Informs Annual Meeting Phoenix 2018

INFORMS Phoenix – 2018

SD49

3 - Evaluating Alternative Models for Evaluating the Daily Deployment of Airtankers for Forest Fire Suppression David L. Martell, University of Toronto, Faculty of Forestry, 33 Willcocks Street, Toronto, ON, M5S 3B3, Canada Forest fire management agencies often use airtankers to assist with initial attack on fires and as is the case with other emergency response systems, response time is crucial. Each day the regional duty officer must decide where to deploy his or her airtankers to minimize their expected response time given the predicted fire arrival rates. Resolution of the daily airtanker deployment problem calls for the design and control of a complex spatial queueing system with time-dependant arrival rates and complex service processes. We explore the merits of using alternative simplified initial attack process models to evaluate daily airtanker deployment strategies. n SD49 North Bldg 230 Optimization and Systems Engineering Methods in Petrochemicals and Energy Sponsored: Energy, Natural Res & the Environment/Natural Resources Petrochemicals Sponsored Session Chair: Christos T. Maravelias, University of Wisconsin-Madison, Madison, WI, 53706, United States 1 - Minimum Number of Matches in Heat Recovery Networks for Energy Efficiency Heat exchanger network synthesis exploits excess heat by integrating process hot and cold streams and improves energy efficiency by reducing utility usage. Determining provably good solutions to the minimum number of matches problem is a bottleneck of designing a heat recovery network. This subproblem is an NP-hard mixed-integer linear program (MILP). We (i) explore this MILP from a graph theoretic perspective, (ii) discuss its symmetric properties, (iii) develop heuristic methods with performance guarantees, and (iv) develop a new MILP formulation without big-M parameters for special cases. Numerical results from a collection of 51 instances substantiate the methods. 2 - Solving Mixed-integer Linear Bi-level Optimization Problems through an Augmented Lagrangean Method Francisco Trespalacios, ExxonMobil Research & Engineering, Clinton, NJ, United States, Stuart M. Harwood, Dimitri Papageorgiou, Myun-Seok Cheon In this work, we present a model that demonstrates the relevance of bi-level optimization in the energy industry. Then, we present a novel method for solving mixed-integer linear bi-level optimization problems by using the Augmented Lagrangean of the subproblem. Finally, we present numerical examples that demonstrate the efficiency of this method against other methods. 3 - Integration of Crude-oil Scheduling and Refinery Planning by Lagrangean Decomposition David E. Bernal, Carnegie Mellon University, Carnegie Mellon University, Pittsburgh, PA, 15213, United States, Haokun Yang, Ignacio E. Grossmann We address the optimization of integrated refinery problem, involving crude oil scheduling and refinery operation planning. We study several modeling approaches for the Crude Distillation Unit (CDU): fixed yield, swing cuts, fractionation index. Our model, which solves both the scheduling and planning simultaneously with linking constraints for the CDU, is compared against a non- integrated approach to the crude oil scheduling and followed by the refinery planning. Furthermore, we present a Lagrangean Decomposition algorithm for this problem, whose solution time is compared versus the case that the monolithically integrated model is solved using state-of-the-art MINLP solvers. 4 - Combining the Advantages of Discrete- and Continuous-time MIP Scheduling Models Christos T. Maravelias, University of Wisconsin-Madison, 1415 Engineering Drive, Engineering Hall 2024, Madison, WI, 53706, United States, Ho Jae Lee We develop a solution algorithm, for large-scale chemical production scheduling problems, that consists of three stages: (1) solution of a discrete-time models to obtain batching, task-unit assignment, and sequencing decisions; (2) mapping of solution, preserving the three aforementioned decisions, onto time grids; and (3) solution of an LP to obtain accurate solution based on the decisions made in the first stage. We also discuss how to use two algorithmic parameters to obtain good first-stage decisions, and how to model complex processing features. We close with an extensive computational study showing the effectiveness of the proposed algorithm. Georgia Kouyialis, Imperial College London, London, United Kingdom, Dimitrios Letsios, Ruth Misener

n SD47 North Bldg 229A

Joint Session Tutorial/Practice Curated: Machine Learning and Data Mining with Combinatorial Optimization Algorithms Emerging Topic: Practice Curated Track Emerging Topic Session Chair: Scott J. Mason, Clemson University, 273 Freeman Hall, Clemson, SC, 29634, United States 1 - Machine Learning and Data Mining with CombinatorialOptimization Algorithms Dorit Simona Hochbaum, University of California-Berkeley, Dept of IEOR, 4135 Etcheverry Hall MC 177, Berkeley, CA, 94720- 1777, United States The dominant algorithms for machine learning tasks fall most often in the realm of AI or continuous optimization of intractable problems. This tutorial presents combinatorial algorithms for machine learning, data mining, and image segmentation that, unlike the majority of existing machine learning methods, utilize pairwise similarities. These algorithms are efficient and reduce the classification problem to a network flow problem on a graph. One of these algorithms addresses the problem of finding a cluster that is as dissimilar as possible from the complement, while having as much similarity as possible within the cluster. These two objectives are combined either as a ratio or with linear weights. This problem is a variant of normalized cut, which is intractable. The problem and the polynomial-time algorithm solving it are called HNC. It is demonstrated here, via an extensive empirical study, that incorporating the use of pairwise similarities improves accuracy of classification and clustering. However, a drawback of the use of similarities is the quadratic rate of growth in the size of the data. A methodology called “sparse computation has been devised to address and eliminate this quadratic growth. It is demonstrated that the technique of “sparse computation enables the scalability of similarity-based algorithms to very large- scale data sets while maintaining high levels of accuracy. We demonstrate several applications of variants of HNC for data mining, medical imaging, and image segmentation tasks, including a recent one in which HNC is among the top performing methods in a benchmark for cell identification in calcium imaging movies for neuroscience brain research. n SD48 North Bldg 229B Joint Session ENRE/Practice Curated: Wildland Fire Decision Support I Sponsored: Energy, Natural Res & the Environment Forestry Sponsored Session Chair: Yu Wei, Colorado State University, Fort Collins, CO, 80523, United States 1 - Identify and Present Large Fire Containment Strategies Yu Wei, Colorado State University, Department of FRWS, Forestry 102, Fort Collins, CO, 80523, United States, Matt Thompson Catastrophic large wildfire could threat human lives, properties and natural resources. Fire containment involves complicated decisions. We build an OR model to use stochastic fire simulation results to support large fire suppression effort. Instead of selecting only one optimal suppression strategy, our analyses provide a range of “good fire containment strategies based on a wide range of fire situation predictions, manager’s risk preferences, resource availability levels, and other management restrictions. Model results lead to alternative suppression solutions that are organized through decision-trees to support fast decisions during a fire event. 2 - A Decision Support System for Dispatching Interagency Hotshot Crews Erin Belval, Colorado State University, 1472 Campus Delivery, Fort Collins, CO, 80523, United States, Dave E. Calkin, Yu Wei, Crystal S. Stonesifer, Alex Taylor Masarie Interagency Hotshot Crews (IHCs) are an important wildland fire suppression resource. During the fire season, they drive long distances to respond to ongoing and emerging fires; previous research has indicated that this driving could be reduced. We spent the summer of 2018 working with dispatchers to refine an existing optimization model to produce 1) a real-time dispatching tool and 2) a model that uses historical data to realistically examine the impacts on IHCs of changing various policies. In this presentation we discuss the process of refining the model with dispatching input, the current state of the real-time tool, and some results from the model utilizing historical data.

107