Informs Annual Meeting 2017

SB65

INFORMS Houston – 2017

SB65

capacity queues. We estimate demand for parking across time and location and the subsequent volume of traffic generated. Combined with travel time data, the impact of this congestion can be estimated. 2 - A Data Driven Study on the Relation Between Parking Garage Occupancy and Traffic Speeds in Urban Areas Rui Ma, University of California, Davis, 2041 Academic Surge, University of California, One Shields Avenue, Davis, CA, 95616, United States, drma@ucdavis.edu, Shenyang Chen, Michael Zhang Theoretical studies have suggested that parking and traffic conditions in cities are related, yet few empirical studies are conducted to quantify this relationship. We develop a data processing procedure to extract and fuse the time-varying parking garage occupancy and traffic speed data at the street or city blocks levels via APIs, and study their relationship. It is found that during the morning and evening commute periods, the relationship between the average speed and the parking occupancy are distinct and both can be modeled by a reversed logistic function. Such a macroscopic relationship can be exploited to evaluate the effectiveness of downtown access control or congestion charging. 3 - A Spatial-temporal Flexible Parking Reservation System Xin Wang, University of Wisconsin-Madison, 8916 Red Beryl Drive, Middleton, WI, 53562-4278, United States, xin.wang@wisc.edu, Xiaotian Wang In a reservation-based parking system, customers departing late cause service failure for subsequent customers. To guarantee the service level, the parking system needs to run at a low utilization, which is a huge waste of parking resources, especially in congested urban area. We propose a flexible reservation system with spatial-temporal flexibility to address the issue. A game-theoretical model is used to capture the customers’ utility over the flexibility. 1 - Optimally Replacing Multiple Systems in a Shared Environment David Tarek Abdul-Malak, University of Pittsburgh, 1702 Jancey Street, Pittsburgh, PA, 15206-1146, United States, dta10@pitt.edu, Jeffrey P. Kharoufeh In this talk we will present a model for replacing multiple stochastically degrading systems using condition-based maintenance. Systems are assumed to degrade in a shared, exogenous, Markov modulated environment. Continuous state variables and a high dimensional state space cause the problem to be computationally intractable. To overcome these complications, structural results are proven and a novel reinforcement learning (RL) approach is employed. 2 - Maintenance Optimization for Partially Observable Multi- component Systems Ayse Sena Eruguz, Postdoctoral Researcher, Eindhoven University of Technology, P.O. Box 513, Eindhoven, 5600 MB, Netherlands, a.s.eruguz@tue.nl, Rob Basten, Lisa M.Maillart We consider a multi-component system in which a condition parameter (e.g., vibration or temperature) is monitored. The outcome of monitoring indicates whether the system is functioning properly, is defective, or has failed. However, the condition signal does not reveal which component in the system is defective or has failed. The decision maker needs to infer the exact state of the system from the current condition signal and the past data, in order to decide when to intervene for maintenance. We model this problem as a partially observable Markov decision process. 3 - Joint Planning of Multiaction Maintenance with Complex Effects Yisha Xiang, Lamar University, 2626 Cherry Engineering Building, Beaumont, TX, United States, yxiang@lamar.edu, Yue Shi, David W. Coit Existing maintenance literature mainly focus on a single type of preventive maintenance action (e.g., replacement) and often make simple assumptions regarding the effects of maintenance activities. Many important maintenance effects are overlooked in these models, for example, deterioration suppression after maintenance interventions, deterioration rate reduction under maintenance effect, and the random duration of the effective maintenance period. In this research, we consider joint planning for multiple maintenance actions with complex effects, and formulate the problem as Markovian decision process. Structural properties of optimal policies are investigated. SB67 371B Condition-Based Maintenance and Support Operations Sponsored: Quality, Statistics and Reliability Sponsored Session Chair: Chiel Van Oosterom, vanoosterom@ese.eur.nl

370F Decision Analytics Applications Sponsored: Data Mining Sponsored Session Chair: Victoria Chen, vchen@uta.edu 1 - A High Dimensional State Transition Development Framework for Deicing Activities at Dallas Fort Worth International Airport Ashkan Aliabadi Farahani, University of Texas, Arlington, TX, United States, ashkan.aliabadifarahani@mavs.uta.edu In complex systems, the system state is often high-dimensional and state transitions are unknown. When system data are available, it may be possible to empirically estimate state transitions to enable simulation and/or optimization of the system. This paper presents a data-driven framework for state transition estimation and development. 2 - TK-MARS: An Efficient Approach for Black-Box Optimization Hadis Anahideh, University of Texas at Arlington, Arlington, TX, United States, hadis.anahideh@mavs.uta.edu, Jay Rosenberger, Hadis Anahideh In this study, a modified version of MARS, TK-MARS (Tree Knot MARS), is developed to improve the use of MARS within the surrogate optimization context. TK-MARS is genuinely able to identify peaks and valleys where optimization needs to know. More importantly, due to the limitations of existing optimization test problems, a novel construction of representative surrogate- optimization test problems are presented to overcome the real world noisy black-box function with a set of important variables. 3 - Inverse Probability of Treatment Weighting in Adaptive Pain Management with Correlated Treatments Nilabh Ohol, University of Texas-Arlington, 900 Greek Row Drive, Apt 201, Arlington, TX, 76013, United States, nilabh.ohol@mavs.uta.edu, Aera Kim LeBoulluec, Victoria C. P. Chen, Li Zeng, Jay Michael Rosenberger, Robert J.Gatchel Interdisciplinary pain management combines biological and psychosocial factors causing patient’s pain and administers best treatments. To improve pain outcomes, our framework employs state transition and outcome models using data from Eugene McDermott Center for Pain at the University of Texas Southwestern Medical Center at Dallas. The sequential treatment structure of the data leads to a form of endogeneity. This research develops a process based on the inverse probability of treatment weighted method to address endogeneity and presents a framework to develop a general approach for correlated treatments. 4 - Machine Learning Approaches to Modeling Category Sales: Implications for Optimal Store-level Pricing and Promotions Durai Sundaramoorthi, Washington University in Saint Louis, 10352 Conway Road, Saint Louis, MO, 63131, United States, dsundaramoorthi@gmail.com, Seethu Seetharaman We discuss and propose five machine learning models as candidate models of product category sales as functions of marketing variables (price, display, feature, price promotion), both within and across categories at the same retail store. Using store-level weekly scanner data from 24 product categories in each of 9 stores of a supermarket chain over a period of 5 years, we estimate the five proposed machine learning models and compare their empirical performance to those of three commonly used statistical models. Taking the overall best-performing machine learning, we optimize the profit made by the chain.

SB66

371A Design and Impacts of Parking Systems Sponsored: Transportation Science & Logistics Sponsored Session

Chair: Xin Wang, University of Wisconsin-Madison, 8916 Red Beryl Drive, Middleton, WI, 53562-4278, United States, xin.wang@wisc.edu Co-Chair: Xin Wang, University of Wisconsin-Madison, Middleton, WI, United States, xin.wang@wisc.edu 1 - Data-driven Modeling and Analysis of Urban Congestion Caused by Demand for Parking Chase Patrick Dowling, PhD Student, University of Washington, Seattle, WA, 98105, United States, cdowling@uw.edu, Tanner Fiez, Lillian Ratliff, Baosen Zhang Congestion caused by drivers searching for parking has been limited to theoretical modeling and empirical study; yet, pervasive installation of digital parking meters providing troves of data on parking transactions. These data can be used to estimate occupancy, but are not amenable to canonical theoretical models.We present a new model, amenable to transaction data, using a network of finite

58