2015 Informs Annual Meeting

WA13

INFORMS Philadelphia – 2015

WA11 11-Franklin 1, Marriott Optimization Large Scale II Contributed Session Chair: Lukas Bach, SINTEF, Forskningsveien 1, Oslo, 0373, Norway, lukas.bach@sintef.no 1 - Assortment Planning for Configurable Products with Consideration for Substitution Ying Tang, Graduate Research Assistant, Wayne State University, 4815 Fourth Street, Detroit, MI, 48202, United States of America, ei3512@wayne.edu, Ratna Babu Chinnam, Alper Murat, Joshua Lyon We develop assortment planning models for vehicle programs of a large automaker to support its strategic product planning efforts. We emphasize two objectives: 1) Utilizing both non-parametric and parametric approaches for characterizing demand. 2) Scalability of the models to support real-world programs. We will also present results representative of a North American vehicle program. 2 - Natural Gas System Operations and Expansion Planning for Reliability Conrado Borraz-sanchez, Associate Postdoctoral Researcher, Los Alamos National Laboratory, 1927 22nd Street Apt. D, We present a MINLP formulation to tackle natural gas network expansion planning problems. Our model captures physical, operational, directionality and on/off constraints. However, given its non-convexity, we propose a second-order cone relaxation that proves to be highly effective on large-scale cases that include existing Belgian and German networks. Comparisons against a piecewise linearization approach also show the advantages of our approximation in terms of its robustness and scalability. 3 - Adaptive Large Neighborhood Search using the Graphics Processing Unit Lukas Bach, SINTEF, Forskningsveien 1, Oslo, 0373, Norway, lukas.bach@sintef.no, Geir Hasle, Christian Schulz We investigate the efficiency of Adaptive Large Neighborhood Search on the Graphics Processing Unit (GPU). We do this by implementing the algorithm for the Distance-constrained Capacitated Vehicle Routing Problem (DCVRP), which we benchmark towards a state of the art CPU implementation. The computational power of the GPU in ordinary computers has increased significantly in recent years. Therefore it is interesting to utilize this computing power. We perform tests on well-known DCVRP instances. WA12 12-Franklin 2, Marriott Optimization Stochastic I Contributed Session Chair: Sebastian Maier, Imperial College London, South Kensington Campus, London, SW7 2AZ, United Kingdom, s.maier13@imperial.ac.uk 1 - A Hierarchical Markov Decision Process for Finding the Best Replacement Policy of Fattening Pigs Reza Pourmoayed, PhD Student, Aarhus University, Department of Economics and Business, Fuglesangs Allé, Aarhus, 8210, Denmark, rpourmoayed@econ.au.dk, Lars Relund Nielsen We use a hierarchical Markov decision process to model the sequential decision problem of replacing fattening pigs for slaughter. State of the system includes the weight and the price information acquired by two statistical models based on a Bayesian updating approach. Transition probabilities and rewards are calculated using the statistical models and a simulation method, respectively. Numerical examples are given to show the functionality of the proposed model. 2 - Quantile Optimization for Heavy-tailed Distributions using Asymmetric Signum Functions Ricardo Collado, Stevens Institute of Technology, Howe School of Technology Management, Hoboken, NJ, United States of America, rcollado@stevens.edu, Jae Ho Kim, Warren Powell We present an algorithm for computing the quantile of a continuous random variable that does not require the existence of expectation or storing all of the sample realizations. We use this to optimize the quantile of a random function satisfying some strict monotonicity and differentiability properties. We apply this to the problem of electricity trading in the presence of storage, where electricity prices are known to be heavy-tailed with infinite variance. Los Alamos, NM, 87544, United States of America, conrado.borraz@gmail.com, Pascal Van Hentenryck, Scott Backhaus, Russell Bent, Hassan Hijazi

3 - A Dynamic Size Sample Average Approximation for Stochastic Optimization Adindu Emelogu, Mississippi State University, 260 McCain Building, Mississippi State, MS, 39762, United States of America, emeloguadindu@yahoo.com, Linkan Bian, Mohammad Marufuzzaman The Sample Average Approximation (SAA) is a method of solving stochastic optimization problems by replacing the objective function with an approximation. The size of the sample affects the convergence of the solution of the approximation and the computation time. We propose an algorithm that dynamically updates the sample size in SAA and ensures both convergence and reasonable computation time. We apply our algorithm to a supply chain problem in health care, and compare it with other algorithms. 4 - Risk-averse Stochastic Path Interdiction Stephan Meisel, University of Muenster, Department of Information Systems, Muenster, Germany, stephan.meisel@uni-muenster.de, Laura Priekule, Ricardo Collado We propose a new risk-averse approach to allocating security resources in a network. Resources are allocated for blocking with high probability an attacker that selects a path for traversing the network. The attacker is characterized by an unknown probability distribution and resources are allocated based on beliefs about the distribution. We formulate the problem as a linear program and use coherent risk measures for getting solutions that are risk-averse with respect to errors in the beliefs. 5 - Appraising Interdependent Physical and Digital Infrastructure Investments: An Option Games Approach Sebastian Maier, Imperial College London, South Kensington Campus, London, SW7 2AZ, United Kingdom, s.maier13@imperial.ac.uk, David Gann, John Polak We present a new option games-based appraisal framework for selecting a portfolio of interdependent physical and digital infrastructure investments. We have used this framework to formulate a multistage stochastic optimisation model that combines the Least Squares Monte Carlo algorithm with the modelling of infrastructure interdependencies. We investigate the sensitivity of the optimised portfolio value and option exercise strategies to changes in competitor’s decisions and strategic behaviour. WA13 13-Franklin 3, Marriott Stochastic Integer Programming Methods and Applications Sponsor: Optimization/Optimization Under Uncertainty Sponsored Session Chair: Lewis Ntaimo, Associate Professor, Texas A&M University, 3131 TAMU, College Station, TX, 77843, United States of America, ntaimo@tamu.edu Co-Chair: Saravanan Venkatachalam, Texas A&M University, saravanan@tamu.edu 1 - Scaling Scenario Decomposition Methods for 0-1 Stochastic Programming Kevin Ryan, Georgia Institute of Technology, kryan31@gatech.edu, Deepak Rajan, Shabbir Ahmed A recently proposed scenario decomposition algorithm for stochastic 0-1 programs finds an optimal solution by evaluating and removing individual solutions discovered by solving scenario subproblems. We develop techniques for applying a parallel implementation of this algorithm to difficult problems with many first stage variables and a moderate number of scenarios. Challenges include problem symmetry and effective parallelization. Computational results from large scale problems are presented. 2 - Robust Multicriteria Risk-averse Stochastic Programming

Simge Kucukyavuz, Associate Professor, The Ohio State University, 210 Baker Systems Building 1971 Neil Ave, Columbus, OH, United States of America, kucukyavuz.2@osu.edu, Xiao Liu, Nilay Noyan

We study risk-averse models for multicriteria optimization problems under uncertainty. We use a weighted sum-based scalarization and consider a set of scalarization vectors to address the ambiguity and inconsistency in the relative weights of each criterion. We introduce a model that optimizes the worst-case multivariate CVaR and develop a finitely convergent algorithm for finite probability spaces. Our computational study illustrates the effectiveness of the proposed methods.

377