INFORMS 2021 Program Book

INFORMS Anaheim 2021

TD43

3 - Economic Analysis Approach to Critical Infrastructure Resiliency Investment Jeffrey Lineberry, University of Oklahoma Gallogy College of Engineering, Annapolis, MD, 21402, United States Critical infrastructure resiliency is an imperative global concern. The consideration of critical infrastructure interdependencies complicates the identification of resilience optimality. Many nations are faced with budgeting constraints and the need to optimally determine infrastructure resilience investment. The ability to identify critical infrastructure essential node vulnerability is paramount to decision makers. Determining overall economic impacts associated with critical infrastructure disruptions is a desirable approach. Real data consisting of Sweden’s rail network, power supply network, and associated economic commodity data is implemented in a tri-level model approach utilized to pinpoint vulnerability considering critical infrastructure interdependencies. This Defender-Attack-Defender model representative of vulnerability reductions, network disruptions, and recoverability enhancements is used to determine vital interdependent nodes associated with the rail and power supply networks. The analysis from this model gives insight into associated economic impacts, thus providing the framework necessary to link economic sectors to critical infrastructure interdependencies in order to determine optimal resilience investment. This model results in an overall ability to guide resilience investment based on overall economic sector considerations. 4 - Stochastic Unit Commitment Problem, An Analytical Approach Carlos Olivos, Auburn University, Auburn, AL, United States, Jorge F. Valenzuela The stochastic unit commitment problem has been modeled using different approaches such as scenario generation, chance-constrained, and robust optimization. These methods tend to provide conservative solutions resulting in higher dispatching and commitment costs. We propose an analytical formulation of the expected dispatch and commitment costs resulting from the probability distribution function of the random load. The model is linearized through a piece- wise linear approximation and solved as a Mixed Integer Linear Program (MILP). The solution is verified by simulating and computing analytically the expected cost. Results, algorithms, and conclusions will be presented. TD43 CC Room 213A In Person: Statistical Learning for Decision Analytics in Complex Systems General Session Chair: Victoria C. P. Chen, The University of Texas at Arlington, Arlington, TX, 76019-0017, United States 1 - Machine Learning Framework Fornonlinear and Interaction Relationships Involving Categorical and Numerical Features Shirish Rao, University of Texas-Arlington, Arlington, TX, 76019, United States, Victoria C. P. Chen, Jay Michael Rosenberger, Shouyi Wang, Atefe Makhmalbaf Certain applications like sustainability assessment in green building have a mix of categorical and numerical features. The relation between response and features in these applications can be highly nonlinear in behavior. Moreover, interactions between features impact sustainability metrics, and addressing interaction modeling for this mix of feature types is another challenge. While some of these challenges have been addressed individually in the literature, there is no methodology which handles these complexities simultaneously. We propose a method combining multivariate adaptive regression splines with group LASSO to screen relevant features and model terms. Using experimental design, we uncover causal understanding and show that models fitted with our methodology have improved prediction capability 2 - Lasso Based State Transition Modeling with Interactions in Adaptive Interdisciplinary Pain Management Amith Viswanatha, University of Texas-Arlington, Arlington, TX, 76013, United States, Victoria C. P. Chen, Jay Michael Rosenberger The McDermott Center for Pain Management at The University of Texas (UT) Southwestern Medical Center at Dallas provides a two-stage interdisciplinary pain management program, where a holistic, integrated approach is employed in treating patients with chronic pain to improve their pain outcomes. Patient data from the McDermott Center includes state variables related to the patient’s past and current health, treatment history, and current treatments. It is important to identify the true underlying features and the interactions between the state and treatment variables for building state transition and outcome models that are employed within a two-stage stochastic programming-based treatment optimization. In this study, we evaluate different LASSO based interaction modelling approaches on a simulated case study in identifying the true features and interactions.

3 - Fast and Reliable Metamodeling of Large-scale Nonlinear Time- dependent Problems Xinchao Liu, University of Arkansas, Fayetteville, AR, United States This research proposes a reduced-order surrogate learning framework for nonlinear structural dynamics governed by unequivocal physics principles. Motivated by the nonlinear spatio-temporal surface displacement process due to aircraft-UAV collisions, this paper shows (i) how the reduced-order physics models (including physics of motion ,fundamental material laws and finite element framework) can be obtained from the Proper Orthogonal Decomposition; (ii) how the reduced-order physics models can be accelerated by gradient boosted ensemble trees; (iii) how the input (force) uncertainty in nature is incorporated into deterministic finite element results; and (iv) how the error is controlled and modelled for governing-equation-based reduced-order models.” TE02 CC Ballroom B / Virtual Theater 2 Hybrid Location Models II Sponsored: Location Analysis Sponsored Session Chair: Zvi Drezner, California State University-Fullerton, Fullerton, CA, 92834, United States Co-Chair: Pawel J. Kalczynski, California State University-Fullerton, Fullerton, CA, 92834-6848, United States 1 - Computational Results for Primal and Dual Algorithms for the Min-Max Location Problem with Weighted Euclidean Distances in N-Dimensions Mark Cawood, Clemson University, Clemson, SC, United States, Lin Dearing Numerical results are presented for primal and dual algorithms for the min-max location problem with weighted distances. During each iteration, the algorithms use a search-path method, where the path is determined by the intersection of bisectors of a set of active points; complementary slackness is maintained for points on this path; primal feasibility is maintained for the primal algorithm; and dual feasibility is maintained for the dual algorithm. The step size is computed explicitly. Computational results are presented for each algorithm for up to 10,000 points, in up to 1000 dimensions. 2 - A Closed Form Solution for the K-centra Location Problem on an Unbalanced Binary Tree Trevor Hale, Mays Business School, College Station, TX, United States, Ryan Pepper, Faizul Huq This research delineates a heuristic that solves the k-centra location problem on a unbalanced binary tree. This treatise differs from most location research in that we employ a novel, graph theoretic based k-centra approach that seeks to find the location that minimizes the sum of the distances to the k furthest existing facilities. The k-centra problem generalizes the classic minimax and minisum location problems: If k = 2, the problem reduces to the center (minimax) location problem; whereas if k = n, the problem reduces to the median (minisum) location problem. This research has direct application to the location of a distribution center on a simple origin-to-destination distribution network to improve service levels. 3 - Obnoxious Facility Location: The Case of Weighted Demand Points Tuesday, 4:30PM-6:00PM The problem considered in this paper is the weighted obnoxious facility location in the convex hull of demand points. The objective function is to maximize the smallest weighted distance between a facility and a set of demand points. Three new optimal solution approaches are proposed. Two variants of the “Big Triangle Small Triangle” global optimization method, and a procedure based on intersection points between Apollonius circles. We also compared the results with a multi-start approach using the non-linear multi-purpose software SNOPT. Problems with 1,000 demand points are optimally solved in a fraction of a second of computer time. Atsuo Suzuki, Nanzan University, Dept of Systems and Mathematical Sciences, Nagoya, 466-8673, Japan, Pawel J. Kalczynski, Zvi Drezner

138