Informs Annual Meeting 2017

TB62

INFORMS Houston – 2017

TB62

TB63

370C Spatio-Temporal Models for Forecasting Sponsored: Quality, Statistics and Reliability Sponsored Session Chair: Yu Ding, Texas A&M University, College Station, TX, 77843- 3131, United States, yuding@tamu.edu Co-Chair: Ahmed Aziz Ezzat, Texas A&M University, College Station, TX, United States, aa.ezzat@tamu.edu 1 - Smart Energy: A Modern Data-Driven Perspective Andrew Kusiak, The University of Iowa, Iowa City, IA, 52242, United States, andrew-kusiak@uiowa.edu The world of energy is transforming towards renewable energy generation with wind energy experiencing the fastest growth. Renewable energy has introduced a shift from centralized to distributed generation. The cost to operate and maintain renewable energy plants is a major determinant of energy prices. Autonomy of renewable energy generators and plants offers a viable path to competitive prices of clean energy. Using wind industry as an example, problems awaiting research attention and solution approaches are outlined. A strategy to democratization of energy research aiming at accelerated response to meeting wind energy needs is presented. 2 - Distributed Ensemble Solar Power Forecasting through the Generalized Dynamic Factor Model Duehee Lee, Assistant Professor, KonKuk University, Kwangjin Gu, Seoul, Korea, Republic of, hello.keanu@konkuk.ac.kr, Ross Baldick, Jaehyung Roh Total daily solar irradiance for the next day is forecasted through an ensemble of multiple machine learning algorithms using forecasted weather scenarios from numerical weather prediction models. We also forecast the distributed solar power through the generalized dynamic factor model. Six interpolation functions are used to interpolate weather scenarios at non-integer grid points and performances are compared. The forecasting models are verified using data from the American Meteorological Society 2014 Solar Energy Prediction competition by observing the post-competition rankings. We achieved 16th place out of 160 in this competition, and our best post-competition ranking is 5th place. 3 - A Markov Switching Vector Autoregressive Stochastic Wind Generator for Multiple Spatial and Temporal Scales Amanda Hering, Baylor University, Waco, TX, United States, Mandy_Hering@baylor.edu Stochastic realizations of wind are important for wind energy grid integration and reliability studies. Here, we introduce a Markov-Switching Vector Autoregressive model (MSVAR) model and demonstrate its flexibility in simulating wind vectors for various temporal and spatial scales. In addition, we demonstrate how the model can be used to simulate wind vectors at multiple locations simultaneously. The parameter estimation and simulation algorithm is presented along with a validation of important statistical properties of each simulation scenario. The MSVAR is very flexible in characterizing a wide range of properties in the wind vector. 4 - Short-term Wind Forecasting using Spatio-temporal Asymmetric Models Ahmed Aziz Ezzat, Texas A&M.University, aa.ezzat@tamu.edu Despite the growing recognition for non-separable spatio-temporal models, a significant reliance on separable, fully symmetric models is still the norm in today’s renewable industry. We explore the reasons behind the disregard of these models in wind farm analytics and devise a special pair of spatio-temporal lens that unearths the fine-scale spatio-temporal variations and interactions, at which the merits of the highly-statistically capable models are perceived. Given this lens, we investigate the potential benefits, in terms of inference and forecasting, achieved through extending non separable asymmetric spatio-temporal models for wind energy applications.

370D Spatio-Temporal Conservation Models for Controlling Biological Invasions Sponsored: Energy, Natural Res & the Environment Environment & Sustainability Sponsored Session Chair: Esra Buyuktahtakin, New Jersey Institute of Technology, Newark, NJ, United States, esratoy@njit.edu 1 - Valuing Monitoring Networks for New Agricultural Pathogens Tiesta Thakur, University of Minnesota, St. Paul, MN, 55108, United States, thaku035@umn.edu, Terry Hurley, Frances Homans, Robert Haight US soybean production, a $30 billion industry, is threatened by soybean rust, a new agricultural pathogen that entered the U.S. in 2004. To monitor rust occurrence, a public-private partnership funds a sentinel plot network, which reached a peak of 900 plots in 2008. Soybean producers use sentinel plot information to assess annual risk of infection and make fungicide application decisions. To value the network, we first develop a dynamic, farm-level decision model and estimate producers’ willingness to pay for sentinel plot information. Then, we use an assignment model to locate sentinel plots in the eastern U.S. to maximize net benefits. Results are compared with current plot locations. 2 - Optimal Survey Strategies in Early Detection Programs of Bio-invasions: Detect or Delimit? Denys Yemshanov, Natural Resources Canada, 1219 Queen Street East, Sault Ste. Marie, ON, p6a2e5, Canada, Denys.Yemshanov@canada.ca, Robert G.Haight, Frank H.Koch, Robert Venette, Tom Swystun, Ronald Fournier, Yongguang Chen, Mireille Marcotte, Jean J.Turgeon We present a scenario-based model that incorporates uncertainty about the arrival of an invasive species and damage from an outbreak and optimizes the deployment of surveys for early detection of the outbreak. The model minimizes time to first detection and overall damage from an outbreak in a landscape and accounts for decision-making aspirations towards worst-case outcomes. We apply the model to allocate early detection surveys of Asian longhorned beetle in the Greater Toronto Area, Canada. Our approach is focused to support early detection programs for novel invasive pests and pathogens 3 - A Multi-stage Stochastic Optimization Approach to Cost-effective Surveillance and Control of Biological Invasions Eyyub Kibis, The College of Saint Rose, Albany, NY, United States, eyyubyunus@gmail.com, Esra Buyuktahtakin, Robert G. Haight, Kathleen S. Knight, Charlie Flower In this study, we develop a new multistage stochastic programming model, which considers all possible scenarios for surveillance, treatment, and removal decisions over a planning horizon to control the emerald ash borer invasion in North America. Our objective is to maximize the net benefits of the ash trees on a given landscape by applying surveillance to the ash population, followed by treatment or removal of trees based on their infestation level. Due to the model’s complexity, we develop new cutting planes to facilitate the solution process. Results provide insights into surveillance and control policies, and provide an optimal strategy to minimize EAB infestation with scarce resources.

TB64

370E Retail Analytics Sponsored: Data Mining Sponsored Session

Chair: Matthew A. Lanham, Purdue University, 403 W. State St., KRAN 466, West Lafayette, IN, 47907, United States, lanhamm@purdue.edu 1 - Assortment Optimization with Big Data: Weather, Social, Halo and Cannibalization Victor Pereira, IBM, Columbus, OH, United States, vpereira@vt.edu This talk showcases a complex and time-constrained data science project to optimize deodorant assortments at dozens of demographics-grounded store clusters. The project was sponsored by a CPG manufacturer on behalf of two major retailers and leveraged the effect of hyperlocal weather, changes in consumer sentiment towards product attributes, cluster-level halo and cannibalization, and minimum inventory for high-velocity SKUs. Over a billion rows of data occupying nearly 5TB were analyzed during this 8-week proof of concept. Model outputs matched the opinions of CPG experts at the product attribute level.

326