Informs Annual Meeting Phoenix 2018

INFORMS Phoenix – 2018

MD39

4 - Optimizing Traffic Signals Using Multi-source Data: From a Practical Perspective

n MD37 North Bldg 225A Markov Lecture Sponsored: Applied Probability Sponsored Session Chair: Jose Blanchet, Columbia University, New York, NY, 10027, United States Co-Chair: Rami Atar, Technion, Technion, Haifa, 32000, Israel Co-Chair: Amber L. Puha, California State University-San Marcos, San Marcos, CA, 92096-0001, United States 1 - Adaptive Estimation using Regularized Empirical Risk Sara van de Geer, PhD in Mathematics in 1987 at Leiden University, ETH Z urich, Zurich, 8092, Switzerland We examine a general class of algorithms based on minimizing an empirical risk function. Examples of empirical risk functions are the least squares risk, and more generally a minus log-likelihood. We consider complex models, for example models with a large number of parameters. Adding a penalty to the empirical risk will help to avoid overfitting. Our focus will be on norm-penalized estimators, where the norm is inducing sparsity. Aim is to show why such penalties lead to favourable theoretical properties namely adaptivity to both complexity and curvature. Moreover, the results do not always require the estimator to be a global minimizer. 2 - Discussant Alexandre Belloni, Duke University, Fuqua School of Business, 100 Fuqua Drive, Durham, NC, 27708, United States Innovative Traffic Control and Operations Sponsored: TSL/Intelligent Transportation Systems (ITS) Sponsored Session Chair: Larry Head, University of Arizona, Tucson, AZ, 75719, United States Co-Chair: Qing He, University at Buffalo, SUNY, Buffalo, NY, 14260, United States 1 - Understanding the Problem: Traffic Safety Analysis Brendan Russo, Assistant Professor, Northern Arizona University, P.O. Box 15600, Flagstaff, AZ, 86011, United States Traffic crashes cost society billions of dollars each year as a result of property damage, injuries, fatalities, and non-recurring delay. This analysis explores factors affecting the frequency and severity of crashes along the Arizona portion of the I- 10 corridor, with a particular focus on freight-related crashes. The safety performance along the I-10 is analyzed through the development of crash frequency and severity prediction models using integrated crash, roadway, traffic, and environmental data. The results showed that several roadway-, crash-, vehicle-, and person-specific variables were associated with the frequency and/or severity of crashes along the study corridor. 2 - Machine Learning Investigation using Integrated Data for Identification of Critical Factors for Safety and Mobility Zuoyu Miao, University of Arizona, Tucson, AZ, United States Potential safety and mobility concerns are rising due to the facts that huge losses could occur by the risks of accidents and travel time degradation. To identify critical factors leading to potential losses, multiple data sources are investigated, including history crash reports, travel time records, weather information, etc. Several statistical machine learning methods are combined for analyzing above data and identifying critical factors. Based on identified critical factors and their related impacts, several different countermeasures are discussed for avoiding severe accident injuries and decreasing the accident frequency. 3 - Quantifying Freight Performance Measures using Multisource Traffic Data Abolfazl Karimpour, Graduate Research Assistant, The University of Arizona, Tucson, AZ, United States, Yao-Jan Wu With the emerging development of Intelligent Transportation System (ITS) technologies, surface-transportation data can now be collected by a wide variety of ITS traffic detectors, including Bluetooth detectors, automatic vehicle location (AVL) devices, inductive loop detectors, and radar-based detectors. It has been challenging to take full advantage of the ITS data from multiple sources by enabling them to exchange information with each other to compensate for their various disadvantages. This presentation will be focused on freight performance measurement using multi-source traffic data collected on the Arizona freeway network. n MD38 North Bldg 225B

Yao-Jan Wu, Assistant Professor, University of Arizona, 1209 E. 2nd St. Room 324F, Civil Engineering, Tucson, AZ, 85721, United States Cities have invested a great deal amount of funding in intelligent transportation systems, especially traffic sensor technologies, to better manage traffic and reduce congestion. However, not all cities/agencies have fully utilized the capacities of the advanced technologies. This presentation aims at providing data-driven solutions to enhancing traffic data usability for optimizing traffic signal timing using multiple sources of data. Different from traditional operation research, this presentation focuses more on the practical side of traffic signal optimization and take other key factors, such as safety and operability into account during the signal re-timing process. n MD39 North Bldg 226A Supply Chain Risk Management Sponsored: Manufacturing & Service Oper Mgmt Sponsored Session Chair: Florian Lucker, Cass Business School, London, EC1Y 8TZ, United Kingdom 1 - Recall Analytics and Supply Chain Risk Ahmet Colak, Clemson University, Greenville, SC, 29601, United States, Robert Louis Bray Product safety is one of the most critical drivers of brand equity and reputation in consumer goods industries. I will focus on how automakers trade off between defect liabilities and product recalls, using a sample of 1M consumer defect reports and 15K component recalls. I will also talk about how the federal government enforces its mandated-recall programs, studying the government- initiated recalls for a period of 20 years. I will then elaborate on the supply chain drivers of quality defects from a sample of 26K auto suppliers. And last, I will discuss about data analytics tools to improve failure and recall prediction accuracy dynamically, mining detailed defect report statements. 2 - Inventory Dispersion in a Sequential Inventory System with Demand Forecast Evolution Isik Bicer, Swiss Federal Institute of Technology in Lausanne (EPFL), Chemin de Veilloud 11, Ecublens, 1024, Switzerland, Florian Lucker Inventory dispersion along a supply chain occurs as raw materials are processed and allocated to the production of different end-products, and also when the end- products are sent to different locations to be sold to the customers. We develop a dynamic optimization model to find the optimal order quantities in a multi- echelon setting such that inventory dispersion may occur in different echelons and the demand forecasts of the end products evolve according to a multiplicative martingale model. We show that postponing high-cost activities leads to higher profits than postponing the point of differentiation. 3 - Examining Resiliency in Pharmaceutical Drug Supply Chain Considering Human Behaviors Component Rozhin Doroudi, Northeastern University, Boston, MA, 02116, United States, Ozlem Ergun The growing epidemic of drug shortages in the United States demonstrates the lack of resiliency within drug delivery supply chain. A key, and often understudied, element in resiliency of supply chains is the role of human behavior. Human decision makers should anticipate and react to the contingencies and disruptions that arise. We propose an integrated simulation framework which allows for instantiating, testing, and improving supply chain resiliency while accounting for human behavior component. We demonstrate the applicability of our method through several experiments. 4 - Scheduling Optimization in Continuous Steel Casting and Rolling Operations Joshua Betz, Colorado School of Mines, CO, 80401, United States In continuous steel casting operations, molten steel is batched into heats inside a ladle that is cast into slabs, which are then rolled into coils. We present a mixed integer program to produce a daily casting schedule that is solved using state-of- the-art software. This model minimizes penalties incurred by violating plant best practices while strictly adhering to safety and logical constraints to manage risk for manufacturing incidents that can occur in the rolling mill. A heuristic produces an initial feasible solution to expedite the generation of near-optimal schedules.

227