Informs Annual Meeting Phoenix 2018

INFORMS Phoenix – 2018

TE46

2 - Leveraging Machine Learning Approach to Optimize the Participation of a Wind and Storage Power Plant Md. Noor-E-Alam, Northeastern University, 360 Huntington Ave, Boston, MA, 02021, United States, Jose L. Crespo-Vazquez, Jose A. Martinez-Lorenzo, C. Carrillo, E. Diaz-Dorado In this talk, we will introduce a robust decision making framework for a wind and storage power plant participating in day-ahead and reserve markets. At first we discuss our proposed two-stage convex stochastic model to maximize the net benefit. Then we discuss a number of machine learning techniques used to generate influential scenarios to be used in our proposed stochastic model. Finally, we will share our results obtained from several simulation experiments to evaluate the quality of the proposed stochastic approach using real-world wind farm data. 3 - Volatility, A Barrier to Renewable Energy Future Jingxing Wang, University of Michigan, 1205 Beal Ave., Ann Arbor, MI, 48109, United States, Romesh Saigal, Eunshin Byon, Abdullah Alshelahi Renewable energies, such as solar and wind, create volatilities in the power system, causing problems for the management of the power grid. The battery technology can be used to stabilize the volatile solar and wind energy power output. We use the principal multi-agent contract theory to model an aggregator who will bid the stabilized wind farm output to the ISO. 4 - A Structural Estimation Analysis of the Strategic Switching of Peaking Generators Stein-Erik Fleten, Norwegian University of Science & Technology, Dept of Ind Econ & Tech Mgmt, Trondheim, Norway, Benjamin Fram, Alois Pichler, Carl J. Ullrich We analyze the real options to shutdown, startup and abandon peak power plants. The plants’ status for a given year is either operating, in standby or retired. The analysis is made using data for 1,121 individual power plants for the period 2001-2009, and peakers in PJM 2001-2016. We estimate the irreversible costs of switching status by structural estimation of a real options model. The results on switching and maintenance costs are new to the literature. There is a markedly difference in behavior after the introduction of capacity payments in PJM. n TE45 North Bldg 228A Environment, Energy, and Natural Resources I Contributed Session Chair: Tatsuya Ishikawa, IBM Research, New York, NY, United States 1 - Modeling and Solving Conservation Management Design Problem Andres P. Weintraub, Universidad de Chile, Dept De Ingenieria Industrial, Republica 701 Casilla 86-D, Santiago, Chile, Jordi Garcia-Gonzalo, Eduardo lvarez-Miranda, Virgilio Hermoso, Jos Salgado We propose a MIP-based framework for modeling and solving a multi-action and multi-species conservation management design problem; we seek for plans that maximize ecological benefit and minimize spatial fragmentation, simultaneously, while ensuring an implementation cost no greater than a given budget. Using a case study from northern Australia, we show how the methodology exploits the trade-offs among the ecological, spatial and cost criteria, enabling to explore and analyze a broad range of conservation plans, and to select the one exhibiting the best quantitative and qualitative outcomes. 2 - New Stochastic Cell Based Fire Management Simulator Cristobal Pais, PhD Student, University of California Berkeley, Berkeley, CA, 94720, United States, Jaime Carrasco, David L. Martell, Andres P. Weintraub, David L. Woodruff Cell2Fire is a new stochastic cell-based woodland ?re spread simulator. Incorporating fuel and fire models from the Canadian Fire Behavior Prediction (FPB) System, it is designed for assessing the impact of harvesting designated forest stands on landscape ?ammability and expected losses in a heterogeneous landscape. Cell2Fire exploits parallel computations allowing users to run large- scale simulations. It can be easily integrated with harvesting management decision models in order to determine the optimal forestry planning. Real instances are tested and compared to existing state-of-the-art simulators outputs for performance comparison. Future extensions are discussed. 3 - Smart Control of Fleets of Electric Vehicles in Smart and Connected Communities

demonstrated. 4 - Real Time Localization of Objects via Passive RFID and KNN with Dynamic Neighborhood Selection Algorithm Bijoy Dripta Barua Chowdhury, University of Arizona, Tucson, AZ, 85716, United States, Sara Masoud, Young-Jun Son In this work, a novel localization framework is proposed to localize objects using a real-time, passive Radio Frequency Identification (RFID) system. After generating a statistically adequate regression model based on Received Signal Strength Indicator (RSSI), a fingerprinting database is generated to localize the tags based on dynamic k-nearest neighborhood (KNN) algorithm. The proposed framework continuously updates the models which makes it adaptable to the environmental changes. Experimental results at different scenarios demonstrate the effective performance of the proposed framework for localizing objects in dynamic environments. 5 - Sparse Modeling Approach for Optimizing Temperature-sensor Placements in Greenhouses Tatsuya Ishikawa, IBM Research, Tokyo, Japan, Kenji Komiya In greenhouse farming, controlling temperature is essential for crop growth. However, farmers often operate thermal equipment without knowing the details of temperature distributions. Deploying a large number of sensors can be a solution but is expensive and obstructive. To minimize the need for such intensive measurements, we developed a model that predicts temperature distributions from a small number of sensors. The sensors are selected using group regularization, such as the multi-task Lasso, that forces the model to reconstruct all sensor values of a pilot deployment from a subset of these values. Our experiments using real-world greenhouse data indicate the effectiveness of our approach. n TE46 North Bldg 228B Natural Gas Markets Sponsored: Energy, Natural Res & the Environment/Energy Sponsored Session Chair: Felipe A. Feijoo, Pontificia Universidad Calotica de Valparaiso, Chile 1 - The Economic Impact of Price Controls on China’s Natural Gas Supply Chain Felipe A. Feijoo, Pontificia Universidad Calotica de Valparaiso, Av. Brasil 2241, Office 6-8, Valparaiso, Chile, Bertrand Rioux, Philipp Galkin, Axel Pierru, Frederic H. Murphy, Kang Wu We developed a Mixed Complementarity Problem (MCP) model of China’s natural gas industry. The model was used to design several scenarios to assess how government pricing policies and restricted third party access to midstream infrastructure impacted the supply logistics of China’s profit maximizing natural gas firms. We find that lifting the price caps for regulated natural gas demand sectors could yield a 4.7% reduction in total system cost and reduce the national average marginal supply cost by 16%. Liberalized prices combined with improved third party access to pipeline and regasification infrastructure would result in 2.2 billion USD costs saved (7.6%) and a 16% reduction in the spot price. 2 - A Framework for Calibrating Transportation Models: A Spatial Price Equilibrium Based Method Charalampos Avraam, Johns Hopkins University, Felipe A. Feijoo, Richard Poulton, Sauleh Ahmad Siddiqui The lack of detailed cost data in the Natural Gas sector implies that there will exist more than one calibration that can reproduce a given baseline scenario for a planning model, leading to different responses under the same scenarios. Given the produced, consumed and traded quantities of all modeled regions, we propose a two - stage method for the exact calibration of such models. Stage 1 reconciles the physical system. Stage 2 uses the results of stage 1 and calibrates model parameters using a variation of Spatial Price Equilibrium that exploits the economic interpretation of the dual variables of the model. We implement our method to a transportation model of the North American Natural Gas Market. 3 - Locational Valuation of Natural Gas Subject to Pipeline Engineering Constraints Anatoly Zlotnik, Los Alamos National Laboratory, Aleksandr Rudkevich We present a pricing mechanism that maximizes social welfare for a pipeline network that delivers natural gas from suppliers to consumers. Engineering constraints on local pressures and energy applied by gas compressors are incorporated. Optimality conditions yield expressions for locational trade values (LTVs) for gas and a decomposition of LTVs into components corresponding to energy, compression, and congestion. We demonstrate that price and pressure differentials between nodes have the opposite sign, so that price cannot decline in the direction of flow, and prove that the pricing mechanism is revenue adequate.

Erotokritos Skordilis, PhD Student, University of Miami, 5202 University Dr, Coral Gables, FL, 33146, United States, Ramin Moghaddas

We present a multi-objective mixed integer-linear program for the electric vehicle (EV) charging scheduling problem. The proposed mathematical framework was developed from the perspective of optimal EV assignment to charging stations between specific time intervals. Static and dynamic model approaches were considered. As solution approaches, the augmented e-constraint method and a modified weighted-sum method were utilized. Scheduling results based on a real world charging station network and a large number of simulated EVs are

395