Informs Annual Meeting 2017

TA80

INFORMS Houston – 2017

2 - Gas Storage Valuation via Least Squares Monte Carlo and Support Vector Regression Alexander Malyscheff, University of Oklahoma, 110 W Boyd Street, Devon Energy Hall 150, Norman, OK, 73019-1102, United States, malyscheff@gmail.com, Theodore B.Trafalis Least squares Monte Carlo (LSMC) approaches represent a computationally efficient method for the valuation of natural gas storage facilities. LSMC methods are computationally tractable while they simultaneously allow for a decoupling of the price path simulation from the optimization of the decision vector. We study the impact of additional pricing information in the form of a forward contract on the value of a gas storage facility. Value function approximation is carried out by employing a kernel-based regression technique in the form of support vector machine regression (SVR). 3 - Optimal Control of Tank Levels with Constrained Chance of Pipeline Shutdown In a petroleum pipeline network, when unexpected power failure or other malfunctions happens during a predetermined shipping and maintenance schedule, the unusual change of pipeline flow rate may lead to excessively high or low inventory level in tank farms, which will result in connecting pipeline shutdowns and network throughput missing. To address this problem, an optimization model is developed to study the optimal tank levels and associated pipeline flow rate adjustment policy with the objective of minimizing the missing throughput, while the chance of pipeline shutdown is bounded. The model is tested on a simplified real-world network owned by a Canadian pipeline company. 381C Modeling Interdependent Infrastructures Sponsored: Energy, Natural Res & the Environment, Energy Sponsored Session Chair: Feng Qiu, Argonne National Laboratory, Lemont, IL, 60439, United States, fqiu@anl.gov Co-Chair: Spada Matteo, Paul Scherrer Institute, Villigen PSI, 5232, Switzerland, matteo.spada@psi.ch 1 - Enhancing Resilience of Powergrid-cyber-transportation Interdependent Infrastructures Yunhe Hou, University of Hong Kong, CYC Building, Room 522, Hong Kong, Hong Kong, yhhou@eee.hku.hk This presentation discusses the coordinating strategies for enhancing the resilience of energy systems. The interaction among power network, cyber system, and transportation systems are modeled. Cyber-attack and extreme weather events trigger the disruptive event of a power system. Some sequential resilient operating strategies to absorb the impacts and to restore the system are discussed. Due to the essential interdependent characteristics of three systems and stages after the trigger events, the proposed strategies are constructed based on a multi- level optimization framework. The solution methods are discussed as well. 2 - Resilience Planning and Operations of Energy and Water Systems Neng Fan, University of Arizona, 1127 E James E. Rogers Way, PO Box 210020, Tucson, AZ, 85721, United States, nfan@email.arizona.edu In this talk, multilevel optimization model will be constructed to model the interdependency between energy and water systems. To ensure the resilience planning and operations, failure scenarios in both systems will be integrated into the model with their respective planning/operations standards. To solve the complicated model, decomposition based algorithms will be designed and validated on some test cases. 3 - Computable Equilibrium Model Analysis for Energy-water Nexus Feng Qiu, Argonne National Laboratory, 9700 S Cass Ave, Building 202, Room C205B, Lemont, IL, 60439, United States, fqiu@anl.gov, Andrew Schreiber, Jianhui Wang Many studies on energy and water are rightfully interested in the interaction of water and energy, and their projected dependence into the future. While many studies are beginning to ask the right questions, the lack of numerical rigor raises questions of concern in conclusions discerned. In this study, we will perform economic analysis using computable general equilibrium models with energy- water interdependencies captured as an important factor. We atempt to answer important and interesting questions in the studies: how can we characterize the economic choice of energy technology adoptions and their implications on water use in the domestic economy. Tianyuan Zhu, University of Calgary, Scurfield Hall, 2500 University Dr NW, Calgary, AB, T2N.1N4, Canada, zhut@ucalgary.ca TA80

4 - Modeling Recovery of Interdependent Civil Infrastructure Systems Eun Jeong Cha, Assistant Professor, University of Illinois at Urbana-Champaign, 205 N.Mathews Ave., Room 2207, Urbana, IL, 61801, United States, ejcha@illlinois.edu, Xian He Civil infrastructure systems depend on each other for product input and/or information sharing and properly modeling such dependencies is essential in assessing the performance of the civil infrastructure systems during and following a disruptive event, which lays the foundation of assessing the community resilience. This presentation introduces the Dynamic Integrated Network model which has been developed under the NIST’s Center for Risk-Based Community Resilience Planning. The model assesses the damage and recovery of a civil infrastructure network consisting of different systems following a disruptive event with considering the facility-level dependencies. 382A Distributionally Robust Optimization Sponsored: Optimization, Optimization Under Uncertainty Sponsored Session Chair: Grani Hanasusanto, The University of Texas at Austin,grani.hanasusanto@utexas.edu 1 - Two-stage Distributionally Robust Linear Programming Over Wasserstein Balls Grani Adiwena Hanasusanto, The University of Texas at Austin, We study two-stage stochastic linear programming problems where the distribution of the uncertain parameters is ambiguous and is only known to belong to a family of all distributions that are close to the empirical distribution with respect to the Wasserstein metric. We derive an exact copositive program for the generic problems and formulate a tractable linear program for instances with only right-hand side uncertainty. We illustrate the effectiveness of our reformulations in numerical experiments and demonstrate their superiority over the classical sample average approximation scheme and the state-of-the-art moment-based model. 2 - Distributionally Robust Mechanism Design Cagil Kocyigit, Ecole Polytechnique Federale de Lausanne (EPFL), Lausanne, Switzerland, cagil.kocyigit@epfl.ch, Garud N. Iyengar, Daniel Kuhn, Wolfram Wiesemann We study a mechanism design problem for auctioning a single good, where the bidders’ values are random variables that may follow any distribution in a commonly known ambiguity set. We assume that the seller is ambiguity averse and the bidders have Knightian preferences. If the bidders’ values are independent, then the added value of the optimal mechanism over a simple second price auction is offset by just attracting one extra bidder. If the bidders’ values are dependent and characterized through moment bounds, we provide a new class of randomized mechanisms, the highest-bidder-lotteries, whose revenues cannot be matched by any second price auction with a constant number of additional bidders. 3 - From Data to Decisions: Distributionally Robust Optimization is Optimal Daniel Kuhn, Ecole Polytechnique Federale de Lausanne (EPFL), EPFL.CDM.MTE I.RAO, ODY 1 01 B.(Odyssea), Lausanne, 1015, Switzerland, daniel.kuhn@epfl.ch, Bart Paul Gerard Van Parys, Peyman Mohajerin Esfahani Data-driven stochastic programming aims to find a procedure that transforms time series data to a near-optimal decision (a prescriptor) and to a prediction of this decision’s expected cost under the unknown data-generating distribution (a predictor). We propose a meta-optimization problem to find the least conservative predictors and prescriptors subject to constraints on their out-of-sample disappointment. Leveraging tools from large deviations theory, we prove that the best predictor-prescriptor pair is obtained by solving a distributionally robust optimization problem. 4 - Robust Map Inference in Markov Random Fields Areesh Mittal, University of Texas at Austin, 1626 West 6th St, Apt F, Austin, TX, 78703, United States, areeshmittal@utexas.edu, Grani Adiwena Hanasusanto In the maximum a posteriori (MAP) inference on Markov Random Fields (MRFs), we seek for the most likely assignment to the given Markov network. The underlying parameters of the MRFs, however, are typically estimated from data and can be subject to substantial errors. In this work, we introduce a new MAP inference model that endeavors to alleviate these estimation errors by leveraging on techniques from robust optimization. We show that the arising optimization model is amenable to standard integer programming solution schemes, and we assess its effectiveness via numerical experiments. TA81 ETC 5.120, Austin, TX, 78712, United States, grani.hanasusanto@utexas.edu, Daniel Kuhn

300