Informs Annual Meeting Phoenix 2018

INFORMS Phoenix – 2018

SA78

Sunday, 10:00AM - 10:50AM

3 - The Facts on the Ground: Using Simulation to Understand Policies in Humanitarian Fleet Management Liyi Gu, University of Maryland, Van Munching Hall, College Park, MD, 20742, United States, Ilya O. Ryzhov, Mahyar Eftekhar Humanitarian field managers are known to lack systematic guidance in decision making and often rely on their intuition. In this work we develop empirical and stochastic models of humanitarian fleet management, on top of which we build a simulator that evaluates intuitive policies that may be used by field managers. We empirically estimate demands (mission distances) and vehicle operational costs using data from a large international humanitarian organization, and evaluate the acquisition, assignment, and disposition of vehicles in field operations. The simulation results provide insight to why managers make decisions the way they do, and could lead to improved policy recommendations. 4 - Scale vs. Impact. Resource Allocation Strategies for Family Planning Outreach Harwin de Vries, INSEAD, Boulevard de Constance, Fontainebleau, 77210, France, Lisanne Van Rijn, Kim Leliveld, Luk N. Van Wassenhove We study an organization deploying over 500 mobile outreach teams to bring family planning services to hard to reach communities. Major objectives are to maximize utilization (scale) and to reach target populations like young and poor people (impact). We discuss how operations affect scale and impact, trade-offs, models, and implications. n SA78 West Bldg 213B Joint Session TSL-FAC/SOLA: Random Stow Strategies in Warehousing Sponsored: Location Analysis Sponsored Session Chair: John Gunnar Carlsson, University of Southern California, Los Angeles, 90089, United States 1 - A Velocity-based Stowage Policy for a Semi-automated Fulfillment System Stephen C. Graves, Massachusetts Institute of Technology, E62- 579, 77 Massachusetts Avenue, Cambridge, MA, 02139-4307, United States, Amy Liu Online retail fulfillment is increasingly performed by semi-automated systems in which inventory is stored in mobile pods that are moved by robotic drives. The stowage decision decides on what pods to store what inventory. We examine the impact of velocity-based stowage policies on the operational performance of the fulfillment system. In particular we model a stowage policy in which received units are categorized as being either fast or slow, and then characterize the impact of velocity-based stowage on the travel distance for the robotic drives. We find that such policies can substantially reduce the travel distance, which reduces the number of drives required for a given system throughput level. 2 - Man vs Machine in Warehouse Picking Gerard P. Cachon, University of Pennsylvania, 3730 Walnut St, Philadelphia, PA, 19104, United States, Omar Besbes We compare warehouse pick operations performed the traditional way (with a human) against picking with robots (e.g., Kiva systems). We are interested in how these different systems scale and under what conditions one outperforms the other. 3 - Random Stow and the Generalized Travelling Salesman Problem John Gunnar Carlsson, University of Southern California, 3750 McClintock Avenue, Los Angeles, CA, 90089, United States The generalized travelling salesman problem is a variation of the traditional TSP in which one is given a collection of sets of points and one seeks a tour of minimal length that visits one member each set. The GTSP is fundamentally important in studying randomized strategies in warehouses, in which one stores a stock keeping unit (SKU) in any available location (as opposed to designating specific regions of the warehouse for different SKUs). We derive asymptotic bounds for the length of a GTSP tour under various assumptions on the magnitude of demand and its distribution. 4 - Fast Solutions for the Dynamic Item Stocking Problem in Amazon Class Online Order Fulfillment Warehouses Sanchoy Das, Professor of Industrial Engineering, New Jersey Institute of Technology, University Heights, Newark, NJ, 07102, United States, Jingran Zhang, Sevilay Onal Amazon class online order fulfillment warehouses use an explosive storage policy, whereby the same item is stocked simultaneously in many bin locations anywhere in the warehouse. The stocking objective is to disperse incoming items through the warehouse to maximize the probability a time-fenced group of future customer orders can be fulfilled from a set of closely located bins (efficient pick list). Demand intelligence described by time-fenced item correlations of incoming orders is a key data input. The problem is dynamic in that both the inventory state and pending order list are temporal.

n Plenary West Bldg 301AB Plenary: Riding Technology Waves: Perspectives and Opportunities for Operations Research Plenary Session Chair: Young-Jun Son, University of Arizona, Systems and Industrial Engineering, Engineering Building #20 Room 111, Tucson, AZ, 85721- 0020, United States 1 - Riding Technology Waves: Perspectives and Opportunities for Operations Research Brenda Dietrich, Cornell University, Ithaca, NY, United States This talk begins with a fly-by of almost six decades of information technology beginning with its use to automate business processes and extending to its current role in consumer self-service, the internet of things, and in intermediating social processes. The resulting “data exhaust together with the availability of low cost computing capacity spawned the age of analytics, the rise of big data, the birth of cognitive computing and the reinvigoration of artificial intelligence. The past, current and potential role of analytic methods in these technology waves will be discussed, with focus on the opportunity to use analytics and automation to create new data. Areas in need of further study by the Operations Research community will be highlighted. n SB01 North Bldg 121A Methods for Multi-Stage Stochastic Optimization Sponsored: Optimization/Optimization Under Uncertainty Sponsored Session Chair: Selvaprabu Nadarajah, College of Business, University of Illinois at Chicago, Chicago, IL, 60607, United States 1 - Basis Function Selection in Approximate Linear Programming Parshan Pakiman, University of Illinois at Chicago, Chicago, IL, United States, Selvaprabu Nadarajah, Negar Soheili, Qihan Lin Approximate linear programs (ALPs) compute a value function approximation (VFA) for Markov decision processes (MDPs) using a linear combination of basis functions. The resulting VFA provides policies and lower and upper bounds on the optimal policy value. Basis functions - typically chosen in a problem-specific manner - affect the quality of these bounds. Constructing VFAs is thus challenging to a non-expert and may not guarantee tight bounds. To ease the use of ALPs, we develop a version using basis functions sampled from a parametric function class, which asymptotically leads to a near-optimal VFA. We establish finite sample guarantees and discuss numerical results on challenging applications. 2 - Computational Experience with Asynchronous Projective Hedging David L. Woodruff, University of California-Davis, Graduate School of Management, One Shields Avenue, Davis, CA, 95616, United States, Jean-Paul Watson, Jonathan Eckstein Recent work by Eckstein and Combettes resulted in development of an algorithm for multi-stage, convex optimization problems with uncertain input data expressed as a set of scenarios. The algorithm is called Asynchronous Projective Hedging (APH). In this talk we describe computational experience with this algorithm primarily based on two well-known problems from the stochastic programming literature as well as experience with large mixed-integer problems. We explore various tradeoffs such as computational resources vs. wall-clock vs. solution quality as well primal versus dual solution quality. 3 - Solving Multi-period Mine Planning Models with Endogenous Uncertainty Tito Homem-de-Mello, Universidad Adolfo Ib á ñez, Santiago, Chile, Denis R. Saure, Tomas Lagos, Guido Lagos, Margaret Armstrong We study a production scheduling problem in open pit mining under ore-grade uncertainty. We consider a multi-stage version of the problem, where there is a planning horizon of several time periods and where at each period the decision is: which blocks to extract and which to process, while satisfying extraction and processing capacity constraints. We propose an optimization with learning approach intended to tackle two of the main challenges presented by this problem: the scalability of the problem due to its large dimensions, and the robustness to uncertainty due to the inherent volatility of some of the parameters of the problem. We report computational experiments on a real-sized mine. Sunday, 11:00AM - 12:30PM

30