: a case study using flow-shop scheduling problems

Résumé

The benefits of hybrid stochastic local search (SLS) methods, in comparison with more classical (non-hybrid) ones are often difficult to quantify, since one has to take into account not only the final results obtained but also the effort spent on finding the best configuration of the hybrid and of the classical SLS method. In this paper, we study this trade-off by means of tools for automatic algorithm design, and, in particular, we study the generation of hybrid SLS algorithms versus selecting one classical SLS method among several. In addition, we tune the parameters of the classical SLS method separately and compare the results with the ones obtained when selection and tuning are done at the same time. We carry out experiments on two variants of the permutation flowshop scheduling problem that consider the minimization of weighted sum of completion times (PFSP-WCT) and the minimization of weighted tardiness (PFSP-WCT). Our results indicate that the hybrid algorithms we instantiate are able to match and improve over the best classical SLS method.
Fichier non déposé

Dates et versions

hal-01094683 , version 1 (12-12-2014)

Identifiants

Citer

Franco Mascia, Manuel López-Ibáñez, Jérémie Dubois-Lacoste, Marie-Eléonore Marmion, Thomas Stützle. Algorithm comparison by automatically configurable stochastic local search frameworks: a case study using flow-shop scheduling problems. HM 2014 - 9th International Workshop Hybrid Metaheuristics , Jun 2014, Hamburg, Germany. pp.30-44, ⟨10.1007/978-3-319-07644-7_3⟩. ⟨hal-01094683⟩
86 Consultations
0 Téléchargements

Altmetric

Partager

Gmail Facebook X LinkedIn More