INFORMS 2021 Program Book

INFORMS Anaheim 2021

SB36

3 - Capacity Scaling Augmented with Unreliable Machine çLearning Predictions Daan Rutten, Georgia Institute of Technology, Atlanta, GA, çUnited States Modern data centers suffer from immense power consumption. As a result, data center operators have heavily invested in capacity scaling solutions, which dynamically deactivate servers if the demand is low and activate them again when the workload increases. We analyze a continuous-time model for capacity scaling, where the goal is to minimize the weighted sum of flow-time, switching cost, and power consumption in an online fashion. We propose a novel algorithm, called the Adaptive Balanced Capacity Scaling (ABCS) algorithm, that has access to black-box machine learning predictions. ABCS aims to adapt to the predictions and is also robust against unpredictable surges in the workload. In particular, we prove that the ABCS algorithm is (1 + )-competitive if the predictions are accurate, and yet, it has a uniformly bounded competitive ratio even if the predictions are completely inaccurate. 4 - Presenter Siddhartha Banerjee, Cornell University, 229 Rhodes Hall, Ithaca, NY, 14853-3801, United States, Sean Sinclair SB34 CC Room 209B In Person: Simulation-I General Session Chair: Raghu Pasupathy, Purdue University, West Lafayette, IN, 47907- 2067, United States 1 - Encouraging Modal Shift Through Green Technologies: çA Multiobjective Simulation Optimization Approach Sebastian Rojas Gonzalez, Postdoctoral Research Fellow, Hasselt University, Hasselt, Belgium, and Postdoctoral Research Fellow, Ghent University, Ghent, Belgium, Maximiliano Udenio, Hamed Jalali, Inneke Van Nieuwenhuyse In this work we consider the problem of inland multimodal transportation of perishable goods involving multiple stakeholders (ports, carriers, and shippers), whose interests are in conflict. We model the problem as a multiobjective optimization, where both financial and sustainability considerations drive the decision making. In particular, we investigate the effect of introducing a green technology for inland barge transport, aimed at increasing modal shift from trucks to barges by lowering the carbon footprint of the latter. As the problem is highly stochastic and analytically intractable, we propose to use a novel multiobjective simulation optimization algorithm to seek for solutions that not only reveal the essential trade-offs, but also account for the intrinsic noise in the observed performance. 2 - Confidence Sets for Parameters Estimated from Time Series Raghu Pasupathy, Purdue University, Department of Statistics, West Lafayette, IN, 47907-2067, United States, Peter W. Glynn We present a procedure for constructing confidence sets using time series data on a “parameter” residing in a metric space. Application contexts include quantile field estimation, nonhomogeneous Poisson process rate estimation, parameter estimation for an ARMA(p,q) process, and stochastic optimization. Since dependence is an important complication to contend, the main instrument that enables the confidence set construction is batching. We detail several set estimators whose weak convergence to what we call OB Type statistics depends on the nature and extent of data batching. In demonstrating the approach’s breadth, we include numerical illustrations on constructing confidence sets in diverse contexts. We also present tables for OB Type distributions analogous to the classical Student’s t distribution. SB35 CC Room 210A In Person: Advances in Material Handling/Stochastic Optimization General Session Chair: Ali Tolooie, Manhattan, KS, 66502, United States 1 - Combining Predictive and Prescriptive Techniques for Optimizing Electric Vehicle Fleet Charging Ehsan Mahyari, Ph.D. Student in Operations Management, The University of Alabama, Tuscaloosa, AL, United States We develop a rolling-horizon approach that combines predictive models and traditional optimization techniques for minimizing the charging cost of an electric vehicle fleet. We use predictive models to account for the uncertainty in vehicle arrivals. The underlying mathematical programming model is NP-hard. Thus, we

develop a simple heuristic to generate initial feasible solutions for a problem instance and use the heuristic to warmstart the mathematical programming model. Experiments based on data for a real mass transit fleet suggest that our approach offers benefits with respect to cost and grid sustainability. 2 - Stochastic Models for Optimizing Unmanned Aerial Vehicle Delivery on Last-mile Logistics Ali Tolooie, Kansas State University, Manhattan, KS, United States, Ashesh Kumar Sinha We propose a two-stage stochastic mixed-integer programming model to design a reliable and efficient supply chain network. The proposed network includes charging stations to extend the delivery coverage of drones. We handle stochasticity in the problem by developing Markov decision process models that evaluate tradeoffs between the number of batteries and drones in the last-mile logistics system. To overcome difficulties computationally, we propose different novel decomposition-based approaches for each problem to provide an exact analysis for our logistics network. SB36 CC Room 210B In Person: Using Health Data to Inform Decision Making General Session Chair: Pooyan Kazemian, Case Western Reserve University, Cleveland, OH, 44114, United States 1 - Mining Temporal Patterns for Prediction a Mixed-integer Programming Approach Farzaneh Mansourifard, Oregon State University Temporal patterns are sets of feature abstractions sequenced by the time that can be used as variables in prediction models. The current state of the art is to select those pattern defining parameters independent to the construction of the prediction model. In this work, we propose a mixed-integer programming framework to determine optimal pattern defining parameters and prediction model coefficients simultaneously. 2 - Site Reassignment for Mobile Outreach Teams: Investigating the Effectiveness of Decentralized Decision-making Lisanne van Rijn, PhD Candidate, Erasmus School of Economics, Rotterdam, Netherlands, Harwin de Vries, Luk N. Van Wassenhove To improve access to healthcare, mobile outreach teams of healthcare workers visit remote sites to provide healthcare services. Dynamics in demand and supply cause once rational site-to-team assignment decisions to become suboptimal. This paper considers the problem to reassign sites to maximize effectiveness. Outreach teams commonly have much decision-making autonomy, but reassignment requires coordination. To study whether and when a decentralized approach is effective, we examine the trade-off between centralization and effectiveness and study how design choices and information gaps induced by centralization affect this trade-off. We use empirical data from six country outreach programs of NGO MSI Reproductive Choices. Our results suggest that simple decision-making systems, when properly designed, tend to perform close to centralized decision- making. 3 - Periodic Vaccination Against SARS-CoV-2: Some Projections for the United States Jade Xiao, Georgia Institute of Technology, Atlanta, GA, United States, Turgay Ayer, Jagpreet Chhatwal With the U.S. nearing the end of its inaugural wave of COVID-19 vaccinations, public health authorities are turning their attention to post-pandemic management. The SARS-CoV-2 virus is expected to become endemic. Given that waning immunity to the virus is highly probable, periodic mass vaccination will be necessary for continual outbreak prevention. However, the exact duration of immunity conferred by both vaccines and natural infection is still unknown, making it difficult at present to plan revaccination efforts. We model different periodic vaccination strategies within an SEIR framework based on the COVID-19 Policy Simulator (www.covid19sim.org) to assess their effect on disease burden over the next several years. This study provides qualitative conclusions to aid policymakers in post-pandemic management of COVID-19.

11