Models and algorithms for the design of integrated supply chain networks.

摘要

In the past few years, both in industry and in academia, interest in supply chain integration has grown dramatically. In today's highly competitive business environment, only efficient supply chains that integrate decisions in various phases can survive. Two major issues in the efficient design of a supply chain network are inventory management and facility location. The literature on supply chain optimization has traditionally considered facility location strategic decisions and inventory management tactical decisions independently. In this dissertation, we introduce two new models that integrate strategic and tactical decisions in designing supply chain networks. This dissertation consists of following three parts.;In the first part, we develop a multi-echelon joint inventory-location model that considers location and inventory decisions simultaneously. We formulate this problem as a non-convex nonlinear mixed-integer program. We propose a Lagrangian relaxation-based technique for solving this problem when the retailers are identical (same demand and inventory costs).;In the second part, we develop a genetic algorithm with three different encoding schemes for solving our integrated supply chain problem when the retailers have different demand and different inventory costs. We also compare the results obtained by the genetic algorithm using each of the three encoding schemes with the results obtained by the Lagrangian relaxation-based technique.;In the third part, we develop a different joint inventory-location model that incorporates random customer demand. We first propose an approximation for the stochastic demand one-warehouse multi-retailer inventory model. We then combine this approximate inventory model with the uncapacitated fixed charge location model. The resulting problem is formulated as a large-scale 0-1 integer program. For solving this problem, we design three different algorithms: a two-phase Lagrangian relaxation-based algorithm, a one-phase Lagrangian relaxation-based algorithm, and a linear programming-based branch-and-bound algorithm.