演讲人：Dr. Dan Zhang,
Leeds School of Business,
University of Colorado Boulder
 
题目：An Approximate Dynamic Programming Approach to a Rolling-Horizon Appointment Scheduling Problem
 
 
主持人：庄伟芬 副教授

时间：2015年5月6日 星期三  下午15:00

地点：嘉庚一409
 
Abstract
Appointment scheduling problems typically involve multiple customer classes with varying resource requirements and priorities. Such problems are often formulated as infinite horizon discounted Markov decision processes (MDPs) where the objectives are either revenue maximization or cost minimization. Due to the rolling horizon nature and the existence of multiple customer classes, MDP formulations suffer from the well-known curse of dimensionality, and are not amenable to exact solutions via standard solution methods. We consider two solution strategies in this work. First, we show that an affine functional approximation based on linear programming based approximate dynamic programming admits a compact representation and therefore can be solved very quickly, leading to a reasonable control policy. Our results generalize some recent work in this area. Second, we exploit the rolling-horizon structure of the problem by showing that a relaxed version of the problem is equivalent to a finite-horizon MDP. We study the structure of the finite horizon MDP and show that it is easily solvable. Unlike the affine approximation, the solution from the finite-horizon MDP captures the nonlinearity of the value function. A numerical study shows that the policy from the finite-horizon MDP improves upon the policy from affine approximation and is a promising candidate for practical applications. (This is joint work with Thomas Vossen at University of Colorado Boulder.)
 
Biography
Dan Zhang is an assistant professor in the Management and Entrepreneurship division at Leeds School of Business, University of Colorado Boulder. Prior to Leeds, he was an assistant professor of operations management at Desautels Faculty of Management, McGill University in Montreal, Canada. Dr. Zhang received his PhD in industrial engineering from University of Minnesota and subsequently did postdoctoral work at Booth School of Business, University of Chicago. His current primary research interest is revenue management and pricing. In the last few years, he worked on approximate dynamic programming methods for large-scale dynamic optimization problems and on consumer behavior models with applications to revenue management and pricing. Dr. Zhang is a frequent reviewer for academic journals, conferences, and grant agencies. For his review work, he recently received meritorious service awards from Management Science and Manufacturing and Service Operations. Dr. Zhang currently serves on the editorial board of Journal of Revenue and Pricing Management, and is a Senior Editor for the journal Production and Operations Management. Dr. Zhang teaches in the area of operations management and data analytics.