Publication detail

Warm-start Cuts for Generalized Benders Decomposition

KŮDELA, J. POPELA, P.

Original Title

Warm-start Cuts for Generalized Benders Decomposition

English Title

Warm-start Cuts for Generalized Benders Decomposition

Type

journal article in Web of Science

Language

en

Original Abstract

In this paper, we describe a decomposition algorithm suitable for two-stage convex stochastic programs known as Generalized Benders Decomposition. For this algorithm we propose a new reformulation that incorporates a lower bound cut that serves as a warm-start, decreasing the overall computation time. Additionally, we test the performance of the proposed reformulation on two modifications of the algorithm (bunching and multicut) using numerical examples. The numerical part is programmed in MATLAB and uses state-of-the-art conic solvers.

English abstract

In this paper, we describe a decomposition algorithm suitable for two-stage convex stochastic programs known as Generalized Benders Decomposition. For this algorithm we propose a new reformulation that incorporates a lower bound cut that serves as a warm-start, decreasing the overall computation time. Additionally, we test the performance of the proposed reformulation on two modifications of the algorithm (bunching and multicut) using numerical examples. The numerical part is programmed in MATLAB and uses state-of-the-art conic solvers.

Keywords

stochastic programming, Generalized Benders Decomposition, L-shaped method, warm{start

Released

31.12.2017

Publisher

UTIA

Location

Prague

Pages from

1012

Pages to

1025

Pages count

13

BibTex


@article{BUT142260,
  author="Jakub {Kůdela} and Pavel {Popela}",
  title="Warm-start Cuts for Generalized Benders Decomposition",
  annote="In this paper, we describe a decomposition algorithm suitable for two-stage convex stochastic
programs known as Generalized Benders Decomposition. For this algorithm we propose a new
reformulation that incorporates a lower bound cut that serves as a warm-start, decreasing the
overall computation time. Additionally, we test the performance of the proposed reformulation
on two modifications of the algorithm (bunching and multicut) using numerical examples. The
numerical part is programmed in MATLAB and uses state-of-the-art conic solvers.",
  address="UTIA",
  chapter="142260",
  doi="10.14736/kyb-2017-6-1012",
  howpublished="print",
  institution="UTIA",
  number="6",
  volume="53",
  year="2017",
  month="december",
  pages="1012--1025",
  publisher="UTIA",
  type="journal article in Web of Science"
}