Informs Annual Meeting Phoenix 2018

INFORMS Phoenix – 2018

SB32

services to bridge the last mile. We base our reliability measures on continuous distributions of historical travel data and apply it to search for reliable itineraries. We also investigate the amount of historical travel data required to predict reliable itineraries. 2 - Intermodal Autonomous Mobility on Demand Systems Maximilian Schiffer, Dr., Technical University of Munich, Munich, 52072, Germany, Mauro Salazar, Federico Rossi, Daniele Vigo, Christopher H. Onder, Marco Pavone Currently, municipalities struggle to implement sustainable modes of transportation that can cope with increasing urban congestion. Against this background, we study the benefits of an autonomous mobility-on-demand system interacting with public transportation. We present a network flow model to jointly optimize the operation of a fleet of self-driving cars providing on-demand mobility and public transit. We design socially optimal control policies, including a pricing and tolling scheme to align the behaviour of selfish agents to the social optimum. Results on a case study of Manhattan show that such an intermodal system can yield significant benefits to an urban transportation network. 3 - Considering Complex Customer Preferences in Multimodal Travel Itineraries Digital travel apps allow for easy combination of an ever-increasing number of multimodal travel options. However, underlying optimization approaches focus on simple objectives such as travel time, price or number of transfers, and neglect complex customer preferences that would provide a customer-oriented selection of multimodal travel itineraries. We highlight challenges of considering complex customer preferences in the creation of multimodal itineraries, present an overview of current research in this area, and, based on standard algorithms available in the literature, present a first solution approach that takes more complex preferences into account. 1- Data Driven Supply Chain Management Practices at JD.com Zuo-Jun Max Shen, University of California Berkeley, 4141 Etcheverry Hall, Mail code 1777, Berkeley, CA, 94720-1777, United States As China’s largest retailer, online or offline, JD leverages advanced technologies such as AI to develop cutting-edge retail solutions that enable more personalize marketing and more efficient supply chain management, all aimed at improving the customer experience. In this tutorial, we will focus on a wide range of scenarios and applications in JD.com’s ecosystem, from both JD’s own e- commerce platform and its external business partners. We will cover the data-driven approaches used in these business scenarios to help improve supply chain efficiency and customer experience. We will also cover how technologies from fields such as operations research, data analytics, and machine learning are transforming the retail landscape. 2- Solving Multi-Objective problems with CPLEX Ed Klotz, IBM, P.O. Box 4670, Incline Village, NV, 89450, United States During this tutorial, you will learn how to use the Multi-Objective feature in the upcoming version of CPLEX Optimizer. With the ability to specify combinations of lexicographic and blended objectives, it allows you to specify your goals very precisely. By using this feature, you can save the trouble of developing your own multi-objective framework, and avoid the numerical difficulties often faced when combining vastly different objectives scales. Jan Fabian Ehmke, Otto von Guericke Universit, Universitaatsplatz 2, Magdeburg, 39106, Germany, Thomas Horstmannshoff n SB34 North Bldg 223 11:00 - 11:45 JD.com/11:45 - 12:30 IBM Vendor Demo Session

n SB32 North Bldg 222B

Joint Session TSL/ICS: Approximate Dynamic Programming and Reinforcement Learning for Routing II Sponsored: TSL/Freight Transportation & Logistics Sponsored Session Chair: Barrett Thomas, University of Iowa, Iowa City, IA, 52242-1000, United States 1 - Reinforcement Learning for Solving the Vehicle Routing Problem Reza Nazari, Lehigh University, Bethlehem, PA, 18015, United States, Afshin Oroojlooy, Lawrence V. Snyder, Martin Takac We present an end-to-end framework for solving Vehicle Routing Problem using reinforcement learning. In this approach, we train a single model that finds near- optimal solutions, only by observing the reward signals and following feasibility rules. Our model represents a parameterized stochastic policy, and by applying a policy gradient algorithm to optimize its parameters, the trained model produces the solution as a sequence of consecutive actions, without the need to re-train for every new problem instance. On capacitated VRP, our approach outperforms classical heuristics and Google’s OR-Tools on medium-sized instances in solution quality with comparable computation time. 2 - Reinforcement Learning for Same-day Delivery with a Heterogeneous Fleet of Drones and Vehicles Xinwei Chen, University of Iowa, Iowa City, IA, 52242, United States, Marlin Wolf Ulmer, Barrett Thomas With ten cities approved in May to test drones, commercial use of drones is nearly a reality. Previous work shows, compared to a homogeneous fleet of vehicles, using a heterogeneous fleet of drones and vehicles for same-day delivery increases the expected number of served customers. In that work, a travel time- dependent threshold determines whether to dispatch a drone or a vehicle when both are available. In this talk, we introduce a time-dependent threshold to account for changing resource availability and heterogeneous customer requests. We use reinforcement learning to learn policies that are represented by neural networks. Computational results demonstrate improvements in solution quality. 3 - Multi-Period Dynamic Technician Routing with Experience-based Service Times Barrett Thomas, University of Iowa, S210 PBB, Iowa City, IA, 52242-1000, United States, Mike Hewitt, Xi Chen We study a multi-period technician routing problem in which technicians gain productivity as they gain experience. Tasks to be performed in the current day are known, but the tasks to be performed in subsequent days are uncertain. We develop an Approximate Dynamic Programming-based approach that incorporates into daily assignment decisions estimates of the long-term benefits associated with experience accumulation. Using an extensive computational study, we derive insights into how an organization can schedule their employees in a manner that enables meeting both near and long-term demands. 4 - Multi-period Workload Balance in Last-mile Urban Delivery Yang Wang, Tsinghua University, Beijing, China, Lei Zhao, Martin W. P. Savelsbergh When dispatching last-mile delivery among couriers, managers need to consider workload (im)balance among couriers besides their total travel cost. We differentiate two types of workload: incentive workload (e.g., number of packages) that affects couriers’ income and effort workload (e.g., travel distance or time). We model the multi-period (e.g., a month) workload balance problem in last-mile delivery as a stochastic dynamic program and study the impact of various dispatching policies on the expected total travel cost and workload (im)balance. Multimodal Passenger Transportation Sponsored: Transportation Science & Logistics Sponsored Session Chair: Jan Fabian Ehmke, Otto-von-Guericke University, Magdeburg, 39106, Germany 1 - Using Analytics to Find Reliable Itineraries for Multimodal Travel Michael Redmond, University of Iowa, 820 Spencer Dr, Iowa City, IA, 52246, United States, Ann Melissa Campbell, Jan Fabian Ehmke Multimodal travel requires the combination of several transport options. In this presentation, we will introduce reliability measures of multimodal travel itineraries. These itineraries may include scheduled travel modes such as flights or buses as well as non-scheduled travel modes such as driving or ride-sharing n SB33 North Bldg 222C

42