Industrial strength theory and algorithms for discrete optimization via simulation and the impact of supply chain structure on product line architecture.

摘要

Stochastic simulation is an important tool to model many real-world systems. In many applications, the decision maker wants to use the simulation model to help optimize the design of the system by adjusting the levels of design variables to search for the best system performance, as predicted by simulation output. Discrete Optimization via Simulation research studies systematic ways to achieve this goal when design variables are discrete. Algorithms proposed by academic researchers focus on convergence guarantees and are often too inefficient to solve real-world problems. Commercial software products are based on metaheuristics designed for deterministic optimization problems and largely ignore the stochastic nature of the problem.;In this dissertation, we contribute to the area of Discrete Optimization via Simulation by proposing a framework for designing locally convergent algorithms and implementing a software package to bridge the gap between academic research and industrial practices. We further propose a locally convergent algorithm for solving high-dimensional Discrete Optimization via Simulation problems. Numerical experiments demonstrate that our algorithms have finite-time performance comparable to a leading commercial product and at the same time deliver statistical convergence guarantees that commercial products lack.;The second part of this dissertation looks into the impact of supply chain structure on product line architecture. We examine product line proliferation strategies for firms with different market conditions and supply chain structures in a modular design environment, where a product is made up of a platform and a component. Using a model that combines a nested logit representation of demand with a macro level depiction of supply chain cost structure, we show that a Make-To-Order (MTO) system makes platforms and components complementary, while they may be substitutes in a Make-To-Stock (MTS) system. Furthermore, an MTO system favors a fatter product line with a proliferation of component options, while an MTS system favors a leaner product line with streamlined component options but possibly more platforms. Finally, when a firm is not able to implement the optimal product line architecture due to legacy considerations or uncertainty, we find that, contrary to common belief, an aggressive approach with more platforms may be more effective than a conservative approach with fewer platforms.