Informs Annual Meeting 2017

MA66

INFORMS Houston – 2017

3 - Resilient Route Design for Collection of Material from Suppliers Zhijie Dong, Assistant Professor, Texas State University, San Marcos, TX, 78666, United States, sasha.dong@txstate.edu As production schedules of manufacturing operations change, the amounts of material to be picked up at individual suppliers can be quite volatile, but it is desirable to have relatively regular route structures in place so that the system can adapt to varying volumes without requiring complete redesign of collection routes every time there is a change in the production schedule at the manufacturing plant. Our model accomplishes this resilient design of routes using a stochastic version of a capacitated clustering formulation. The model is constructed as a two-stage stochastic mixed integer program. 4 - Service Network Design Problem on an Hierarchical Hub-and-Spoke Network for a Major Carrier In this study, we present mathematical models and solution procedures for the routing and scheduling of the Caribbean air feeder network of a major express cargo service provider. The problem is modeled as a service network design problem on a hierarchical hub-and-spoke network. The proposed solution methodology first executes an initial data processing that involves route generation and elimination in order to contain the size of the solution space for the mathematical model. We consider two versions of the proposed model based on two scenarios: dedicated runs and milk runs. We compare two approaches in terms of operational costs, flight hours, and fleet sizing. 371A Network Resilience Sponsored: Transportation Science & Logistics Sponsored Session Chair: Yan Deng, Cornell University, Cornell University, 2406 Hasbrouck Apartment, Ithaca, NY, 14850, United States, yd256@cornell.edu 1 - Disaster Induced Traffic Network Analysis using Percolation Theory Yaoming Zhou, University of Hong Kong, University of Hong Kong, Pokfulam, Hong Kong, Hong Kong, youngme910@hku.hk, Junwei Wang, Jiuh-Biing Sheu The impact of traffic networks’ structure change on operations management during disasters is usually ignored. We propose a stochastic mathematical framework utilizing percolation theory to systematically study how will different magnitude of disasters affect the global and local connectivity of traffic networks. This method also helps find out when and to what extent can the impacts be ignored. The impacts of natural and man-made disasters are compared. We discussed the application of our method to a real-world traffic network with some important insights. 2 - An Optimal Stopping Problem to Mitigate Productivity Losses of a Transportation System Utilizing Hurricane and Storm Surge Forecasts Rodney Middlebrooks, Graduate Student, North Carolina A&T State University, Greensboro, NC, United States, middlebrooks.r@gmail.com In the context of a transportation network, a system-wide operational shutdown can be ordered prior to the realization of expected hurricane landfall or storm surge inundation. The decision is binary, with the choice to either continue with routine operations or suspend them, initiating protective measures. The critical infrastructure decision problem is modeled as a finite-horizon optimal stopping problem. The objective is to minimize the loss of productive output of a transportation system, given real-time hurricane and storm surge forecast information. 3 - Vulnerability Assessment and Mitigation of Complementary Transportation Systems under Worst-case Saptially Localized Attacks Liu Hong, Associate Professor, Huazhong University of Science and Technology, 1037 Luoyu Road, Wuhan, 430074, China, liu.hong@hust.edu.cn, Min Ouyang, Xin Zhong Spatially localized attacks, caused by natural hazards or malicious attacks, can make serious loss to the transportation systems. This paper proposes an approach to assess and mitigate the vulnerability of complementary transportation systems under worst-case spatially localized attacks. The national level complementary railway and airline systems, and the urban level complementary bus and subway systems, are used as numerical experiments to illustrate the tractability and effectiveness of the proposed method. Yusuf Secerdin, University of Miami, 1251 Memorial Drive, McArthur Engineering Building, Coral Gables, FL, 33146, United States, yusufsecerdin@miami.edu, Murat Erkoc MA66

4 - Socially Profitable Transport Infrastructure Investment Strategies under Demand Uncertainty: A Real Options Based Approach Yan Deng, Cornell University, Cornell University, 2406 Hasbrouck Apartment, Ithaca, NY, 14850, United States, yd256@cornell.edu, Oliver Gao This paper developed a real options based methodology to address the optimal transport investment strategies under travel demand volatility. A socially profitable demand threshold is derived under the market share and competition. The methodology is subsequently applied to the High Speed Rail investment in the United States. The results compared the investment criteria from real options with the NPV valuation approach, and illustrated the investment returns from consumer, producer and environmental surplus. Sensitivity analysis was carried out to examine the characteristics of the optimal transport investment rules and the policy implications under the effects of key parameters. 371B Diagnostic/Predictive Data Analytics and Data-Driven Decision Making for Smart and Connected Systems Sponsored: Quality, Statistics and Reliability Sponsored Session Chair: Raed Al Kontar, University of Wisconsin-Madison, 1513 Mechnaical Engineering Building Room 3255, Madison, WI, 53706, United States, alkontar@wisc.edu Co-Chair: Shiyu Zhou, University of Wisconsin-Madison, Madison, WI, 53706-1572, United States, szhou@engr.wisc.edu 1 - A Novel Pattern-frequency Tree Approach for Dynamic Transition Analysis Cheng-Bang Chen, Pennsylvania State University, 310 Leonhard Building, University park, PA, 16802, United States, czc184@psu.edu, Hui Yang, Soundar Kumara This paper presents a novel methodology of pattern-frequency tree for transition analysis in dynamic nonlinear systems. First, we propose Hyperoctree State space Aggregation Segmentation to delineate the high-dimensional dynamic processes in a continuous state space. Then, we develop a pattern-frequency tree to characterize and model the pattern distribution. Finally, we leverage pattern- frequency distribution information to develop a k-Maximin deviation algorithm for e ective and e cient detection of process anomalies. Experimental results show that the proposed method performs better than the conventional methods in multi-sensor settings and high-dimensional environments. 2 - Contamination Source Identification Based on Sequential Bayesian Approach for Water Distribution Network with Stochastic Demands Chao Wang, University of Wisconsin-Madison, Rm. 3255, ME building 1513 University Ave., Madison, WI, 53706, United States, cwang436@wisc.edu, Shiyu Zhou The contamination source identification problem in the water distribution network is studied under the Bayesian framework. Simulations for the contamination propagation are conducted, where sensor alarms are recorded to establish the observation probability distributions. Then, the posterior probability of each possible source can be calculated. Finally, the contamination source is identified based on the ranking of the posterior probability. The contribution of this work is that the observation probability distributions are organized into hierarchical tree structures and the challenge of combinatorial explosion is avoided. The effectiveness of this method is verified by a case study. 3 - Active Learning for Gaussian Process Classification using a Data-based Kernel Hang Li, Pennsylvania State University, University Park, PA, 16803, United States, hul180@psu.edu, Enrique del Castillo, George Runger Active learning and semi-supervised learning are important techniques in the machine learning field, especially when the labeled instances are very difficult, time-consuming or expensive to obtain. In this presentation we discuss a Gaussian Process Classification (GPC) model with a data-based kernel that exploits the manifold structure hidden in the labeled and unlabeled data. Based on the GPC model, we develop an uncertainty-based active learning scheme which queries the most informative instances to label. We also investigate the hyperparameters learning problem by maximizing the likelihood of labeled data. Promising experimental results are presented for synthetic data and real world classification problems. MA67

156