Informs Annual Meeting Phoenix 2018

INFORMS Phoenix – 2018

TD01

10 - Optimal Bidding for Highly Valued IT Service Contracts Theoretical Results and Practical Implications Xiangyu Zhang, Cornell University, Ithaca, NY, 14850, United States Information technology service providers compete to win highly-valued IT service contracts in a tender process. Prior literature shows that features other than price, including the service provider’s relationship with the client, contribute to the client’s selection because good relationship increases the provider’s chance of winning the deal. Thus, it might be beneficial for the provider to lower their price for improving the client relationship and increasing their potential future contracts profits. In this work, we provide theoretical and numerical results illustrating the optimal price is lower than the myopic price which tries to maximize the expected profit of the current deal. 11 - A New Collision Avoidance Algorithm for Autonomous Vehicles Considering Physical Constraints Elnaz Torkamani, Rutgers University-New Brunswick, Piscataway, NJ, United States, Zhimin Xi Vehicle navigation autonomously in a dynamic environment is a challenging task because it should not be pre-programmed under a given handful situations and the vehicle must be able to avoid collisions under numerous unforeseeable but reasonable situations to a human being. From the practical perspective in autonomous cars, the algorithm must be efficient and reliable. This study proposes a new vehicle collision avoidance algorithm for multiple obstacles by eliminating the need of running sampling or optimization approaches, so that the algorithm could be practical useful in real-time collision avoidance under high vehicle speed. 12 - Reliability Based Optimal Design of a Microgrid System under Natural Disasters Zhetao Chen, Rutgers University-New Brunswick, Piscataway, NJ, United States, Zhimin Xi This study proposes reliability-based optimal design of a micro-grid system under service disruptions due to natural disasters. The objective is to determine the minimum number of generators and their distributions in the micro-grid so that the system’s recoverability can be guaranteed under random failure scenarios of the power transmission lines. Power flow analysis combing with the Monte Carlo simulation (MCS) are used for uncertainty propagation analysis to quantify the system’s recoverability distribution under random failure scenarios of the transmission lines. The proposed work is demonstrated through a 12-bus power system. 13 - Characterization of Battery Model Uncertainty for Effective State of Charge Estimation Modjtaba Dahmardeh, Rutgers University, Piscataway, NJ, United States, Zhimin Xi SOC estimation accuracy of batteries has been extensively studied through the development of various battery models and online dynamic estimation algorithms. All battery models, however, contain inherent model bias due to the simplifications and assumptions, which cannot be effectively addressed by the already established algorithms like EKF. Consequently, as observed in our study, battery SOC estimation using a typical EKF is not very accurate depending on the battery characteristics. This poster reports great potential for improving SOC estimation accuracy by proposing two bias characterization methods (i.e. polynomial and GP regression) without the complexity of novel algorithms. 14 - Optimal Smoothed Variable Sample-size Accelerated Proximal Methods Afrooz Jalilzadeh, Pennsylvania State University, University Park, PA, 16801, United States, Uday Shanbhag We consider a class of structured nonsmooth stochastic convex programs. Traditional stochastic approximation schemes are generally characterized by far poorer convergence rates compared to the deterministic counterparts. In this study, we consider variable sample-size schemes in which the bias in the sampled gradient is reduced by taking increasingly larger sample-sizes. We prove that the deterministic rates can indeed be attained while maintaining an optimal complexity in the number of sampled gradients. Moreover, optimal rates are obtained by combining an iteratively smoothing framework with this variable sample-size structure. 15 - Modeling the Impact of Large-scale Disruptions on Supply Chain Networks Ni Ni, Rensselaer Polytechnic Institute, Troy, NY, 12180, United States, Thomas Sharkey We provide optimization models to capture the impact of disruptions to different segments of supply chains. For disruptions to the production segment, we build two-stage stochastic programming models to optimize the pre-event and post- event strategies in terms of the impact of unmet demand of customers. The insights include a æchain reaction’ among recovery efforts in the production segment. For disruptions to the distribution segment, we create models to predict the cascading failure on supply chains based on damage to infrastructures and restoration models to provide the importance of coordination between efforts for elements in the distribution segment.

16 - Lignocellulosic Biomass to Biofuel Supply Chain Optimization with Mobile Densification and Farmers’ Choices Nibal Albashabsheh, Kansas State University, Manhattan, KS, 66502, United States, Jessica Heier Stamm We present an optimization model for lignocellulosic biomass-to-biofuel supply chains that considers mobile densification and farmers’ choices, factors not considered in prior studies. Case study results based on Kansas data indicate that omitting mobile densification increases logistics costs, and ignoring farmers’ choices results in significantly overestimating available supply. 17 - OD Flow Prediction with Vehicle Trajectory Data and Semi-supervised Recurrent Neural Network Yintai Ma, DiDi Research America, Mountain View, CA, 94043, United States, Tao Huang, Zhiwei Qin, Jianfeng Zheng With the emergence of smartphone usage and ride-sharing economy, a large amount of trajectory data of travelers are produced every day, which can be and should be used to leverage the OD flow estimation and prediction on a complex OD network. In this work, We propose a deep neural network to make the prediction of OD flows of an urban network solely based on the trajectory data of DiDi Drivers in both supervised and semi-supervised setting. Our model reaches median MAPE around 25% for peak hours. Our experiments indicate that the app-recorded trajectory data contains essential information of traffic status, encouraging future exploration to use app-recorded trajectory data for traffic patterns studies. 18 - Investigating Autonomous Air Operations Centers for On-demand Mobility Networks Victoria Chibuogu Nneji, PhD Candidate, Duke Robotics, Durham, NC, United States Several manufacturers are developing new vehicles for high-speed intra-city air taxi on-demand mobility (ODM) services. With diverse vehicle design requirements, how might airline operations centers need to innovate to support these new fleet demands We used task- and time-based data from a collective case study of dispatchers to develop a discrete event simulation to serve as a predictive model of human-system performance in these air operations centers. With this tool, companies can rapidly prototype future ODM concepts of operations to make better informed strategic, tactical, and operational decisions in staffing and designing centers. 19 - Reliable Routing of Road-rail Intermodal Freight under Uncertainty Majbah Uddin, University of South Carolina, Columbia, SC, 29210, United States, Nathan Huynh Transportation infrastructures, particularly those supporting intermodal freight, are vulnerable to natural disasters and man-made disasters that could lead to severe service disruptions. These disruptions can drastically degrade the capacity of a transportation mode and consequently have adverse impacts on intermodal freight transport and freight supply chain. To address service disruption, this paper develops a model to reliably route freight in a road-rail intermodal network. The proposed methodology is demonstrated using a real-world intermodal network in the Gulf Coast, Southeastern, and Mid-Atlantic regions of the United States. n TD01 North Bldg 121A Decision Making with Noisy Information Sponsored: Optimization/Optimization Under Uncertainty Sponsored Session Chair: Bo Zeng, University of Pittsburgh, Pittsburgh, PA, 15260, United States 1 - Incorporating Clustering in Decision Making with Noisy Information Chaosheng Dong, University of Pittsburgh, 4200 Fifth Avenue, Pittsburgh, PA, 15260, United States, Bo Zeng Clustering is the problem of partitioning unlabeled data points into a number of clusters so that data points within one cluster are homogeneous. In this study, we show that clustering will increase the robustness of the decision made with noisy information. Meanwhile, it will dramatically improve the efficacy of the decision- making process. Numerical results on both synthetic data and real-world problem show the merit of our method. Tuesday, 2:00PM - 3:30PM

345