Informs Annual Meeting 2017

WB55

INFORMS Houston – 2017

WB52

allows us to solve a set of deterministic recourse problems sharing the same first stage action. Then, the global metamodel accounting for the errors introduced in the second-stage optimization can efficiently search for the optimal first-stage decision. Our algorithm is global convergent, and numerical study indicates it achieves substantial efficiency and accuracy. 3 - A Simulation Calibration Framework for the Production Control Wei Xie, Assistant Professor, Rensselaer Polytechnic Institute, 110 8th Street, CII, Room 5207, Troy, NY, 12180, United States, xiew3@rpi.edu A simplified stochastic simulation model can be used to guide real-time decision making for a complex real system. To support a reliable guidance, we propose a simulation calibration framework for the production control. The calibrated simulation system is able to capture the dynamic behaviors in the historical output data and predict the future outputs for the real system. The proposed framework delivers credible intervals for calibration parameters and also a prediction interval for the future output. Empirical studies demonstrate efficacy of our approach. 4 - Optimal Robust and Tolerance Design for Computer Experiments with Mixture Proportion Inputs robust and tolerance design on the mixture proportions. Traditionally, manufacturing of mixture products is controlled via interval tolerances for mixture amounts. In this paper, an optimal tolerance region for proportions, which gives optimal quality cost among all possible tolerance regions for mixture proportions with the same acceptance probability, is proposed for integrated parameter and tolerance design in mixture computer experiments. Real examples are given to demonstrate the advantages of the optimal tolerance region. 5 - An Efficient Power Network Damage Assessment via UAVs under Uncertain Environments Hoyoung Na, PhD Candidate, University of Arizona, Tucson, AZ, 85721, United States, nhy4201@email.arizona.edu, Jaeyoung Cho, Young-Jun Son This study proposes a novel meta-algorithm for efficient power network damage assessment using unmanned aerial vehicles (UAVs) in the events of power failures. A genetic programming-based hyper-heuristic (GPHH) as an offline path planner is developed to generate the paths of UAVs, which are expected to minimize the total operating cost and the damage assessment completion time under uncertain environments (e.g., wind speed and direction, remaining battery times). The effectiveness of the proposed algorithm is demonstrated for a power network system in Roanoke, VA, USA which is represented by physics-based real time simulation model in Unity3D. 6 - Fixed Budget Ranking and Selection under Input Uncertainty Di Wu, Georgia Institute of Technology, 755 Ferst Drive, NW, Atlanta, GA, 30318, United States, dwu80@gatech.edu, Enlu Zhou Ranking and selection has been studied mostly under the assumption of accurate input model. When there is input uncertainty, a direct application of traditional algorithms may result in poor probability of correct selection. We consider a new problem where a finite budget is available for collecting input data as well as running simulation. Four algorithms are developed to select the best system for single-stage and multistage scenarios. We also investigate the large deviations property in a fixed allocation asymptotic regime, and reveal that the probability of false selection still enjoys exponential decay even in the presence of input uncertainty. Mei Han, City Unversity of Hong Kong, Hong Kong, meihan2-c@my.cityu.edu.hk, Matthias Hwai Yong Tan Computer experiments often have inputs that are proportions of components in a mixture. In these mixture computer experiments, it can be of interest to perform 362B Economics Contributed Session Chair: Fernando Garcia Freitas, National Confederation of Services, Sao Paulo, Brazil, fernando.garcia.freitas@gmail.com 1 - Approximate Random Allocation Mechanisms Mohammad Akbarpour, Assistant Professor, Stanford University, Stanford, CA, United States, mohamwad@stanford.edu, Afshin Nikzad We extend the scope of random allocation mechanisms, where the mechanism first identifies a feasible “expected allocation” and then implements it by randomizing over nearby feasible integer allocations. Previous literature had shown that the cases in which this is possible are sharply limited. By introducing a new rounding algorithm, we show that if some of the feasibility constraints can be treated as goals rather than hard constraints then, subject to weak conditions that we identify, any expected allocation that satisfies all the constraints and goals can be implemented by randomizing among nearby integer allocations that satisfy all the hard constraints exactly and the goals approximately. WB55

361E Environmental Operations Contributed Session Chair: Christine Shoemaker, National University of Singapore, Singapore, Singapore, ceesca@nus.edu.sg 1 - Firm’s Multi Periodic Decision Making about Carbon Emission Reduction under Cap & Trade Yuan Baiyun, Henan Polytechnic University, 2001 Century Avenue, jiaozuo, China, yuanbaiyun@hpu.edu.cn Under the government’s Cap and Trade pilot policy, multi-period dynamic optimization about carbon emissions reduction of enterprises is discussed. Considering the policy that government limits the total amount of enterprise’s carbon emissions and allows that the carbon emissions quota can be traded among enterprises in multi-period, we study enterprise’s optimal carbon emission reduction strategy to maximize the sum of discounting profit per year by optimal control theory, and analyze how some parameters affect enterprise’s optimal carbon emission reduction strategy of each period, such as the trading prices of carbon emissions quota and initial carbon emission quotas. 2 - Automation’s Effect on Green Production: Does it Translate to Improved Competitiveness?

Kelly Weeks, Assistant Professor, Lamar University, 4400 MLK Blvd., Beaumont, TX, 77110, United States, kweeks1@lamar.edu, Mahdi Safa

This research builds upon and, as recommended, expands the work of Rao and Holt (2005). This paper will provide a longitudinal study of the impact of automation in a manufacturing setting. We will then try to determine automations impact, perhaps even inadvertently, on green production, the overall supply chain and ultimately upon competitiveness. This study provides multiple

international samples, which few predecessors, if any, have done. 3 - Parallel Surrogate Multi Objective Optimization of Computationally Expensive Problems with Environmental Applications

Christine A. Shoemaker, Distinguished Professor, National University of Singapore, office E1A 06-12, Singapore, 14853, Singapore, ceesca@nus.edu.sg, Taimoor Akhtar A “Multi-Objective Population-based Parallel Local Surrogate-Assisted” candidate search algorithm (MOPLS) is proposed for efficient and parallel multi-objective optimization of computationally expensive simulation optimization problems. Application to test problems and a 15-dimensional watershed model, and comparison against GOMORS, ParEGO and AMALGAM, shows that MOPLS is more efficient within a limited evaluation budget, especially when parallel speed- up is considered. 362A Simulation Contributed Session Chair: Di Wu, Georgia Institute of Technology, Atlanta, GA, United States, dwu80@gatech.edu 1 - Computation Methods for Simulation Optimization using Gaussian Markov Random Fields Eunhye Song, The Pennsylvania State University, 310 Leonhard Building, The Pennsylvania State University, Room C210, University Park, PA, 16802-4400, United States, eunhyesong2016@u.northwestern.edu, Mark Semelhago, Barry L. Nelson, Andreas Waechter At the core of Gaussian Process (GP) regression-based simulation optimization algorithms is a GP representing knowledge about the objective function whose conditional distribution is updated as more of the feasible region is explored. Computing the conditional distribution requires inverting a large, dense covariance matrix whose size grows as more solutions are sampled. When a GP is a Gaussian Markov Random Field, then it is particularly suited for a problem with discrete solutions since updating its conditional distribution involves a very sparse precision matrix. In this talk, we discuss how to exploit this sparse-matrix structure to reduce the computation overhead even further. 2 - A Simulation Optimization for Two Stage Decision Making Wei Xie, Assistant Professor, Rensselaer Polytechnic Institute, 110 8th Street, CII, Room 5207, Troy, NY, 12180, United States, xiew3@rpi.edu, Yuan Yi To support the dynamic decision making, we study discrete two-stage optimization with the unknown objective estimated by simulation. We introduce a global-local metamodel-assisted approach that can efficiently use the simulation resource to solve for the first- and second-stage decisions. The local metamodel WB54

493