Informs Annual Meeting 2017

MD23

INFORMS Houston – 2017

MD23

2 - Naloxone Procurement and Distribution in the Presence of High Drug Prices Yijia Wang, University of Pittsburgh, Pittsburgh, PA, United States, YIW94@pitt.edu, Daniel Jiang Naloxone, a drug for treating opioid overdose, is vital for public health agencies’ efforts to combat the opioid epidemic. However, naloxone prices have recently surged so it is essential for public health agencies to effectively manage procurement and distribution of naloxone. The goal is to maximize naloxone distribution into affected communities under funding and storage constraints; this problem will be solved through a novel formulation of a hierarchical MDP along with an associated approximate dynamic programming technique. The main policy question we will answer is: under optimal operations, what is the effect of increasing prices on our ability to distribute Naloxone kits? 3 - An Origin-Destination Decomposition Method for Network Revenue Management Rui Zhang, University of Colorado Boulder, 5377 Racegate Run, Boulder, CO, 21045, United States, rui.zhang@colorado.edu, Dan Zhang In this paper, in contrast to resource-based decomposition method, we consider a decomposition method where the approximate linear programming of the network revenue management problem is decomposed into a collection of subproblems, each corresponding to an origin-destination pair, which we call the OD- based decomposition. We show that OD-based decomposition provides an upper bound to the total expected revenue and alternative heuristic control policies. We identify conditions under which the OD-based decomposition gives tighter upper bounds and better heuristic policies than the resource-based decomposition. 4 - An Approximate Dynamic Programming Approach to a Rolling- horizon Appointment Scheduling Problem Fan You, University of Colorado-Boulder, 995 Regent Dr, Boulder, CO, 80302, United States, fayo4612@colorado.edu, Thomas Vossen, Dan Zhang We consider a rolling-horizon appointment scheduling problem with multiple patient classes. The problem is formulated as an infinite horizon discounted cost Markov decision process. We consider affine and finite-horizon approximations and show that they admit compact representations and can be efficiently solved as small scale linear programs. A numerical study illustrates the performance of the heuristic control policies based on the approximations. 350A Power Generation Scheduling Invited: Energy Systems Management Invited Session Chair: Kai Pan, University of Florida, Gainesville, FL, 32611, United States, kpan@ufl.edu 1 - A Novel Matching Formulation for Startup Costs in Unit Commitment Bernard Knueven, University of Tennessee, 525K.Tickle Building, Knoxville, TN, 37996, United States, bknueven@vols.utk.edu, Jim Ostrowski, Jean-Paul Watson We present a novel formulation for startup cost computation in unit commitment (UC). Both the proposed formulation and existing formulations in the literature are placed in dominance hierarchy based on their respective linear programming relaxations. The proposed formulation is tested empirically against existing formulations on large-scale unit commitment instances drawn from real-world data. Our proposed formulation is empirically demonstrated to be as tight as a perfect formulation for startup costs. This tightening reduces the computational burden in comparison to existing formulations, especially for UC instances with large variability in net-load due to renewables production. 2 - Data-driven Chance-constrained Stochastic Unit Commitment under Wind Power Uncertainty Ali Bagheri, Oklahoma State University, 320 E McElroy Road, Apt 508, Stillwater, OK, 74075, United States, ali.bagheri@okstate.edu Rapid integration of cheap, clean but highly intermittent wind energy into power systems brings challenges to ISOs to maintain the system reliability. Stochastic programs may result in biased and unreliable unit commitment (UC) and economic dispatch (ED) decisions by fixing the probability distribution of wind output. Robust optimization (RO) approaches sacrifice system’s cost-effectiveness in exchange of reliable UC and ED schedules. We develop a data-driven chance- constrained stochastic UC model, where the data-driven chance constraint limits the worst-case chance of load imbalance to be no more than a specified tolerance, by taking advantage of historical data. MD25

342E Revenue Management and Control of Operational Problems Sponsored: Revenue Management & Pricing Sponsored Session

Chair: Arnab Bisi, Johns Hopkins Carey Business School, 100 International Drive, Baltimore, MD, 21202, United States, abisi1@jhu.edu 1 - Healthcare Payment Models under Competition Zheng Han, University of Kansas, Lawrence, KS, United States, hanzheng@ku.edu, Mazhar Arikan, Suman Mallik We consider two hospitals competing as profit-maximizers by choosing their own quality levels under different payment schemes (i.e., bundled payment (BP) vs. fee-for-service (FFS)). A game-theoretic model is built to analyze equilibrium outcomes. We show that: BP (FFS) is not always associated with high (low) equilibrium quality, and identify threshold payments for both FFS and BP that can lead the corresponding hospital to choose the highest or lowest level of quality; a higher reimbursement for hospital readmissions under FFS may result in quality improvement, rather than deterioration; a higher reimbursement from either FFS or BP does not necessarily result in a higher beneficiary welfare. 2 - Dynamic Pricing or Dynamic Logistics? Thunyarat (Bam) Amornpetchkul, NIDA Business School, Bangkok, Thailand, thunyarat.a@nida.ac.th Thunyarat (Bam) Amornpetchkul, Boston College (Visiting), Chestnut Hill, MA, United States, thunyarat.a@nida.ac.th, Hyun-Soo Ahn, Ozge Sahin Retailers who sell their products through multiple channels often hold inventory at separate locations, dedicated for demand arriving at each channel. When inventory level does not match with demand, transshipment across channels may be beneficial. We characterize and compare the effectiveness of the retailer’s optimal pricing and transshipment policy. Furthermore, we investigate how the optimal pricing policy is affected by the transshipment policy, and vice versa. 3 - An Inventory System with Partial Backorders and Inventory Holdback Arnab Bisi, Johns Hopkins Carey Business School, 100 International Drive, Baltimore, MD, 21202, United States, abisi1@jhu.edu, Yanyi Xu, Maqbool Dada We consider a periodic-review base stock inventory system with partial backorders. In each period, an order of q is placed for delivery one period later and as much as possible of the realized demand is filled immediately from inventory on-hand. Of the excess demand above the inventory on-hand, up to q - k units are backordered to be filled from future orders. The on-order quantity k denotes the reservation quantity held back for use in subsequent periods. Our model yields the full lost sales and full backorder models as special cases. For both finite and infinite horizon problems, we determine structural properties of the optimal policy and discuss how to find the optimal base stock and hold back quantity. 342F Topics in Approximate Dynamic Programming Sponsored: Revenue Management & Pricing Sponsored Session Chair: Dan Zhang, University of Colorado, Boulder, CO, 80309, United States, dan.zhang@colorado.edu Co-Chair: Rui Zhang, University of Colorado Boulder, Columbia, MD, 21045, United States, rui.zhang@colorado.edu 1 - Index Policies and Performance Bounds for Dynamic Selection Problems James E.Smith, Duke University, Fuqua School of Business, Box 90120, Durham, NC, 27708-0120, United States, jes9@duke.edu, David Brown In this work, we study heuristics and performance bounds for dynamic selection problems. Examples of dynamic selection problems include assortment planning with demand learning, design of clinical trials, and screening a pool of applicants (e.g., for admission to colleges). We study a number of index-based policies, including one based on a Lagrangian relaxation of the problem. We characterize the performance of this Lagrangian index policy compared to an optimal policy. We also develop performance bounds that combine Lagrangian relaxations with information relaxations and show that these new performance bounds outperform standard Lagrangian relaxation bounds. MD24

244