Progressive selection method for the coupled lot-sizing and cutting-stock problem

Wu, Tao and Akartunali, Kerem and Jans, Raf and Liang, Zhe (2017) Progressive selection method for the coupled lot-sizing and cutting-stock problem. INFORMS Journal on Computing, 29 (3). pp. 523-543. ISSN 1526-5528

[img]
Preview
Text (Wu-etal-IJC-2016-Progressive-selection-method-for-the-coupled-lot-sizing)
Wu_etal_IJC_2016_Progressive_selection_method_for_the_coupled_lot_sizing.pdf
Accepted Author Manuscript
License: Creative Commons Attribution-NonCommercial 4.0 logo

Download (1MB)| Preview

    Abstract

    The coupled lot-sizing and cutting-stock problem has been a challenging and significant problem for industry, and has therefore received sustained research attention. The quality of the solution is a major determinant of cost performance in related production and inventory management systems, and therefore there is intense pressure to develop effective practical solutions. In the literature, a number of heuristics have been proposed for solving the problem. However, the heuristics are limited in obtaining high solution qualities. This paper proposes a new progressive selection algorithm that hybridizes heuristic search and extended reformulation into a single framework. The method has the advantage of generating a strong bound using the extended reformulation, which can provide good guidelines on partitioning and sampling in the heuristic search procedure so as to ensure an efficient solution process. We also analyze per-item and per-period Dantzig-Wolfe decompositions of the problem and present theoretical comparisons. The master problem of the per-period Dantzig-Wolfe decomposition is often degenerate, which results in a tailing-off effect for column generation. We apply a hybridization of Lagrangian relaxation and stabilization techniques to improve the convergence. The discussion is followed by extensive computational tests, where we also perform detailed statistical analyses on various parameters. Comparisons with other methods indicate that our approach is computationally tractable and is able to obtain improved results.