Research

year 2005 
author Sungwon Jung 
Keyword Supply Chain Management, Master Planning, Memetic Algorithm, Multi-Level Lot-Sizing, Multi-Level Capacitated Lot-Sizing 
Abstract Nowadays, the rapid development of information technology has created the environment in which production plants in supply chain can cooperate to make a production plan for the master product of each facility units in the aspect of the supply chain. This type of multi facility production plans can be viewed as a multi stage production planning which is well known and documented in literature.


Multi stage production planning is known as a so-called NP hard problem which can be solved optimally in a reasonable CPU time only when the problem instance is very small. Hence, many researchers have focused on the heuristic methods to address large scale problems of this kind. Previous studies on multi stage production planning can be divided into two types: the multi-level lot sizing (MLLS) which considered only cost optimization and the multi-level capacitated lot sizing (MLCLS) which considered capacity constraints as well as cost optimization. MLLS is used when it is necessary to make a production plan to meet a given demand in a relatively short time, with a sufficient extra production capacity. MLCLS aims to develop a production plan taking into account production capacity limits.


In this thesis, the author identifies limitations of the algorithms of previous studies for the multi stage production planning and proposes the cost-effective heuristics to provide high quality solutions for MLLS and MLCSL. The proposed algorithms are developed based on the memetic algorithms which are population-based heuristic search approaches for optimization problems. The most crucial and distinctive feature of the algorithm for MLLS proposed in this study is the incorporation of local refinement procedure based on the concept of benchmark. The proposed local refinement procedure allows recovering any temporary degradation of the solution in the operation of a population. In the case of the algorithm for MLCLS, we contrive the solution representation scheme which guarantees to generate feasible solution with regard to production capacity constraints.


The benchmark-based local-refinement procedure proposed by this study will be well applicable to other problems which are otherwise difficult to refine their solution. In addition, the solution representation considering the capacity constraint will be a good foundation for the development of algorithms for optimizing the more complex supply chain problems. 
c PhD 

Download :