INFORMS 2021 Program Book

INFORMS Anaheim 2021

SD38

3 - An Exact Algorithm for the Parallel Drone Scheduling Traveling Salesman Problem Using Benders-decomposition Jerimi Lee, Hankuk University of Foreign Studies, Yongin-si, Korea, Republic of, Jaegwan Joo, Youngjoo Roh, Chungmok Lee The parallel drone scheduling TSP (PDSTSP) combines the drone deliveries at the depot and the traditional vehicle routing to serve a given set of customers. Due to the limited operating range of the drones, only customers close to the depot can be served by the drones, while the remaining customers should be visited by the vehicle. We present an exact algorithm based on the logic-based Benders decomposition incorporated into a branch-and-cut framework. The computational experiments on the well-known benchmark show that the proposed algorithm outperforms the previous heuristic approaches, including the state-of-the-art MIP solvers. 4 - An Adaptive Large Neighborhood Search Method for Drone Truck Arc Routing Problem Xufei Liu, University of South Florida, Tampa, FL, United States, Sung Hoon Chung, Changhyun Kwon Arc Routing Problems (ARP) are widely used in many fields, including traffic monitoring, infrastructure inspection, and security. This talk considers ARP by a mixed fleet of drones and trucks. While trucks follow road networks, drones can fly directly between any two points on the network. With a limited flying range and battery capacity, drones need to fly from and to trucks to recharge. A metaheuristic method based on Adaptive Large Neighborhood Search (ALNS) is proposed to solve the Drone-Truck ARP. The performance of ALNS is evaluated using randomly generated ARP instances. SD38 CC Room 210D In Person: Commodity and Energy Market Operations General Session Chair: Bo Yang, Carnegie Mellon University, Pittsburgh, PA, 15213- 4226, United States 1 - The Term Structure of Optimal Integrated Hedges Danko Turcic, University of California, Riverside, Graduate School of Business, CA, 92521-9800, United States We show how a commodity processor facing stochastic demand and stochastic selling price has a capacity and lead-time preference and how that preference can be advantageously manipulated with hedging. The results apply in industries in which firms’ revenues are significantly affected by movements in commodity prices. 2 - Pathwise Reinforcement Learning for Informationally Rich Models: Coordinated Decomposition and Regression Bo Yang, Carnegie Mellon University, Pittsburgh, PA, 15213-4226, United States, Selvaprabu Nadarajah, Nicola Secomandi Pathwise reinforcement learning (PRL) has been used to obtain high quality bounds and control policies for Markov decision processes with rich information structures. Beyond optimal stopping, the state of the art for solving underlying linear program is a block coordinate descent (BCD) procedure that exhibits high per iteration computational complexity. We propose a coordinated decomposition methodology with improved complexity that (i) finds a solution to the dual of the sampled LP and (ii) recovers a primal solution by approximately enforcing complementary slackness via regression. We conduct a numerical study in the context of merchant energy production. Compared to BCD, our technique can solve both existing instances more efficiently and with similar accuracy and new larger size ones that are out of reach for this alternative, achieving near optimal performance. SD39 CC Room 211A In Person: Healthcare Operations and Technology Management General Session Chair: Minje Park, Boston University, Boston, MA, 02215-1704, United States 1 - Multi-channel Referrals and Patient Outcomes Sokol Tushe, Emory University, Atlanta, GA, 30322-1059, United States, Diwas S. Kc A physician consultation has traditionally required the collocation of the physician and the patient. However, the wide adoption of telemedicine creates multiple channels for delivering healthcare, including core processes such as patient diagnosis. We investigate how primary care physicians adjust their referral behavior when they can choose to refer patients to specialists through an in- person channel or an online channel. Specifically, we look at how patient classes are separated into different referral channels based on their complexity level. We

also study the implications for patient outcomes. 2 - Impact of Pharmaceutical Supply Chain Disruptions on Medication Safety: Synthetic Control Method Approach Minje Park, Boston University, Boston, MA, 02215-1704, United States, Anita L. Tucker, Rena Conti We investigate the impact of pharmaceutical supply chain disruptions on medication safety by studying a drug shortage case caused by Hurricane Maria in 2017. By applying the synthetic control method (Abadie et al. 2003, 2010), we measure the increase in medication errors and adverse drug events after the supply disruption. With our results, we provide implications for safe substitution between drugs during the supply disruption period. SD40 CC Room 211B In Person: Issues in Energy Market Design, Regulation, and Evolution General Session Chair: Ramteen Sioshansi, The Ohio State University, The Ohio State University, OH, United States 1 - Can an Energy-only Market Design Yield Electricity Decarbonisation? Insights from a System Dynamics Approach Olivier Massol, IFP School, Paris, France, Alexis Lebeau, Marcelo Saguan, Yannick Perez In contemporary power systems, an important policy issue is whether an energy- only type of market design (EOM) is capable to yield a transition towards net carbon neutrality. In this research, we adopt a simulation framework and propose a system dynamics representation to investigate the two follwoing questions: (1) what assumptions about investor behavior and available information are needed to ensure that an EOM achieves the desired decarbonization trajectory and the desired target mix?; (2) How robust is an EOM (as measured by deviations between realized vs. optimal mix trajectories) when different assumptions are considered? Our results extend the standard analyses by stressing the crucial importance of a series of conventionally admitted assumptions (e.g., the role of perfect foresight, that of full information and the agents’ type of rationality). 2 - Multi-period Pricing under Price History Dependent Investments in Consumption Infrastructure: An Application in Natural Gas Sector Baturay Calci, The University of Texas at Austin, Austin, TX, 78751-5031, United States, Benjamin D. Leibowicz, Jonathan F. Bard, Gopika Jayadev We build a bilevel model of the interaction between two agents where the leader sets prices over the planning horizon, and then the follower determines the investments at each period that set the future demand based on the price history until that period. This framework is applied to a natural gas producer (leader) and an electric utility company (follower) which decides investments in natural gas- fired power generation infrastructure based on past average gas prices. There is a trade-off in the leader’s problem between high prices (high current revenue) and low prices (high future revenue due to investments). Preliminary results are presented as well as the formulation and solution approach. 3 - Data-driven Piecewise Linearization for Distribution Three-phase Stochastic Power Flow Jiaqi Chen, University of Wisconsin–Madison, Madison, WI, United States, Wenchuan Wu, Line A Roald As the penetration of distributed renewable energy increases, stochastic power flow (SPF) becomes an essential tool to analyze the uncertainties in active distribution networks. The Monte Carlo (MC) method is the most straightforward and accurate technique to calculate the three-phase SPF. However, the computation burden of the MC method is significant since it involves numerous calculations of three-phase nonlinear AC power flow. This talk will introduce a piecewise linear, data-driven power flow approach for the MC-based three-phase SPF calculation. An improved K-plane regression algorithm is proposed while considering the collinearity of the training data. We demonstrate that the proposed SPF approach can handle complex operational conditions such as the correction of random variables and three-phase unbalance with high accuracy and efficiency.

42