Matheurísticas para o problema de dimensionamento de lotes com múltiplas plantas

• Ano de defesa: 2015
• Autor(a) principal: Carvalho, Desiree Maldonado [UNIFESP]
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• dARK ID: ark:/48912/001300001dqkf
• Idioma: por
• Instituição de defesa: Universidade Federal de São Paulo (UNIFESP)
• Palavras-chave em Português: lot-sizing problem; multi-plant; lagrangian heuristic; kernel search; problema de dimensionamento de lotes; múltiplas plantas; heurística lagrangiana
• Link de acesso: https://sucupira.capes.gov.br/sucupira/public/consultas/coleta/trabalhoConclusao/viewTrabalhoConclusao.jsf?popup=true&id_trabalho=2400547; https://repositorio.unifesp.br/handle/11600/47204

This Dissertation presents the lot sizing problem with multiple plants (factories) that produce several types of items, where every plant has a single machine. Each machine has setup time and costs, limited capacity and manufactures all types of items. In these plants, the planning horizon is finite and divided into periods. Besides that, each plant has its own demand. The demands of the items in a plant can be met by any plant, without delay, as the problem allows transfers of production lots among the plants and storage of production, both operations subject to costs. This problem, referred here as PDLCLMP, therefore, aims at defining a production planning to minimize the costs of setup, production, storage and transfers between plants, respecting the available resources and attending the demand previously determined in each plant. Due to the limitations found by recently proposed methods in the literature to find feasible solutions to the problem, in this Dissertation we propose matheuristics to solve the PDLCLMP. Three solution methods are presented: a Lagrangian heuristic; a Lagrangian heuristic hybridized with a meta-heuristic path-relinking; and a kernel search method. Besides these instances, others of large scale were generated and employed in experiments performed in this Dissertation. According to these experiments, both pure Lagrangian heuristic and the hybridized Lagrangian version presented an excelent performance when comparing with the best heuristic from literature. Meanwhile the proposed kernel search method, which had as primary goal to keep the solutions quality found by the CPLEX, however in a inferior computacional time, successfully succeded for the instances with small dimensions.