Informs Annual Meeting Phoenix 2018

INFORMS Phoenix – 2018

SA35

3 - Route-based Approximate Dynamic Programming for Demand Management in Attended Home Delivery Sebastian Koch, University of Augsburg, Universitatsstrasse 16, Augsburg, 86159, Germany, Robert Klein Attended home delivery describes the delivery of goods or services to a customer at an appointed time, often called the service time window, in order to avoid failure of delivery. In this talk, we re-consider the problem of integrated dynamic pricing for demand management regarding the customers’ time window choices and the subsequent operational vehicle routing. Thereby, we propose a novel approximate dynamic programming approach, anticipating the future value of the demand management decisions based on temporal features from dynamically generated route plans. In a simulation study, we compare our approach with

4 - A Generalized Routing and Accessibility Analysis System for Flooding: Iowa Testbed Peng Chen, The University of Iowa, Ann Melissa Campbell, Ibrahim Demir In this study, a generalized routing and accessibility analysis system is developed for flooding events. The system helps people find alternative routes and understand which areas in the disaster zone have limited travel accessibility. Our system can support emergency response activities with a real-time and dynamic routing framework for evacuation, rescuing civilians, delivering supplies, and deciding emergency center placement. n SA34 North Bldg 223 8:00 - 8:45 AnyLogic /8:45 - 9:30 Uber Vendor Workshop Session 1- AnyLogic: The Most Comprehensive Simulation Modeling Platform for Business and Research Rainer Dronzek, AnyLogic North America, Chicago, IL, United States, Arash Mahdavci In this tutorial, we will discuss how you can leverage unique features of AnyLogic simulation software and AnyLogic Cloud to solve your business challenges or perform scientific research. We demonstrate application of simulation in various domains and demonstrate state-of-the-art technologies that can take your simulation models to the next level in terms of sophistication and usefulness. 2 - Marketplace Optimization and Data Science at Uber Hamid Nazerzadeh, UBER & University of Southern California, Bridge Memorial Hall - BRI 401B, 3670 Trousdale Parkway, Los Angeles, CA, 90089, United States Marketplace is the center of Uber’s business, where riders and drivers come together at extraordinary scale. The data science team tackles problems such as optimizing Uber’s short and long term pricing mechanisms; efficiently matching incoming trip requests in Uber’s dispatch system; developing innovative incentive schemes that reward riders and drivers for choosing our network; and providing optimal routes and positioning suggestions to save time for everybody. In this presentation, we discuss in more details some of these challenging and innovative projects. n SA35 North Bldg 224A Joint Session AAS/TSL-Air: Passenger Flow Forecasting and Airport Operations Sponsored: Aviation Applications Sponsored Session Chair: Heng Chen, University of Nebraska-Lincoln, Lincoln, NE, 68588, United States 1 - Forecasting Airport Transfer Passenger Flow Using Machine Learning and Real-time Data Xiaojia Guo, PhD Student, University College London, International Hall, Lansdowne Terrace, London, WC1N 1AS, United Kingdom, Yael S. Grushka-Cockayne, Bert De Reyck Air passengers missing their connection can have a major impact on satisfaction and airline delays. Accurate forecasts of the flow of passengers and their journey times through an airport can help improve the experience of connecting passengers and support airline, airport, and air space punctuality. In collaboration with Heathrow Airport, we utilize real-time data to develop a predictive system based on a regression tree and Copula-based simulations. These real-time predictions can be used to inform target off-block time adjustments and determine resourcing levels at security and immigration. 2 - Runway Scheduling Under Winter Conditions Maximilian Pohl, Technical University of Munich, Munich, Germany, Rainer Kolisch We address the runway scheduling problem under consideration of winter operations. During periods of snowfall, runways have to be intermittently closed in order to clear them from snow, ice and slush. We propose an integrated discrete optimization model to simultaneously plan snow removal for multiple runways and to assign runways and starting and landing times to aircraft. To improve the computation times of our model formulation, problem specific pruning rules and valid inequalities are presented. We also discuss a method to derive initial start solutions heuristically. The model is tested and validated with real-world data from Munich International Airport.

state-of-the-art benchmarks, showing its profit potential. 4 - Restoration of Disrupted Services Through Reinforcement Learning Aybike Ulusan, Northeastern University, Boston, MA, United States, Ozlem Ergun

In the case of a disruptive event, the roads of a transportation network may be blocked with debris. Thus, many emergency response operations are hampered. In order to build a post-disruption resiliency, connectivity between supply and demand points within the service network should be maintained by effective clearance of the debris. We tackle this problem (road network recovery problem) with stochastic demands. We model this problem as MDP and develop a framework to aid the scheduling of road clearance decisions by modeling and simulating an intelligent clearance agent through reinforcement learning. Then, we use a linear function approximation to generate policies for large-sized instances.

n SA33 North Bldg 222C

Disaster and Humanitarian Logistics Sponsored: Transportation Science & Logistics Sponsored Session Chair: Zhijie Dong, Texas State University, San Marcos, TX, 78640, United States 1 - Hospital Coordination Strategies for Managing Emergency Patients in Disaster Elise Miller-Hooks, Professor, George Mason University, 208 Rosalie Cove Ct, Silver Spring, MD, 20905, United States, Bahar Shahverdi, Mersedeh Tariverdi A discrete event simulation framework is presented for assessing the benefits of coordinated response of hospitals in a disaster incident involving area-wide damage and mass casualties. Impacts on critical resources, physical spaces and demand are modeled. Findings from numerical experiments on a case study involving multiple hospitals spaced over a large metropolitan region show the potential of strategies involving resource sharing and joint capacity enhancement alternatives. 2 - Pre-positioning of Relief Supplies and Supplier Selection in Humanitarian Relief Zhijie Dong, Texas State University, RFM 4227B, 601 University Drive, San Marcos, TX, 78640, United States This paper integrates supplier selection into the pre-positioning of relief supplies. The supplier selection criteria include price discounts offered by suppliers based on order quantity and required lead time as well as suppliers’ own physical inventories. By considering failure risks, this paper presents a two-stage stochastic programming model to produce plans including facility location and inventory, supplier selection, and distribution of relief supplies. A case study focused on hurricane threat in the Gulf Coast area of the US illustrates application of the model. Sensitivity analysis of comparison experiments offers managerial insights for relief agencies. 3 - A Dynamic Team Orienteering Model of Dynamic Urban Search and Rescue Deployment Decision-making with Social Data Inputs Erin Mullin, University of Arkansas, Walnut Creek, CA, 94596, United States, Ashlea Bennett Milburn Deployment decisions for urban search and rescue (USAR) teams are critical in saving as many lives as possible. One challenge is having accurate information regarding the locations and needs of disaster survivors. Another is the simultaneous consideration of multiple inputs and constraints in a highly dynamic and uncertain environment. This research addresses these challenges by permitting social data during deployment planning and by modeling the USAR deployment problem as a dynamic variant of the team orienteering problem with time windows. A first-come-first-served strategy, often employed in practice, is compared with an optimal offline approach for a small set of test instances.

13