2016 INFORMS Annual Meeting Program

MA60

INFORMS Nashville – 2016

MA60 Cumberland 2- Omni Fleet and Marketplace Optimization for Mobility-on- Demand (MOD) Systems Sponsored: TSL, Urban Transportation Sponsored Session Chair: Samitha Samaranayake, Cornell University, 317 Hollister Hall, Ithaca, NY, 14853, United States, samitha@alum.mit.edu 1 - Queueing-theoretical Models For Mobility-on-demand Systems: Theory And Algorithms Frederico Rossi, Stanford University, Stanford, CA, United States, frossi2@stanford.edu, Marco Pavone In this talk I will present recent advances towards modeling and controlling autonomous mobility-on-demand (AMoD) systems, an emerging mode of personal transportation wherein robotic, self-driving vehicles transport customers on-demand. First, I will present queueing-theoretical models inspired by the theory of Jackson and BCMP queueing networks. Such models provide structural insights about the performance of AMoD systems and guidelines for the design of routing algorithms for the robotic vehicles. Then, I will discuss large-scale coordination algorithms for AMoD systems that are aimed at throughput maximization and can handle congestion and charging constraints. 2 - Fleet Management In Mobility-on-Demand Systems With Shared Rides Samitha Samaranayake, Cornell University, samitha@alum.mit.edu We consider a MoD system with ridesharing between passengers. Inherent to the formulation are two important attributes: (i) the need to rebalance empty vehicles and (ii) the ability to identify lucrative ridesharing corridors via trip chaining. We present a mixed-integer linear programming (MILP) formulation of the problem and show how a heuristic (feasible) solution to the problem can be obtained in polynomial-time by independently solving the ride-matching and rebalancing problems. This approximate solution can be used as a initial guess when solving the coupled problem via an MILP solver. 3 - Dynamic Pricing In Ride-share Platforms Siddhartha Banerjee, Cornell University, sbanerjee@cornell.edu Much of the success of ride-sharing platforms like Lyft and Uber is ascribed to their ability to do fast-timescale dynamic pricing - where prices can react to instantaneous system state, and across very small geographic areas. We explore the value of such dynamic pricing via a model which combines a queueing model for the dynamics of the platform’s operations with strategic models of both passenger and driver behavior. In particular, we suggest that dynamic pricing may not be better than the optimal static price, but rather, allows the platform to realize the optimal price with limited knowledge of system parameters. Joint work with Ramesh Johari, Carlos Riquelme, and the data science team at Lyft. 4 - Marketplace Optimization At Uber Robert Phillips, Uber, Palo Alto, CA, United States, robert.phillips@uber.com The rapid acceleration of the sharing economy has introduced a myriad of challenges for two-sided marketplaces. This talk will address how optimization and machine learning are powering the dynamic marketplace at Uber, a platform that has connected over one billion riders and drivers across more than 60 countries. Topics that will be surveyed include dynamic pricing, matching riders in uberPOOL, and real-time on-demand delivery services.

2 - Intermodal Hub Simulation Mike Prince, BNSF Railway, Contact: mike.prince@bnsf.com Intermodal hubs are the facilities at which BNSF Railway’s intermodal trains interface with customers. This presentation will discuss an AnyLogic simulation model that was developed for the purpose of assisting in the capital expansion planning process for these facilities. 3 - Utilizing Rail Information In Intermodal Operations Georgi Tasev, Schneider, Contact: TasevG@Schneider.com Accurate train ETA information is critical to intermodal dray operations and directly influences the ability to serve customers effectively. In this session, we will review how Schneider uses train information provided by our rail partners to optimize key operations, such as appointment setting and dispatch. In addi- tion, we will cover the analysis that was completed to study the accuracy of rail ETA information at key time intervals. Lastly, we will discuss the implementation and results of building a direct feed for train ETA information into Schneider’s system. MA62 Cumberland 4- Omni Determinants of Aviation Strategies and Market Outcomes Sponsored: Aviation Applications Sponsored Session Chair:SufficMartin E Dresner, University of Maryland-College Park, R H Smith School of Business, College Park, MD, 20742, United States, mdresner@rhsmith.umd.edu 1 - The Impact Of Predicted Quality On Customer’s Quality Assurance Behaviors In The Us Airline Industry Woohyun Cho, University of New Orleans, New Orleans, LA, United States, wcho@uno.edu Dong-jun Min, Pamela Kennett-Hensel We empirically examine the drivers of customers’ voluntary quality assurance behaviors (QAB). Using survey data and archival data from the US airline industry, we show that whereas an increase in predicted quality of airlines departure operations (e.g., on-time performance and flight frequencies) leads to a decrease in the level of QAB (i.e., customer wait time for their flight at the airport), an increase in price leads to an increase in QAB. Our finding also indicates that the expense of exercising QAB reduces QAB. We emphasize the importance of properly measuring the impact of predicted quality and price on the customer’s role, as it may help share the cost of managing quality with their customers. 2 - Passenger Facility Charge Vs. Airport Improvement Program Funds: A Dynamic Network Dea Analysis For U.S. Airport Financing Bo Zou, University of Illinois at Chicago, 2073 Engineering Research Facility, 842 West Taylor Street, Chicago, IL,n Passenger Facility Charge (PFC) and the Airport Improvement Program (AIP) are two major sources to finance U.S. airports. This paper develops a novel dynamic network DEA framework to investigate the substitutability between PFC and AIP funds. We find that the studied U.S. airports can substitute PFC for 8-35% of the current AIP funds and contribute significantly to the proposed plan of the US congress to cut AIP funding. In addition, the amount of PFC-for-AIP funds substitution negatively correlates with the productive efficiency of airports. The findings send an important message for future policy reforms on U.S. airport financing. 3 - Measuring Competition Intensity And Product Differentiation: Evidence From The Airline Industry Benny Mantin,University of Waterloo, bmantin@uwaterloo.ca David Gillen, Tuba Delibasi Measuring the degree of competition in markets is essential for policy and decision makers. Commonly used structural indices (e.g., HHI) overlook how firms compete with each other and the intensity of the competition. We propose a new competition measure: Schedule (Temporal) Differentiation Metric, STDM, which encapsulates firms’ market shares as well as the degree of overlap and substitution between the competing services—critical elements in service industries. We demonstrate the STDM using aviation markets revealing a significant improvement in explaining prices, how the effect varies across fare percentiles, and how the insights change with the business models of the competing firms. Univ60607, United States, bzou@uic.edu Young-Tae Chang, Hyosoo Park, Nabin Kafle

MA61 Cumberland 3- Omni

Intermodal Transportation Sponsored: Railway Applications Sponsored Session Chair: Mike D Prince, BNSF Railway, Fort Worth

1 - Intermodal Empty Railcar Distribution Optimization Shantih Spanton, CSX Transportation, Jacksonville, FL, Shantih_Spanton@csx.com, Jagadish Jampani Optimization models to effectively reposition the empty railcars. The forecasting model predicts the future demand for the containers and trailers, which is subse- quently translated into railcars. This demand data is converted into equivalent number of railcars which is input into the optimization model. In addition, train profiles, network and terminal attributes are input into the optimization model. The model also predicts when and where the loaded railcars will become avail- able in the selected optimization time horizon. This optimization model is embedded with a real time tool that is used by the intermodal railcar distribution team.

142