2016 INFORMS Annual Meeting Program

WE06

INFORMS Nashville – 2016

2 - A Convex Optimization Model For The Combined Electricity And Natural Gas Expansion Planning Problem Under Gas-price Volatility Considerations Conrado Borraz-Sánchez, Associate Postdoctoral Researcher, Los Alamos National Laboratory, Los Alamos, NM, United States, conradob@lanl.gov, Russell Bent, Pascal Van Hentenryck, Seth Blumsack, Hassan Lionel Hijazi Recent trends towards installation of gas-fired power plants have increased the economic growth and mutual dependency of electric power and natural gas industries. These industries, however, have commercial, political and technical constraints that often force them to plan, operate and manage in isolation. As a result, adverse upshots may arise such as those experienced by both systems during the winter of 2013/2014 in the US Northeast. Here, we present a joint gas- grid elastic model to optimize required expansions to meet peak demand under consideration of gas-price volatility caused by congested areas. We conduct experiments on integrated test systems that include the New England area. 3 - Electricity Capacity Expansion And Cost Recovery With Renewables Ramteen Sioshansi, The Ohio State University, Integrated Systems Engineering, 1971 Neil Avenue, Columbus, OH, 43210, United States, sioshansi.1@osu.edu, Yixian Liu High levels of renewables can suppress electricity prices, reduce revenue for all generation resources, and lead to uneconomic retirements and failure to make needed investments. To analyze this problem quantitatively, two models are utilized: (1) a multi-stage stochastic model seeking an optimal investment plan with consideration of uncertainties and operating constraints; (2) a unit commitment model giving electricity and reserve prices based on the investments. Policies such as emission restrictions and renewable subsidies are considered in this analysis. The investment model is large in scale and solved effectively by the progressive hedging algorithm. WE05 101E-MCC Wildland Fire Management II Sponsored: Energy, Natural Res & the Environment II Forestry Sponsored Session Chair: Matthew Thompson, U.S. Forest Service, 800 E Beckwith, Missoula, MT, 59801, United States, mpthompson02@fs.fed.us 1 - Filling Requests For National Interagency Hotshot Crews Erin Belval, Colorado State University, Fort Collins, CO, United States, mccowene@gmail.com, David Calkin, Matthew Thompson, Yu Wei Interagency Hotshot Crews are groups of 18 to 20 highly trained personnel used in the United States for fighting wildland fires. Dispatching criteria for these crews include minimizing travel distance and time, not taking crews away from areas with high levels of predicted near-term fire activity and balancing workloads across crews. We employ a simulation-optimization routine to examine the effects of these factors on dispatching decisions during a fire season. We compare our solutions with historical decisions using data from the Resource Ordering and Status System. 2 - Advancements In Spatial Fire Planning Matthew Thompson, U.S. Forest Service, 800 E Beckwith, Missoula, MT, 59801, United States, mpthompson02@fs.fed.us, Yu Wei, Christopher Dunn, Christopher O’Connor, Greg Dillon, David Calkin Pre-fire assessment and planning can support incident management decision making by dampening time pressures, reducing uncertainties, expanding options, and clarifying risk-benefit tradeoffs that unfold over different time horizons. This presentation will highlight the role of simulation and optimization in spatial fire planning on federal lands in the western US, with an emphasis on factors relating to cost, responder exposure, probability of success, and consequences. 3 - A Stochastic Optimization Model To Account For Climate Change In Forest Planning Jordi Garcia-Gonzalo, Centre Tecnològic Forestal de Catalunya (CTFC)., Solsona, Spain, j.garcia@ctfc.es, Andres Weintraub, Cristóbal Pais We consider a short/medium term multi-objective forest planning problem considering harvesting decisions in the presence of uncertainty due to climate change which impacts in the forest production. We introduce a multistage Stochastic model considering multiple climate change scenarios and including the corresponding non-anticipativity constraints. This enables the planner to make more robust decisions than using a single average scenario.

4 - Learning Optimal Mobility And Demand Pressure from Fire Resource Ordering Data Alex Taylor Masarie, CSU and U.S. Forest Service, Fort Collins, CO, 80523, United States, alex.masarie@gmail.com, Yu Wei, Matthew Thompson, Michael Bevers, Iuliana Oprea, Erin Belval, David Calkin An inverse partial differential equation model is used to evaluate how fire suppression resource allocation patterns vary with environmental and managerial factors. This presentation will convey the physical basis for the applied math technique demonstrating a finite difference approach to the spatial dynamics of allocation. A calibration case study will relate features of continuous optimization to operational research methodologies and present preliminary results.

WE06 102A-MCC Optimization in Data Mining 2 Sponsored: Data Mining Sponsored Session

Chair: Taghi Khaniyev, University of Waterloo, 200 University Ave., CPH 3669, Waterloo, ON, N2L 3G1, Canada, thanalio@uwaterloo.ca 1 - A Linear Separation Based MILP Model For Multi-class Data Classification Problem Fatih Rahim, Koç University, Rumelifeneri Yolu, Sarıyer, stanbul, 34450, Turkey, frahim@ku.edu.tr, Metin Turkay We address the multi-class data classification problem by a mixed integer linear programming model (MILP). We split each class’s data set into subsets such that the subsets of different classes are separable by a hyperplane. The hyperplanes that separate a subset form a polyhedral region and the regions of different classes are disjoint. A MILP model is used to find the optimal separation by minimizing the total number of regions and misclassified data points. A preprocessing step is proposed to decompose or simplify the problem considering pairwise separation of classes. We evaluated two approaches for the testing phase, based on the convex hulls of subsets and the regions defined by the hyperplanes. 2 - Robust Multicategory Support Vector Machines Using Difference Convex Algorithm The Support Vector Machine (SVM) is one of the most popular classification methods in the machinelearning literature. In this paper, we focus on classification in the angle-based framework,which is free of the explicit sum-to- zero constraint, hence more efficient, and propose two robust MSVM methods using truncated hinge loss functions. We show that our new classifiers can enjoy Fisher consistency, and simultaneously alleviate the impact of outliers to achieve more stable classification performance. To implement our proposed classifiers, we employ the difference convex algorithm (DCA) for efficient computation. 3 - Scaling For Training Set Selection In Classification Walter Dean Bennette, Air Force Research Lab, 7280 Lake View Dr, Ava, NY, 13303, United States, wdbennette@gmail.com To allow for faster and better predictions from instance based classifiers such as k- Nearest Neighbors, Training Set Selection techniques can be used to intelligently select the classifier’s training dataset. However, Training Set Selection techniques are limited in the size of datasets to which they can be practically applied. In this work scaling approaches are introduced that improve the execution time of Training Set Selection techniques. Results show that scaling methods maintain data reduction rates and achieve acceptable levels of accuracy for experimental datasets. 4 - Structure Detection In Mixed Integer Programs Taghi Khaniyev, PhD Student, University of Waterloo, 200 University Ave., CPH 3669, Waterloo, ON, N2L 3G1, Canada, thanalio@uwaterloo.ca, Samir Elhedhli, Safa F. Erenay Bordered block diagonal structure in constraint matrices of integer programs lends itself to Dantzig-Wolfe decomposition. We introduce a new measure of goodness to capture desired features in such structures. We then use it to propose a new approach to identify the best structure inherent in the constraint matrix. The main building block of the proposed approach is the use of community detection which alleviates one major drawback of the existing approaches in the literature: predefining the number of blocks. When compared against the state-of-the-art techniques, the proposed algorithm is found to identify very good structures, require shorter CPU time, and lead to comparable dual bounds. Minh Pham, Postdoc Associate, University of Virginia, 1555 Montessori Ter, Charlottesville, VA, 22911, United States, ptuanminh@gmail.com

485