Presenter: Xinru Wu

Title: Dynamic Inventory Control with Nonstationary Demands and Random Supply: Application of Reinforcement Learning

Abstract

Most existing stochastic inventory literature are based on the assumption that the product demands are stationary. However, in real business environment, the demands are often non-stationary due to the seasonality and diverse external economics factors. In addition, due to the supply chain disruptions, the product supply is often uncertain. In this paper, we extend the research about dynamic inventory management problems to consider both the non-stationary demands and the stochastic supply. The demand distributions are modeled by a Markovian process. The stochastic supply is captured by a random factor. Thus, the problem is formulated as a stochastic dynamic optimization problem. Given the high computational complexity, the traditional dynamic programming based approximate algorithm become inefficient, especially when solving the large scale problems. We solve the model by the state of art reinforcement learning algorithms. Moreover, we conduct an intensive numerical study to illustrate the efficiency and effectiveness of the algorithms, the value of considering the non-stationary demands and random supply. Finally, multiple managerial insights are summarized.

Main supervisor: Hongyan Jenny Li

Discussant: Lars Relund Nielsen

A PhD seminar is part of the mandatory 1st or 3rd year presentations for PhD students. The time is extended to one hour: 40 minutes for the PhD student, 5-10 minutes for the discussant, and 10-15 minutes for questions. The PhD seminar is hosted by the PhD student’s main supervisor.

Organisers: Surabhi Verma and Hartanto Wong