Lagrangian cuts generated by batch to efficiently solve two-stage stochastic mixed-integer program

Fuente: arXiv
Enregistré dans:
Détails bibliographiques
Auteurs principaux: Xiaoyu, Luo, Chuanhou, Gao
Format: Preprint
Publié: 2024
Sujets:
Accès en ligne:
Tags: Ajouter un tag
Pas de tags, Soyez le premier à ajouter un tag!
_version_ 1866913213341237248
author Xiaoyu, Luo
Chuanhou, Gao
author_facet Xiaoyu, Luo
Chuanhou, Gao
contents We propose to generate Lagrangian cut for two-stage stochastic integer program by batch, in contrast to the existing methods which solve each Lagrangian subproblem at every iteration. We establish two convergence properties of the proposed algorithm. Then we demonstrate that the improvement in the lower bound achieved by incorporating the Lagrangian cut adheres to the `triangle inequality', thereby showcasing the superiority of our proposed method over existing approaches. Moreover, we suggest acquiring Lagrangian cuts for unresolved scenarios by averaging the coefficients of the acquired Lagrangian cuts, ensuring the quality of this cut with a certain probability. Computational study demonstrates that our proposed algorithm can significantly improve the lower bound of the linear relaxation of the Bender master problem more quickly with much fewer Lagrangian cuts.
format Preprint
id arxiv_https___arxiv_org_abs_2401_15901
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle Lagrangian cuts generated by batch to efficiently solve two-stage stochastic mixed-integer program
Xiaoyu, Luo
Chuanhou, Gao
Optimization and Control
90C06, 90C11, 90C15
We propose to generate Lagrangian cut for two-stage stochastic integer program by batch, in contrast to the existing methods which solve each Lagrangian subproblem at every iteration. We establish two convergence properties of the proposed algorithm. Then we demonstrate that the improvement in the lower bound achieved by incorporating the Lagrangian cut adheres to the `triangle inequality', thereby showcasing the superiority of our proposed method over existing approaches. Moreover, we suggest acquiring Lagrangian cuts for unresolved scenarios by averaging the coefficients of the acquired Lagrangian cuts, ensuring the quality of this cut with a certain probability. Computational study demonstrates that our proposed algorithm can significantly improve the lower bound of the linear relaxation of the Bender master problem more quickly with much fewer Lagrangian cuts.
title Lagrangian cuts generated by batch to efficiently solve two-stage stochastic mixed-integer program
topic Optimization and Control
90C06, 90C11, 90C15
url https://arxiv.org/abs/2401.15901