Informs Annual Meeting 2017

SA57

INFORMS Houston – 2017

SA58

4 - Modeling Global Facility Decisions for a Supply Chain Coordinator Ram Gopalan, Rutgers, The State University of New Jersey, 227 Penn Street, Camden, NJ, 08102, United States, ram.gopalan@gmail.com, Snehamay Banerjee, Damodar Golhar A Supply Chain Coordinator (SCC) is an intermediary between raw material suppliers, contract manufacturers and retailers. The SCC does not own plant capacity or raw material, but performs a disintermediating role in supply chain finance by orchestrating the contract manufacturing of customized products for various globally dispersed retailers. The SCC exploits differences in manufacturing costs at various candidate facilities, but also bears the operational risk if demand at end retailers is significantly different from forecasts. This research addresses a facility choice and capacity selection (FC-CS) problem for a SCC who is customizing a product with pronounced counter-seasonal demand. 362D Advances in Simulation Optimization and its Applications Sponsored: Simulation Sponsored Session Chair: Loo Hay Lee, National University of Singapore, Singapore, 119260, Singapore, iseleelh@nus.edu.sg 1 - Improving Ordinal Transformation by Combining Multiple Low-fidelity Models Si Zhang, Shanghai University, Shang-Da Road No. 599, Shanghai, 200444, China, zhangsi817@sina.com, Chun-Hung Chen, Jie Xu, Edward Huang Simulation optimization problems are usually large-scaled and reqire a very high computing cost to solve them. Previous works on a new framework known as ordinal transformation (OT) provides a method that makes use of a low-fidelity approximate model to speed up optimization. In this work we study how to improve the quality of using low-fidelity models when there are two or more low-fidelity models. 2 - Efficient Nested Simulation for Risk Management of Variable Annuities Ben Feng, University of Waterloo, Waterloo, ON, Canada, ben.feng@uwaterloo.ca Variable annuities are popular insurance products that enable policyholders to enjoy both insurance protections and investment returns.Dynamic hedging for variable annuities requires computationally intensive nested Monte Carlo simulation experiments. We propose, analyze, and examine efficient simulation procedures to estimate the expected shortfall of the hedging errors for different variable annuities.Our procedure first infers the ranking of hedging losses from a baseline simulation model then concentrate the simulation budget on the inferred “tail scenarios”. As shown by our numerical experiments, such concentration of simulation effort greatly improves efficiency. 3 - Machine Learning and Sample Path Analysis for Large Scale Ranking and Selection Giulia Pedrielli, Arizona State University, Tempe, AZ, 85281, United States, giulia.pedrielli.85@gmail.com, John W. Fowler The advancement of machine learning methods as well as computing capabilities has opened possibilities to solve discrete optimization problems over large sets. We propose an algorithm that uses unsupervised learning to support ranking and selection when a large number of solutions are present. Paired comparisons are used to provide a three level scoring to individual solutions for clustering and then optimally allocate computational budget to the solutions for estimation. The comparison uses sample path information relying on memory availability on the machine, while trying to minimize the amount of new observations required. 4 - Exploring an Input Parameter Space with a Limited Simulation Budget Xilun Chen, Arizona State University, 600 S Dobson Road, Unit 110, Mesa, AZ, 85202, United States, Xilun.Chen@asu.edu, K. Selcuk Candan Complex simulators, such as for predicting epidemics, require a large number input parameters and generate a large space of observations. Consequently, simulation ensembles are expensive to generate and difficult to explore. In practice, given a simulation budget, one needs to identify an ensemble that includes the most informative simulations to help explore the simulation space. Recognizing this challenge, we propose a novel SimExplorer method that uses partial observations to guide the selection process for new simulations to execute. We show that this process can effectively identify new interesting simulations to run, while balancing the overall fit against the partial observation. SA57

362E Energy and Climate Themes in Transportation Sponsored: Transportation Science & Logistics Sponsored Session Chair: Benjamin Leibowicz, The University of Texas-Austin, Graduate Program in Operations Research and Indust, Austin, TX, United States, bleibowicz@utexas.edu 1 - Hydrogen Supply Chain Analysis for Fuel Cell Vehicles using Wind Energy in Texas Kazunori Nagasawa, University of Texas-Austin, Austin, TX, United States, nagasawa@utexas.edu, F. Todd Davidson, Michael E. Webber This work will develop methods to help private and public stakeholders better assess the opportunity for manufacturing, transmitting, and distributing renewable hydrogen, and will use Texas as a case study. This work will include identifying potential locations for hydrogen production via electrolysis and wind power, while accounting for the spatial variance in future consumption centers, and the trade-offs in different infrastructure solutions. 2 - Shared Autonomous Electric Vehicle (SAEV) Operations and Emissions Savings Krishna Gurumurthy, University of Texas-Austin, Stop C1700, Austin, TX, 78712, United States, Kara Kockelman Shared Autonomous Vehicles (SAVs) will offer new travel options, with smaller vehicles and better engine use delivering lower emissions. Use of electric vehicles can further reduce energy demands and human exposure to pollution. These studies simulate SAV and SAEV vehicle use around Austin for emissions and cost estimates. One uses different battery-range, fleet-size and charging-speed scenarios to assess fleet profitability and customer service times. While a gasoline hybrid-electric (HEV) SAV fleet typically performs better than EV fleets, SAEVs come close with longer ranges and fast-charging. Higher gas prices, carbon taxes and environmental policies may stimulate their demand. 3 - Policy Recommendations for a Transition to Sustainable Mobility Based on Historical Diffusion Dynamics of Transport Systems Benjamin D.Leibowicz, University of Texas-Austin, ETC 5.128D, 204 E. Dean Keeton St. C2200, Austin, TX, 78712-1591, United States, bleibowicz@utexas.edu This study analyzes historical data on the diffusion dynamics of transport systems in the United States. The methodology systematically compares the relative timing of diffusion processes for infrastructure, vehicles, and travel. A striking regularity observed across transport systems is that the diffusion of infrastructure precedes the adoption of vehicles, which precedes the expansion of travel. On the “chicken-and-egg conundrum” of infrastructure provision, findings thus support the view that infrastructure comes first. 362F Airport Capacity Management Sponsored: Aviation Applications Sponsored Session Chair: Konstantinos G. Zografos, Lancaster University, Lancaster, LA1 4YX, United Kingdom, k.zografos@lancaster.ac.uk 1 - Airport-centric vs Airline-centric Approaches in Demand Management Alexandre Jacquillat, Carnegie Mellon University, 5000 Forbes Avenue, Office 2118J, Pittsburgh, PA, 15213, United States, ajacquil@andrew.cmu.edu, Vikrant Vaze Airport scheduling interventions can mitigate, or control, congestion through limited changes in flight schedules. However, significant flexibility exists regarding the design of the underlying schedule optimization. The outcomes of centralized (airport-centric) approaches may not coincide with the distributed (airline- centric) objectives. This talk presents two advances: (i) scheduling approaches that balance centralized objectives of efficiency and reliability with distributed objectives of inter-airline equity, and (ii) game-theoretic approaches that identify opportunities of airline strategic behaviors, or lack thereof. Theoretical and computational results are shown. SA59

22