2016 INFORMS Annual Meeting Program

MC17

INFORMS Nashville – 2016

2 - Scheduling Commercial Spots Using Mixed Integer Programming Vivek Vasudeva, Turner Broadcasting System, Inc.,

3 - Distribution Sensitivities For Quantile, Distortion Risk Measure, And Inference Yijie Peng, Fudan University, pengy10@fudan.edu.cn Michael Fu We treat quantile sensitivity, sensitivity of distortion risk measure, and statistical inference under a single umbrella of distribution sensitivities. A new stochastic derivative estimation technique called generalized likelihood ratio method is proposed to address three applications in a uniform manner. We illustrate advantages of the proposed method over existing stochastic derivative estimation techniques for distribution sensitivities estimation, and provide supporting numerical evidences. 4 - Small Data Optimization Vishal Gupta, University of Southern California, guptavis@usc.edu, Paat Rusmevichientong Notwithstanding press about “Big Data,” many real-world problems exhibit both a large number of uncertain parameters and a small amount of data per parameter. We propose a novel approach to linear optimization in this “small-data regime” inspired by empirical Bayes methods. Our approach uses the large-scale structure to circumvent the insufficient data; as the size of the optimization tends to infinity while the amount of data remains fixed, our approach performs comparably to an oracle best-in-class policy. Other popular methods do NOT enjoy this property. Empirical evidence confirms that our approach significantly outperforms state-of- the-art data-driven methods in this small-data regime. MC17 105B-MCC Robust Optimization Sponsored: Optimization, Optimization Under Uncertainty Sponsored Session Chair: Anirudh Subramanyam, Carnegie Mellon University, 5000 Forbes Avenue, Pittsburgh, PA, 15213, United States, asubramanyam@cmu.edu 1 - Exploiting The Structure Of Two-stage Robust Optimization Models With Integer Adversarial Variables Seyed Hossein Hashemi Doulabi, Polytechnique Montréal, This paper addresses a class of two-stage robust optimization models with integer variables in the adversary’s problem. We apply Dantzig-Wolfe decomposition to exploit the structure of these models and show that the original problem reduces to a single-stage robust problem. We propose a Benders algorithm for the reformulated problem. Since the master problem and subproblem in the Benders algorithm are mixed integer programs it is computationally burdensome to optimally solve them in each iteration of the algorithm. Therefore, we develop novel stopping conditions for these mixed integer. Some computational experiments are performed on a two-stage nurse planning problem. 2 - A New Algorithmic Framework For Two-stage K-adaptable Robust Optimization Problems With Mixed-integer Recourse Anirudh Subramanyam, Carnegie Mellon University, Pittsburgh, PA, 15213, United States, asubramanyam@cmu.edu Wolfram Wiesemann, Chrysanthos Gounaris We present a new algorithm for solving K-adaptability versions of two-stage robust mixed-integer linear programs (MILPs), in which we commit to K recourse policies here-and-now and implement the best policy once the uncertain parameters are observed. Viewing such problems as semi-infinite disjunctive MILPs, our framework is able to address mixed-integer and random recourse in K-adaptability problems for the first time. It is also able to incorporate tailored solution approaches for the corresponding deterministic problems and decomposition techniques widely used in stochastic programming. We conduct extensive numerical experiments on benchmark data from a number of popular applications. 3 - Weekly Two Stage Robust Generation Scheduling For Hydrothermal Power Systems Hossein Dashti, University of Arizona, hdashti@email.arizona.edu, Antonio J. Conejo, Ruiwei Jiang, Jianhui Wang As compared to short-term forecasting, it is challenging to accurately forecast the volume of precipitation in a medium-term horizon. As a result, fluctuations in water inflow can trigger generation shortage and electricity price spikes in a power system with major hydro resources. In this work, we study a two-stage robust scheduling approach for a hydrothermal power system. We consider water inflow uncertainty and employ a vector autoregressive (VAR) model to represent its seasonality and construct an uncertainty set in the robust optimization approach. We design a Benders’ decomposition algorithm to solve the problem. Results are presented for the proposed approach on a real-world case study. Montreal, QC, Canada, hashemi.doulabi@polymtl.ca Patrick Jaillet, Gilles Pesant, Louis-Martin Rousseau

Atlanta, GA, United States, vivek.vasudeva@turner.com, Jose Antonio Carbajal Orozco, Wassim (Wes) Chaar

A broadcasting company sells airtime to advertisers to air their commercial spots in accordance with certain agreed-upon rules. We examine a mixed integer programming-based formulation that can be used to schedule spots such that a desired measure can be optimized while honoring the deal rules. We also analyze how the measure changes as we relax different sets of constraints, to gain insights into the relative impact of these constraints on the optimization measure.

MC15 104E-MCC Disaster and Emergency Management I Contributed Session 1 - Apropos Resilient System Design

Henry Lester, University of South Alabama, P.O. Box 8172, Mobile, AL, 36689, United States, hlester@southalabama.edu System resiliency signifies the ability to resist and recover from an extreme event. Critical measures of this resiliency are extreme event damages and disaster recovery time. This paper presents an analytical approach to capturing extreme event system behaviors with respect to system resiliency in order to predict disaster recovery time. The approach isolates significant system factors to estimate recovery function parameters to enhance a resilient system design. The resultant resilient system design can provide situational awareness for future decision analysis pertaining system deployment and operations while subject to extreme events. 2 - A Follow-Up Sharing Method For Post-Event Response Resource Distribution With Group Information Updates Yong Ye, Wenzhou Medical University, Chashan Street, Ouhai District, Wenzhou, 325035, China, yong_ye@foxmail.com, Guiling Liu, Lingle Pan This paper addresses a Follow-up Sharing Character (FSC), which coordinates resources between different phases. Based on FSC, this paper proposes a general model by minimizing the RAEL (the losses caused by the mismatch between supply and demand in impacted areas) of all phases, the RAEL of all affected areas in the present phase, and the ELTL of the distribution plan in the present phase. We also apply the Bayesian information updates approach to deal with uncertainties of demand and traffic condition, by using historical and sample information. Then, a solution algorithm is proposed to solve the model; and a simulation study is presented. Insights derived from the model are provided in the conclusion. MC16 105A-MCC Data-Driven Optimization Methods Sponsored: Optimization, Optimization Under Uncertainty Sponsored Session Chair: Vishal Gupta, University of Southern California, University Park Campus, Los Angeles, CA, 90089, United States, guptavis@usc.edu 1 - Machine Learning & Portfolio Optimization Gah-Yi Ban, London Business School, gban@london.edu, Noureddine El Karoui, Andrew Lim We adapt two machine learning methods, regularization and cross-validation, for portfolio optimization. First, we introduce performance-based regularization (PBR), where the idea is to constrain the sample variances of the estimated portfolio risk and return. We consider PBR for both mean-variance and mean- CVaR problems. We show that the PBR models can be cast as robust optimization problems with novel uncertainty sets and establish asymptotic optimality of both Sample Average Approximation (SAA) and PBR solutions and the corresponding efficient frontiers. We also develop new, performance-based k-fold cross- validation algorithms. 2 - Smart Predict Then Optimize Paul Grigas, UC Berkeley, 4177 Etcheverry Hall, University of California, Berkeley, CA, 94720-1777, United States, pgrigas@berkeley.edu, Adam Elmachtoub We consider a class of optimization problems where the objective function is not explicitly provided, but contextual information can be used to predict the objective based on historical data. A traditional approach would be to simply predict the objective based on minimizing prediction error, and then solve the corresponding optimization problem. Instead, we propose a prediction framework that leverages the structure of the optimization problem that will be solved given the prediction. We provide theoretical, algorithmic, and computational results to show the validity and practicality of our framework.

187