Informs Annual Meeting 2017

WB60

INFORMS Houston – 2017

4 - Effective Methods for Solving the Bicriteria Pcenter Pdispersion Problem Golbarg Kazemi Tutunchi, Senior Operations Research Specialist,

edited English-language sources, such as news wire articles. We demonstrate the application of rule-based processes and open-source Natural Language Processing libraries on a domain-specific corpus in a closed-loop system. This process standardizes references to common aviation terms such as SIDs and STARs, abbreviations, misspellings, and key phrase identification.

SAS.institute Inc., 100 SAS.campus Dr., Cary, NC, 27513, United States, golbarg.tutunchi@sas.com, Yahya Fathi

We propose an IP model for the bi-criteria p-center and p-dispersion problems via the set covering and the node packing problems. We further study the structure of the proposed model, and we introduce a new family of valid inequalities to solve this IP model more effectively. Through a limited computational experiment we

WB60

370A Forecasting Contributed Session Chair: Fabian Taigel, University of Wuerzburg, Wuerzburg, Germany, fabian.taigel@uni-wuerzburg.de 1 - Spare Parts Planning under Uncertain Demand Rates Erwin van Wingerden, PhD Student, Eindhoven University of Technology, Eindhoven, Netherlands, e.v.wingerden@tue.nl, Tarkan Tan, Geert-Jan Van Houtum In this paper we consider a multi-item inventory problem with a backorder constraint. Demand follows a Poisson distribution but instead of assuming the demand rate is given, as is commonly done, we allow for uncertainty in the demand rate. When dealing with new products and a lack of data, the demand rate commonly based on estimates and uncertain. First we show that uncertainty of the demand rate is similar to an uncertain lead time. Next, we show that base stock levels and costs are always higher in the case when there is more uncertainty regarding the demand. Finally, we present numerical results where we consider different distributions for the demand rate among which uniformly distributed demand rates. 2 - Event Forecasting in Call Centers Ger M. Koole, Vrije Universiteit Amsterdam, Department of Mathematics, De Boelelaan 1081A, Amsterdam, 1081 HV, Netherlands, ger.koole@vu.nl In this talk we will present recent advances in call center forecasting, especially those related to the forecasting of events such as holidays, billions, and marketing actions. We show how this leads to highly accurate and accountable forecasts. 3 - On Estimation of Cross Elasticity of Demand to Forecast Sparse Time-series Data for E-commerce Applications Arash Pourhabib, Walmart Labs, 10 Pacific Bay Cir, Apt 204, San Bruno, CA, 94066, United States, apourhabib@walmartlabs.com, Antony Joseph, John Bowman, Jagtej S. Bewli Cross elasticity characterizes the change in demand for an item as a result of a change in price, or any other features, for another item. Accurate demand forecast hinges upon accurate estimation of cross elasticity, which is a challenging task in eCommerce operations, mainly due to the existence of a large number of low-selling items. We propose a data-driven method to estimate cross elasticity by taking into account the sparse structure of the eCommerce data. We use an error correcting model approximation to a system of differential equations. We demonstrate that such an approach improves the accuracy of demand forecast. 4 - Hierarchical Forecasting and Multiple Aggregation Prediction Algorithm Combined Intermittent Demand Forecasting and Inventory Control This paper reviews the intermittent demand forecasting methods and compares hierarchical forecasting methods and multiple aggregation prediction algorithm forecasting methods and then proposes a new forecasting method by combining these two methods. We use the procurement data of the State Grid Corporation of China to study these demand forecasting methods. In terms of forecasting performance, we use two forecasting error measurements as well as a real data based inventory simulation experiment. The results of the case study indicate that forecasting methods will generate smaller forecasting error and are better than the previous forecasting methods in terms of inventory performance. 5 - Prescriptive Analytics in Practice - Integrating Machine Learning with Inventory Management for a Fast-casual Restaurant Fabian Taigel, University of Wuerzburg, Wuerzburg, Germany, fabian.taigel@uni-wuerzburg.de, Jan Meller We propose a new, integrated approach for data-driven inventory management by combining tree-based machine learning techniques with the well-known newsvendor logic. We compare our model with various competing methods from the literature and benchmark it in a real -world application for a restaurant chain. We obtain very promising first results and provide relevant insights both for practice and theory, as we are able to identify fundamental drivers of the model’s performance and apply our approach to a real world data set. Li Shalang, Peking University, Beijing, 100078, China, 1301110862@pku.edu.cn, Lei Ming, G.Keong Leong, Honghui Deng

demonstrate the effectiveness of the proposed valid inequalities. 5 - A Manufacturing Network Design and Contingency Planning Model

Tan C. Miller, Director of Global Supply Chain Management Program, Rider University, 12 Winding Way, Morris Plains, NJ, 07950, United States, tanjean@verizon.net, Renato De Matta We enhance a traditional network facility location manufacturing and distribution model to provide extensive contingency and scenario planning capabilities. Specifically, we develop and test a combination strategic network design model that in addition to solving for the optimal profit maximizing locations, also facilitates trade-off analyses between different levels and costs of manufacturing capacity investments and inventory investments.

WB59

362F Aviation Applications Contributed Session

Chair: Dan Larsen, University of Cincinnati, 1807 Sycamore Valley Drive, Apartment T2, Reston, VA, 20190-4561, United States, asel.dan@msn.com 1 - A Novel Approach to Model and Solve Discrete Truss Sizing Problems Mohammad Shahabsafa, Lehigh University, 200 West Packer Ave,, Bethlehem, PA, 18015, United States, mos313@lehigh.edu, Tamas Terlaky, Ali Mohammad-Nezhad, Luis F.Zuluaga, Sicheng He, Joaquim Martins We consider various mathematical optimization models for the truss topology design problem with the objective to minimize the weight of the structure. The non-convex Euler bucking constraints and Hooks law are also considered. We are interested in truss design problems when the cross sectional areas of the bars take only discrete values. We propose Mixed Integer Linear Optimization Model (MILO) reformulation of the non-convex models. The resulting MILO models are not solvable with existing MILO softwares. We propose a solution methodology to solve the MILO models, and present numerical experiments to demonstrate the performance of the methodology. 2 - Capacity Modeling of Mixed-mode Runway Operations Rajesh Piplani, Associate Professor, Nanyang Technological University, School of MAE, 50 Nanyang Avenue N3-2C-84, Singapore, 639798, Singapore, mrpiplani@ntu.edu.sg, Wai Lun Cheung Capacity estimation of airports (defined by runway capacity) is important in planning to meet the long-term growth of air traffic as seen by most regions around the world. In this research, we develop analytical models for capacity estimation (maximum throughput) of a runway operating in mixed-mode. Existing models only consider certain arrival-departure sequences whereas our models take into account optimized arrival departure sequences developed as part of a sister project. We plan to extend the models to two or more runways arranged in different configurations, and also plan to study the relationship between delay and realized throughput, which in turn is affected by traffic mix. 3 - Deliver or Not?: Revenue Management for Future Drone-based Delivery Services Heng Chen, University of Nebraska-Lincoln, CBA 250, 1240 R.Street, Lincoln, NE, 68588, United States, heng@unl.edu, Senay Solak Drones are expected to fulfill future commercial delivery services for retailers and courier companies. We study some revenue management problems for these companies to determine when the drone-based delivery options should be enabled under stochastic demand. A dynamic programming model is developed. Both analytical and numerical analyses are provided. 4 - Natural Language Processing Dan Larsen, The MITRE Corpororation, 7525 Colshire Dr, MS.N329, McLean, VA, 22102, United States, dalarsen@mitre.org Free-form, natural language aviation safety reporting narratives are an integral part of aviation safety analysis. However, it is difficult to apply machine learning algorithms to domain-specific natural language data using libraries trained on

495