INFORMS 2021 Program Book

INFORMS Anaheim 2021

WB15

3 - A Sparse Expansion for Deep Tensor Markov Gaussian Processes Rui Tuo, Texas A & M University, College Station, TX, 77845-7399, United States, Liang Ding, Shahin Shahrampour Deep Gaussian Processes (DGP) enable a non-parametric approach to quantify the uncertainty of complex deep machine learning models. DGP models can suffer from high computational complexity as they require large-scale operations with kernel matrices for training and inference. In this work, we first introduce a class of Gaussian Processes, called Tensor Markov Gaussian Processes (TMGP). We then develop a deep TMGP (DTMGP) model which is the composition of multiple TMGPs. We formulate our DTMGP based on entropic optimal feature (EOF) expansion. The EOF expansion yields a highly accurate approximation and a sparse representation of DTMGP, based on maximizing the metric entropy among kernel features. Numerical experiments show the computational efficiency of DTMGP compared to other DGPs models. WB17 CC Room 202A In Person: Renewable and Rmerging Technologies General Session Chair: Alexandra M. Newman, Colorado School of Mines, Golden, CO, 80401-1887, United States 1 - Improving Fidelity of Dispatch Decisions for Concentrated Solar Power Plants Phillip Buelow, Colorado School of Mines, Golden, CO, 80401, United States Concentrated solar power (CSP) plants paired with thermal energy storage present a promising path towards developing utility-scale renewable energy. To support CSP operator decisions in a real-time setting, a revenue maximizing non- convex mixed-integer, quadradically-constrained program was developed for dispatch scheduling. Amongst commercial CSP plants, the reliability of the steam generator is the most noted issue regarding availability. Thermo-mechanical stress is a main contributor to premature leak-failure within the shell-and-tube heat exchangers (STHX). This work develops a predictive modeling tool that evaluates the thermo-mechanical stress within STHXs for off-design operations. The results from this model inform dispatch decisions such as ramping rates and maintenance forecasting within the optimization model. 2 - Real-time Dispatch Optimization for Concentrating Solar Power with Thermal Energy Storage John Cox, MS, Colorado School of Mines, Golden, CO, 80401, United States Concentrating solar power plants with thermal energy storage present a promising path towards utility-scale renewable energy. To support operator decisions in a real-time setting, we develop a revenue-maximizing non-convex mixed-integer, quadradically-constrained program which determines a dispatch schedule with sub-hourly time fidelity and considers temperature-dependent efficiency. We present exact and inexact techniques to improve tractability. Our approach admits solutions within 5\% of optimality, on average, within a five- minute time limit, demonstrating its usability for decision support in a real-time setting. 3 - Using Concentrating Solar Power Plants as Capacity Resources Ramteen Sioshansi, The Ohio State University, Department Of Integrated Systems Engineering Baker, Columbus, OH, 43210- 1273, United States, Kenjiro Yagi, Paul Denholm In this talk, we explore the use of concentrating solar power plants as capacity resources in electric power systems. WB19 CC Room 203A In Person: Improving Patient Outcomes/Learning vs Earning Trade-offs in Healthcare General Session Chair: Seungyup Lee, Vanderbilt University Medical Center 1 - Service Chains’ Operational Strategies: Standardization or Customization? Lu Kong, University of South Florida, Sarasota, FL, United States, Kejia Hu, Rohit Verma In this research, we investigate within a customer-recognizable service chain, how standardization and customization across chain-belonging units impact that chain’s performance outcomes. We study this question in the nursing home industry. Using rich archival data, we study the Degree of Standardization (DoS) in three operational dimensions: customer mix, service offering, and service delivery, and its impact on three nursing home outcomes: financial outcome, clinical outcome, and resident welfare. We also discuss the implication of our results during a public crisis such as the COVID-19.

WB15 CC Room 201C In Person: Advances in Transportation Management General Session Chair: Hamid R. Sayarshad, WPI, Worcester, MA, 01609, United States 1 - Public Transit for Special Events: Analysis, Ridership Prediction, and Train Optimization Anthony J. Trasatti, ISyE Georgia Tech, Atlanta, GA, United States, Pascal Van Hentenryck Many special events, including sport games and concerts, often cause surges in demand and congestion for transit systems. This paper proposes a suite of data- driven techniques that exploit entry-exit Automated Fare Collection (AFC) data for evaluating, anticipating, and managing the performance of transit systems during these recurring congestion peaks. Using rail data from the Metropolitan Atlanta Rapid Transit Authority (MARTA), simulations show decreased crowdedness and improved wait times for post-game ridership using proposed predictive analytics to create train schedules. 2 - Designing Intelligent Public Parking Strategies For Autonomous Vehicles’ Behaviors Hamid R. Sayarshad, Worcester Polytechnic Institute, Worcester, MA, United States With emerging technologies like autonomous vehicles (AVs), travelers do not need to park close to their destination. A bid price for the daytime parking of AVs that considers urban land use is proposed to evaluate parking strategies possibly chosen by AV users. I determine an actual parking demand function by incorporating individual preferences into a p-median problem that controls user- optimality. A novel dynamic optimization formulation is proposed to design the location of parking facilities for AVs, considering AVs’ individualized characteristics for parking such as bid price, waiting time for searching parking lots, and travel time from a set of demand nodes to the nearest parking facility. WB16 CC Room 201D In Person: Probabilistic Modeling in Predictive Analytics General Session Chair: Rui Tuo, Texas A & M University, College Station, TX, 77845- 7399, United States 1 - Probabilistic Shortest Electric Vehicle Paths: Balancing Efficiency And Reliability Ridvan Aksu, University of Alabama, Tuscaloosa, AL, United States, Mesut Yavuz Sustainability efforts and the recent developments in battery technologies are driving the popularity and the market share of Electric Vehicles (EVs) higher than ever. The main obstacles before conquering the market are limited driving autonomy and long charge times. With the state-of-the-art battery technology, we investigated pre-determining a route that minimizes the total waiting and charging costs and employing recourse to update the path in case of severe waiting times. Our analysis includes finding an initial shortest path that allows low expected waiting times and improved worst case times. 2 - Allocating Shelf Space for Fresh Produce under a Probabilistic Waste Bound Belleh Fontem, Assistant Professor, University of Massachusetts Lowell, Lowell, MA, United States, Cuibing Wu We examine a supplier duopoly serving a risk-averse retailer who faces compound Poisson demand for (initially) fresh produce that deteriorates exponentially over time. Each supplier sells a unique food item to the retailer, and upon arrival, a customer elects to purchase at most one item. Moreover, a customer’s expected demand quantity is proportional to the amount of remaining fresh stock. The retailer’s quest is a profit-maximizing allocation of shelf space subject to a probabilistic upper bound on the total space inadvertently wasted on stale inventory. We determine the retailer’s optimal allocation, and characterize the Nash equilibria arising from the suppliers’ pricing rivalry.

152