Informs Annual Meeting Phoenix 2018

INFORMS Phoenix – 2018

TC36

n TC36 North Bldg 224B Joint Session Drones/Practice Curated: Operations Research Applications in Drones Emerging Topic: Robotics, Drones and Autonomous Vehicles in Logistics Emerging Topic Session Chair: Gino Lim, Univeristy of Houston, Houston TX Co-Chair: Seon Jin Kim, University of Houston, Houston, TX, 77094, United States 1 - A Hybrid Approach for Extending Drone Flight Duration in Real Time Seon Jin Kim, University of Houston, 422 Cypress Vista, Houston, TX, 77094, United States, Gino J. Lim This study focuses on continuous flight missions using battery powered drones. There are two types of battery charging techniques: stationary and dynamic wireless charging systems. However, the stationary system requires frequent visits of drones to the ground control center to replace or charge the batteries, while the dynamic system does not have reliable battery charging efficiency. To overcome both of these issues in a battery charging system, a hybrid battery charging system combining stationary and dynamic wireless charging systems has been proposed. 2 - Carrier-drone Vehicle Routing Problem: Multiple Deliveries by a Single Flight A carrier-drone hybrid delivery vehicle uses a truck as a station for delivery drones. The truck with a larger cargo capacity carries and launches drones to customers in locations where traffic congestions or physical obstructions are present. Because of its significant advantages, various hybrid routing models have recently been proposed and researched. We’d like to introduce a new model that allows drones to make multiple deliveries by a single flight. This property is expected to further enhance logistical efficiency compared to existing models limited to single delivery per flight. In this study, this new hybrid vehicle routing problem is presented along with evidence of its practicality. 3 - Operation Based Vehicle Routing Problem for Drone Assisted Truck Delivery Jinkun Lee, East Carolina University, Greenville, NC, United States, Sung Hoon Chung We consider a drone assisted truck delivery problem where the truck serves as a mobile station of single or multiple drones while delivering items to customers at the same time. We formulate the model as a mixed integer programming problem where the objective is to minimize the total delivery time. We describe the pre- process of the exact algorithm where the possible operations, defined as drone-truck delivery combination units, are identified to reduce the overall computational complexity. We present case studies and discuss efficacy and efficiency of the proposed approach. 4 - Deliver or Not?: Revenue and Capacity Management for Drone-based Delivery Services Zhangchen Hu, University of Massachusetts Amherst, Amherst, MA, 01003, United States, Heng Chen, Senay Solak Drones are expected to become a key component of commercial delivery services for retailers and courier companies in the near future. Using currently available data, we develop both queueing-based and dynamic programming-based models to determine when drone delivery options should be enabled by retailers. Both analytical and numerical analyses are provided. 5 - An Unmanned Aircraft System Path Planning Model for Inspecting Wind Turbines Hyeoncheol Baik, PhD Candidate, Auburn University, Shelby 3301, Auburn, AL, 36849, United States, Jorge F. Valenzuela Rope access is a common technique for maintaining and servicing wind turbines. This technique is risky and inefficient. An alternative approach is to use an unmanned aircraft system (UAS). The challenge is to determine an efficient way to deploy the UAS over a region with multiple wind turbines. In this presentation, we describe a mathematical programming path planning model for a UAS that works cooperatively with a ground vehicle. The objective of the model is to minimize the total inspection time. We use a real wind farm to test our model. Ho-Young Jeong, Purdue University, 315 N. Grant St, West Lafayette, IN, 47907, United States, Seokcheon Lee

Sponsored: Applied Probability Sponsored Session Chair: Jose Blanchet, Columbia University, New York, NY, 10027, United States 1 - Nonconvex Optimization via Accelerated Langevin Diffusion Yi Chen, Northwestern University, Evanston, IL, United States Nonconvex optimization arises from numerous machine learning problems. However, due to the existence of spurious local minima, nonconvex optimization remains challenging in both theory and practice. We investigate the method of solving optimization via Langevin diffusion. We consider some techniques accelerating the convergence, which enable us to obtain a better solution within a shorter time. 2 - Even Faster Algorithms for Optimal Transport Carson Kent, Stanford University Over the past decade, optimal transport and Wasserstein metrics have come to play a major role in a wide variety of tasks throughout optimization and machine learning- e.g. resource allocation problems, unsupervised learning, distributionally-robust optimization, etc. Typically, a major bottleneck for these applications is the computation of a transportation plan between discretized measures. Recent algorithmic advances have resulted in a bevy of new methods for this problem which have linear-time complexity guarantees and can scale to large problem sizes. In this work, we will present several new developments in this lineage of results which obtain superior complexity/performance. Additionally, we will provide perspective for where further computational improvements for this problem are likely and unlikely to be found. 3 - Space-filling Design for Nonlinear Models Chang-Han Rhee, Centrum Wiskunde and Informatica, Jakoba Mulderplein 164, Amsterdam, 1018 MZ, Netherlands, Enlu Zhou, Peng Qiu Traditional space-filling designs for computer experiments aim to fill the parameter space with design points that are as “uniform as possible. However, the resulting design points may be non-uniform in the model output space failing to provide a reliable representation of the output manifold, and becoming highly inefficient or even misleading in case the computer experiments are non-linear. In this talk, we propose and analyze an iterative algorithm that fills in the model output manifold uniformlyùrather than the parameter space uniformlyùso that one could obtain a reliable understanding of the model behaviors with the minimal number of design points. 4 - Approximating Performance Measures for Non-stationary Markov Chains Zeyu Zheng, Stanford University, Stanford, CA, 94305, United States, Harsha Honnappa, Peter W. Glynn Non-stationarity naturally arises in realistic modeling of many problems in operations research and operations management. Unfortunately, the development of closed-form theory for the great majority of stochastic models requires an assumption of stationary dynamics. We develop rigorous approximations to various expectations associated with generic non-stationary Markov models. These approximations are typically no harder to compute, either analytically or numerically, than are the linear systems associated with the stationary case. Asymptotic validity is established with slowly changing transition probabilities (which is only required over a small portion of the entire horizon). n TC38 North Bldg 225B APS-Special Speaker Session Sponsored: Applied Probability Sponsored Session Chair: Henry Lam, Columbia University, New York, NY, 10027, United States Chair: Harsha Honnappa, Purdue University, Purdue University, West Lafayette, IN, 47906, United States 1 - Maintaining Privacy in Statistical Estimation: Definitions, Optimality, and Estimators In this tutorial, I will describe the techniques for maintaining privacy in statistical estimation, machine learning, and stochastic optimization. I will begin the tutorial with several recent definitions, beginning with differential privacy, and describe their implications. After this, I will describe several recently developed tools for deriving optimal estimators, and give examples of their use in theory and practice, including machine learning, statistical, and optimization applications. John Duchi, Department of Statistics - 390 Serra Mall, Stanford University, Stanford, CA, 94305, United States

n TC37 North Bldg 225A Recent Advances in Applied Probability

n TC39

320