INFORMS 2021 Program Book

INFORMS Anaheim 2021

TC15

TC13 CC Room 201A In Person: Data Analytics and Optimization General Session Chair: Ozge Surer, Northwestern University, Evanston, IL, 60201, United States Co-Chair: Tulay Flamand, Colorado School of Mines, Golden, CO, 80401-1878, United States 1 - Uncertainty Quantified Functional Curve Comparisons and its Application in Wind Energy Yu Ding, Texas A&M University, Dept Industrial & Systems Engineering, College Station, TX, 77843-3131, United States, Abhinav Prakash, Rui Tuo Wind turbine performance comparison can be framed as a data science problem, which is to compare nonparametric functional curves. This talk discusses how to quantify uncertainty for such nonparametric functional curve comparison and how to use the resulting method to track and compare performance changes of the same turbine over different periods or different wind turbines over the same period. 2 - Calibration Using Emulation of Filtered Simulation Results Ozge Surer, Northwestern University, Evanston, IL, 60201, United States, Matthew Plumlee A scalable method for calibration involves building an emulator after conducting an experiment on the simulation model. However, when the parameter space is large, the resulting simulator responses drastically differ from the observed data. One solution to this problem is to simply discard, or filter out, the parameters that gave unreasonable responses and then build an emulator only on the remaining simulator responses. In this article, we demonstrate the key mechanics for an approach that emulates filtered responses but also avoids unstable and incorrect inference. 3 - An Optimization Framework for the Optimal Investment Technology to Support the Grid Hosting Electric Vehicles’ Fast-charging Demand Harprinderjot Singh, PhD Student, Michigan State University, East Lansing, MI, United States, Farish Jazlan, Mohammadreza Kavianipour, Mehrnaz Ghamami, Ali Zockaie The rapid market growth of electric vehicles (EV) and the energy demand will affect grid stability and supply-demand balance in the electricity distribution system. Distributed energy resources (DER) (i.e., solar, battery, flywheel) can mitigate these impacts. This study proposes an optimization framework capturing EVs travel and grid requirements to support the EV demand at fast-charging stations while minimizing the total system cost, including potential grid upgrades, electricity cost, and solar and/or energy storage installation. The case study (Michigan) shows that the results are sensitive to the unit cost of DER, weather conditions, seasonal variation in solar and grid conditions. 4 - Retail Analytics for Store-Wide Shelf Space Allocation Tulay Flamand, Colorado School of Mines, Division of Economics And Bus. Engineering Hall, Golden, CO, 80401-1878, United States, Ahmed Ghoniem, Bacel Maddah We address a store-wide shelf-space allocation problem that seeks to maximize the profit from shoppers’ impulse buying. By analyzing thousands of customer baskets, we build a predictive model for in-store traffic, as a function of the space allocation and the store layout and then embed it in a non-linear mixed-integer programming model. The latter is linearized by using linear equivalent constraints and piecewise linear approximations. This helps prescribe improved store configurations and yields managerial insights for retailers. 5 - Sports Analytics for an NBA Team to Optimize Team-Building Decisions Megan Muniz, Colorado School of Mines, Golden, CO, United States, Tulay Flamand We address a team-building problem for a basketball team, where the team decides on players to draft, current players to trade and free agents to acquire. We develop a predictive model to predict the value of new players who can be drafted, and a methodology to create a new metric that encompasses the synergy potential for each player. A predictive method is also developed to predict the synergy potential between players who have not yet played together. These inform a 0-1 integer programming model for the team-building decisions that maximizes the total team value. A case study is conducted using 2018-2019 NBA data. Results show that prescribed decisions are comparable with actual decisions, and we provide insights to the General Managers based on these decisions.

TC14 CC Room 201B In Person: Optimization and Machine Learning General Session Chair: Arthur J. Delarue, MIT, Cambridge, MA, 02139-4310, United States Chair: Vassilis Digalakis, Massachusetts Institute of Technology, United States 1 - Slowly Varying Regression under Sparsity Vassilis Digalakis, Massachusetts Institute of Technology, Cambridge, MA, United States, Dimitris Bertsimas, Michael Lingzhi Li, Omar Skali Lami We consider the problem of parameter estimation in slowly varying regression models with sparsity constraints. We formulate the problem as a mixed integer program and reformulate it exactly as a binary convex program, through a novel exact relaxation that utilizes a new equality on Moore Penrose inverses. This allows us to solve it to optimality using a cutting plane type algorithm; we develop an optimized implementation of such algorithm and a heuristic method that generates warm start solutions. We show on both synthetic and real world datasets that the algorithm outperforms competing formulations in comparable times across a variety of metrics and scales to problems with 10000s of parameters. 2 - Screening Rules for Sparse Regression Andres Gomez, University of Southern California, Los Angeles, CA, 90014-3287, United States, Alper Atamturk We propose techniques to quickly reduce the number of variables in large-scale sparse regression without affecting the quality of the resulting solution. The propose methods are based on using tight convexifications of sparse regression problems to determine variables which are necessarily fixed to zero or one, and adapt formulations to find the maximum number of such variables. We illustrate with computational experiments that the resulting formulations can lead to significant speedups over alternatives that do not use screening. 3 - Convexifications for Mixed Integer Quadratic Programs Linchuan Wei, Northwestern University, Evanston, IL, 60201- 4589, United States, Simge Kucukyavuz, Andres Gomez We study the convexification of mixed-integer convex quadratic problems by decomposing the Hessian matrix Q into a sum of two-by-two positive semidefinite matrices. We give a convex hull description for the corresponding two- dimensional mixed-integer quadratic problem using the disjunctive programming approach. When a proper two-by-two decomposition is not straightforward, we formulate the problem of finding the tightest relaxation via two-by-two decomposition as a semidefinite programming problem (SDP). TC15 CC Room 201C In Person: Data Analytics in Service Operations General Session Chair: Shuai Hao, University of Illinois at Urbana-Champaign, IL, United States 1 - Lower-Tier Products: Friends or Foes? The Impact of Carpool on Ride-hailing Platforms Tingting Nian, University of California, Stern School of Business, New York, NY, 10012, United States, Vidyanand Choudhary, Jinan Lin, Rambo Tan, Cheng Gong The introduction of a new product to existing product lines typically gives rise to two opposing effects to the firm market expansion and cannibalization. In this study, we seek to understand and evaluate the causal impacts of introducing carpool rides on both riders’ and drivers’ welfare. In doing so, we use a unique dataset with fine-grained trip-level information provided by a leading ride-hailing platform, and exploit a natural experiment of the introduction of carpool rides. We are among the first studies investigating how new product introduction affects their ecosystems and revenues.

115