INFORMS 2021 Program Book

INFORMS Anaheim 2021

WC11

attention of optimization theorists, researchers, and industries because of its noteworthy successes in solving sequential decision-making problems. This is evident by the steady rise in publications displaying the applicability of RL for complex optimization problems in domains like operations research, supply chain, autonomous vehicles/drones, Industry 4.0, finance, health science, and many more. This presentation aims to highlight RL’s growing prominence by talking about RL’s used cases, scalability, benchmarking, the evolution of Deep RL, and future scope. Along with that, the parallel goal is to encourage researchers to work on the challenges addressed for applying RL in engineering applications. 2 - Integrating Reinforcement Learning with a Discrete Event Simulation Environment for Queueing Networks Sahil Belsare, Northeastern University, Boston, MA, 02120-2175, United States, Mohammad Dehghanimohammadabadi In this project, RL is applied to solve a M/M/C queuing system in a simulated environment. To conduct this, SimPy, a discrete event simulation (DES) library in Python, is integrated with RL algorithms. This integration provides a unique platform to link a DES environment with RL techniques and enable a new approach to solve traditional simulation-optimization problems. WC14 CC Room 201B In Person: Energy Systems Integration: Linking Platforms and Stakeholders across Systems, Scales, Chair: Jacob Garner Monroe, University of Victoria Institute for Integrated Energy Systems, Aberdeen, NC, 28315, United States 1 - Interfacing the CODERS Database with Energy System Models Using the SPINE Platform Jacob Garner Monroe, North Carolina State University, 36240 U S. Hwy 1 South, Aberdeen, NC, 28315, United States Governmental and social forces motivated by the onset of climate change have created demand for policy development that addresses decarbonization pathways. Successful decarbonization policy depends on a well built and maintained system of modeling infrastructure to support policy development efforts. Canada’s energy system modelling capacity is currently fragmented in the institutional sense and does not have a set of standard tools to assess the operational implications of decarbonization policy. Further, there has been limited effort to apply tools that connect modeling software packages for complex analyses. This research standardizes the input data necessary for energy modeling efforts and develops tools that effectively query and process that data. An open-source platform with an intuitive user interface, the Spine Toolbox, is applied together with the CODERS database to give developers a public instrument to structure, standardize and share energy systems data with work-flow process models. This study develops work-flow process models for both a production cost modeling framework (SILVER) and a generation capacity expansion model (COPPER). The Spine Toolbox weaves CODERS queries into the energy modeling frameworks as input data, then sends the output of those models to a suite of visualization software to illustrate the results. The software tools released here are built in a generic way so that the work-flow process models can be reapplied for other energy system modeling frameworks, thus increasing Canada’s modelling capacity. These models will help to bridge the modeller/stakeholder divide by enabling richer engagement sessions during the decarbonization policy development process. 2 - Virtual Microgrids: Implications for Peer-to-Peer Trading of Renewable Energy Seulchan Lee, PhD student, Texas A&M University, College Station, TX, United States, Alexandar Angelus, Chelliah Sriskandarajah A blockchain-enabled virtual microgrid has the potential to disrupt the traditional buyer-seller relationships in electricity markets. We examine the impact of virtual microgrids on electricity consumer investment on renewable energy resources: the level of investment, the resulting cost savings from virtual microgrids. 3 - A Simple Panel Data Method under Different Data Generating Processes Kang-Bok Lee, Auburn University, Auburn, AL, United States, Yeasung Jeong, Joonhwan In, Han Sumin Using an explanatory variable at time point t1 (first panel data) and at time point t2 (second panel data) with different probability distributions may cause estimation bias that stems from extrapolation, since conventional panel data methods are usually based on the presupposition that the data generating process will not change over time. In this study, we showed that the ordinary least squares method (applied to the first panel data) was not useful for estimating the dependent variables of the second panel data; thus, for simplicity of explanation, we only considered the simple panel data, which contained two time periods. To mitigate this problem, we propose a model that uses the selected informative first panel data in a systematic way by considering the importance of each data at t1 in the estimation of the dependent variable at t2. and Vectors General Session

WC11 CC Room 304C In Person: Operations Management for Urban Air Mobility General Session Chair: Zhangchen Hu, Isenberg School of Management, University of Massachusetts, Isenberg School of Management, Amherst, MA, 1003, United States 1 - UAV Path Planning under Weather Uncertainty and Environmental Impact Considerations Zhangchen Hu, University of Massachusetts Amherst, Isenberg School of Managemen, Amherst, MA, 1003, United States, Heng Chen, Senay Solak Unmanned aerial vehicles (UAVs) are expected to be widely used in the near future as an alternative transportation mode to mitigate congestion and pollution in a variety of applications. We design a dynamic and data-driven decision support system for UAV path planning through a stochastic programming based implementation, where both weather uncertainty and environmental impacts are directly considered. WC12 CC Room 304D In Person: Political Redistricting Part II of II General Session Chair: Hamidreza Validi, Rice University, Stillwater, OK, 74078, United States 1 - Partitioning a Graph Into Low-diameter Clusters with Applications to Districting Hamidreza Validi, Rice University, Houston, TX, 74078, United States, Logan Smith, Austin Buchanan, Illya V. Hicks In this paper, we study the problem of partitioning the vertices of a graph into s- clubs (s-clustering problem). An s-club is a subset of vertices for which the diameter of its induced subgraph is at most s. We propose new Mixed Integer Programming (MIP) formulations and compare them with the existing ones theoretically and computationally. Also, we develop heuristics and fixing procedures to improve the performance of our MIP formulations. Finally, we employ our s-clustering models to restrict the diameter of political districts for a US districting problem with the objective of minimizing population deviation. We implement our formulations on a large set of instances. For transparency purposes, our code and instances will be available on GitHub. 2 - Fairmandering: A Column Generation Heuristic for Fairness- optimized Political Districting David B. Shmoys, Cornell University, Ithaca, NY, 14853-3801, United States, Wes Gurnee, Wes Gurnee The USA winner-take-all congressional district system empowers politicians to engineer electoral outcomes by manipulating district boundaries. Known computational solutions focus on drawing unbiased maps, ignoring political & demographic input, and optimize for compactness. We introduce a scalable 2- stage method to explicitly optimize arbitrary piecewise-linear definitions of fairness, combining a randomized divide-and-conquer column generation heuristic, which produces an exponential number of distinct district plans, and a set partitioning IP. Our decoupled design allows for great flexibility in defining fairness-aligned objective functions. In the largest ever ensemble study of congressional districts, we use our method to understand the range of possible expected outcomes & the implications of this range on potential definitions of fairness. WC13 CC Room 201A In Person: Reinforcement Learning with Engineering Applications I General Session Chair: Mohammad Dehghanimohammadabadi, Northeastern University, Boston, MA, 02115-5005, United States Co-Chair: Sahil Belsare, Boston, MA, 02120-2175, United States 1 - The Increased Applicability of Reinforcement Learning in Engineering Applications Sahil Belsare, Northeastern University, Boston, MA, 02120-2175, United States In recent years, Reinforcement Learning (RL) has attracted the significant

160