STOCHASTIC OPTIMIZATION METHODS FOR MULTI-FACILITY CAPACITY INVESTMENT PROBLEMS

Abstract

Strategic capacity planning for multiple-facility systems with flexible designs is critical to engineering systems or supply chains. The difficulties of this problem lie in the multi-dimensional nature of its random variables and its action space. Stochastic optimization methods such as multi-stage stochastic programming and approximate dynamic programming (ADP) have been proposed in the literature to solve this problem. However, these methods can be inefficient when the capacity of the facilities is discrete. Against this background, this dissertation develops 1) a decision-rule based stochastic programming method and 2) a neural network-based ADP to solve a generic multi-facility capacity expansion problem (MCEP) and its variants (i.e. MCEP with fixed costs, MCEP with risk-aversion, and MCEP with capacity contraction). Theoretical results of the problems and the developed algorithms are provided, and the numerical studies herein illustrate the effectiveness of the algorithms.