In the era of Big Data, with new and emerging technologies, data become easily attainable for companies. However, acquiring data is only the first step for the company. The second and more important step is to effectively integrate the data through the learning process (mining the data) in the decision-making process, and to utilize the information extracted from data to improve the efficiency of the company’s supply chain operation.The primary focus of this dissertation is on multistage stochastic optimization problemsarising in the context of supply chains and inventory control problems, and on the design ofefficient algorithms to solve the respective models. This dissertation can be categorized intotwo broad areas as follows.The first part of this dissertation focuses on the design of non-parametric learning algorithms for complex inventory systems with censored data. We address two challenging stochastic inventory control models: the periodic-review perishable inventory system and the periodic-review inventory control problem with lost-sales and positive lead times. We assume that the decision maker has no demand distribution information available a priori and can only observe past realized sales (censored demand) information to optimize the system;;s performance on the fly. For each of the problems, we design a learning algorithm that can coverage to the best base-stock policy with tight regret rate. The second part of this dissertation focuses on the design of approximation algorithms for stochastic perishable inventory systems with correlated demand. In this part, we consider the perishable inventory system from the optimization perspective. Different from traditionalperishable inventory literature, we allow demands to be arbitrarily correlated and nonstationary, which means we can capture the seasonality nature of the economy, and allowthe decision makers to effectively incorporate demand forecast. For this problem, we develop two approximation algorithms with worst-case performance guarantees. Through comprehensive numerical experiments, we have shown that the numerical performances of the approximation algorithms are very close to optimal.
【 预 览 】
附件列表
Files
Size
Format
View
Data-Driven Algorithms for Stochastic Supply Chain Systems: Approximation and Online Learning