Thesis Title: Managing Operations in Health and Humanitarian Systems Considering Consistency or Uncertainty

Advisor: 

Dr. Pinar Keskinocak, School of Industrial and Systems Engineering, Georgia Tech

Committee members:

Dr. Alejandro Toriello, School of Industrial and Systems Engineering, Georgia Tech

Dr. Chip White, School of Industrial and Systems Engineering, Georgia Tech

Dr. Ozlem Ergun, Department of Mechanical & Industrial Engineering, Northeastern University

Dr. Mathieu Dahan, School of Industrial and Systems Engineering, Georgia Tech

Date: Tuesday, April 13th

Time: 3 pm - 4 pm, EDT

Meeting URL (for BlueJeans): https://bluejeans.com/837081913

Meeting ID (for BlueJeans): 837 081 913

Abstract:

This dissertation focuses on various problems motivated by health and humanitarian systems. When handling the problems, we pay attention to practical aspects of the real applications and consider intangible costs as well as expected and unexpected uncertainties that may arise.

Chapter 2 addresses a home health care scheduling problem (HHCSP) faced by home care agencies. In home health care scheduling, there is a desire to retain consistency with respect to the home health aide servicing each patient; this consistency is referred to as continuity of care. To address this preference for continuity of care, we propose a rolling horizon approach to the scheduling problem and introduce the consistent home health care scheduling problem (Con-HHCSP). We present two different constructive methods to solve HHCSP on a daily basis: an integer programming-based method with approximations and a variant of a petal heuristic. We present adjustments on these methods to address Con-HHCSP, where the goal is to be able to quantify and control the deviation of the new schedule suggested each day from the existing schedule in place, so that some of the existing assignments may be retained in the new schedule that is produced. We discuss the performance and computational efficiency of these methods.

In Chapter 3, we address Endogenous Network Restoration under Uncertainty (ENRU), where the edges are disrupted and the connection between the nodes are lost, and the goal is to restore some of the edges to connect all nodes to the source node. Motivated by post-disaster debris clearance operations, the restoration activities are endogenous such that the current restoration decision depends on the previous actions, and restoration times are stochastic, which are revealed as the network is explored. We analyze the structural properties of this unique variant of a network restoration problem, which are then used to develop an effective heuristic method. We model ENRU as a POMDP, and propose alternative solution methods such as MIP, SMIP, Monte Carlo Search Tree (MCST) based heuristic, and a greedy heuristic. Based on computational experiments, we show that static solutions perform worse than a simple greedy heuristic under uncertainty, and that MCST-based heuristic which utilizes shortest paths in the graph provides a powerful and adaptable framework to handle this type of problems.

Chapter 4 further explores consistency in multi-period decision-making using Newsvendor problem. Motivated by Chapter 2, as well as different research streams in the literature, we consider consistency in the decisions over a time horizon. Consistency in the decisions has some value due to operational restrictions observed in practice, which is often ignored in favor of finding the optimum solution that maximizes the tangible benefits. However, not planning for the infeasibilities that may arise due to inconsistency, or not considering the intangible value of being consistent, would result in suboptimal solutions. In this work, we consider multi-period Newsvendor problem, and present ways of keeping the order quantities consistent over multiple periods. We analyze the impact of consistency on the operation costs, as well as quantified benefits of it in certain use cases.