INFORMS 2021 Program Book

INFORMS Anaheim 2021

MB42

2 - Guaranteeing a Physically Realizable Battery Dispatch Without Charge-discharge Complementarity Constraints Nawaf Nazir, Pacific Northwest National Lab, Richland, WA, United States, Mads Almassalkhi The non-convex complementarity constraints present a fundamental computational challenge in energy constrained optimization problems. In this work, we present a new, linear, and robust battery optimization formulation that sidesteps the need for battery complementarity constraints and integers and prove analytically that the formulation guarantees that all energy constraints are satisfied which ensures that the optimized battery dispatch is physically realizable. In addition, we bound the worst-case model mismatch and discuss conservativeness. Simulation results further illustrate the effectiveness of this approach. MB42 CC Room 212B In Person: Analytics Contributed Session Chair: Divya Mehrish, CapsicoHealth Intern; Stanford University Student, CapsicoHealth, Palo Alto, CA, United States 1 - Detecting Bias in Jury Selection Using Optimal Trees Daisy Zhuo, Interpretable AI, Cambridge, MA, United States, Jack W. Dunn To support 2019 U.S. Supreme Court case Flowers v. Mississippi, there was a previous analysis using backward stepwise logistic regression to assess whether the State exhibited a racial bias in striking potential jurors. Their method is only a heuristic, and additionally cannot consider interactions between features. We apply Optimal Feature Selection to identify the globally-optimal subset of features and affirm significant evidence of racial bias. We also use Optimal Classification Trees to segment the juror population subgroups with similar characteristics and probability of being struck, and find that three groups exhibit significant racial disparity, pinpointing specific areas of bias. 2 - Visualizing the Intellectual Structure of the Impact Of COVID-19 on E-learning Hyaejung Lim, Kyungpook National University, Daegu, Korea, Republic of, Chang-Kyo Suh E-learning platforms developed enormously over time since the appearance of the Internet. However, COVID-19 pandemic made the ways of e-learning change on another level. This study intended to explore the visualization of the intellectual structure of the e-learning field concentrated on the appearance of the COVID-19 pandemic using CiteSpace(Chen, 2017). In this research, we collected the articles through the Web of Science on e-learning field related to the COVID-19 crisis. We analyze the references of the papers through author-co-citation analysis. Then, we classify the major research domains and characteristics. The results and interpretation will be further discussed in the conference. 3 - Extensions on Antminer Algorithms for Rule-based Classification Sayed Kaes Maruf Hossain, PhD Candidate, New Mexico State University, Las Cruces, NM, United States, Sajia Afrin Ema, Hansuk Sohn In this research, we have suggested multiple extensions on the AntMiner algorithms for rule-based classification. Firstly, we incorporated a strategy to dynamically balance the weight of exploration and exploitation during the rule discovery process. Secondly, we have suggested a probabilistic approach to improve the existing exhaustive rule pruning procedures. Thirdly, we performed a modular analysis to explore how the algorithm behaves for a range of probability functions. The early experimental results show competitive results for the proposed strategies over their counterparts. 4 - The Analytics of a Hybrid Workforce Edward Tuorinsky, Managing Principal, DTS, Arlington, VA, United States Missions haven’t changed, but day-to-day operations have. The pandemic is driving a modernization of the government workforce, introducing a truly hybrid model. Though the situation seems new, existing organizational data can reveal how agencies meet their mission today and provide direction for the future. This session will draw on our experience with the U.S. Fish and Wildlife Service to illustrate how human data analytics can be used to understand the impact of a hybrid workforce and better position government agencies for change. We will cover tools and techniques to capture and organize relevant data, using data visualization to make data actionable, and leveraging human data analytics.

MB40 CC Room 211B In Person: Policy-Enabling Models in the

Energy Sector General Session Chair: Afzal Siddiqui, Stockholm University, London, WC1E 6BT, United Kingdom 1 - Energy Expenditure Incidence in the Presence of Prosumers Yihsu Chen, Professor, University of California Santa Cruz, Santa Cruz, CA, United States, Makoto Tanaka, Ryuta Takashima DERs owned by prosumers are considered an effective way of fortifying grid resilience and enhancing sustainability. We analyze how their growing presence in the market may negatively affect less affluent consumers who are financially unable to adopt new technologies. Comparison of the energy expenditure incidence among different income groups when prosumers are subject to a net- metering and a net-billing policy demonstrates that policies exclusively based on volumetric consumption for recovering fixed costs are likely to favor the affluent income group. A hybrid policy, which also features an income-based fixed charge and an annual (re)connection fee or a grid access fee on prosumers, may improve energy equity by leveling the energy expenditure incidence. The policy is also more acceptable by and appealing to the utilities because of revenue certainty. 2 - Equilibrium-based Modeling of Carbon Intensity-based Standards for Transportation Fuels Several jurisdictions have enacted or proposed market-based lifecycle carbon intensity standards to promote the use of renewable transportation fuels, e.g., California’s Low Carbon Fuel Standard (LCFS). We develop a policy-oriented market model by combining multiple discrete agent (primal) optimization problems into a single mixed complementarity problem using the Extended Mathematical Programming (EMP) syntax available in GAMS. 3 - Lyapunov-regularized Reinforcement Learning for Power System Transient Stability Wenqi Cui, University of Washington, Seattle, WA, United States, Baosen Zhang Transient stability of power systems is becoming increasingly important with the growing integration of renewable resources. Their power electronic interfaces can implement almost arbitrary control laws, which provide increased flexibility in frequency responses. To design optimal non-linear policy for these controllers, reinforcement learning (RL) has emerged as a powerful method. A key challenge is to enforce that a learned controller must be stabilizing. This paper proposes a Lyapunov regularized RL approach for optimal frequency control for transient stability in lossy networks. Because the lack of an analytical Lyapunov function, we learn a Lyapunov function parameterized by a neural network. The learned Lyapunov function is then utilized as a regularization to train the neural network controller by penalizing actions that violate the Lyapunov conditions. Adam Christensen, Technical Staff, GAM S. Development Corporation, Fairfax, VA, United States, Colin Murphy, Julie Witcover, Daniel Mazzone In Person: Operation Research for Emerging Resources: Hybrid Power Plants, Virtual Power Plants, Batteries and Beyond General Session Chair: Nawaf Nazir, University of Vermont, Burlington, VT, 05401, United States 1 - Reducing Forecasting Error by Optimally Pooling Wind Energy Generation Sources Through Portfolio Optimization Alexander Vinel, Auburn University, Auburn, AL, 36832-5418, United States, Chanok Han It is widely documented that it is often possible to reduce the severity of generation intermittency by pooling together geographically diverse renewable sources. This paper aims at evaluating the potential for a similar approach targeted at addressing the related issue of limited predictability of wind energy generation. We design a portfolio optimization model based on Conditional Value- at-Risk methodology for intelligently constructing a wind energy portfolio for a given harvesting region. We then employ it to evaluate potential improvement in (day ahead) generation predictability for a collection of locations in the USA. The study concludes that if intelligent pooling is used, wind energy generation forecasting error can be significantly reduced without sacrificing much efficiency, with the effect directly related to the size of the harvesting region. MB41 CC Room 212A

55