Informs Annual Meeting 2017

MC82

INFORMS Houston – 2017

4 - Long-term Coordination of Transmission and Storage to Integrate Wind Power Antonio J. Conejo, The Ohio State University, Department of Integrated Systems Engineering, 210 Baker Systems Building, Columbus, OH, 43210, United States, conejonavarro.1@osu.edu, Ning Zhang, Chongqing Kang, Yaohua Cheng A power system with a high wind power integration requires extra transmission capacity to accommodate the intermittency inherent to wind power production. Storage can smooth out this intermittency and reduce transmission requirements. This presentation describes a stochastic optimization model to coordinate the long-term planning of both transmission and storage facilities to efficiently integrate wind power. Both long-term and short-term uncertainties are considered in this model. Long-term uncertainty is described via scenarios, while short-term uncertainty is described via operating conditions. 381C Optimizing Distributed Energy Generation Sponsored: Energy, Natural Res & the Environment, Energy Sponsored Session Chair: Alexandra M Newman, Colorado School of Mines, Golden, CO, 80401, United States, anewman@mines.edu 1 - Parameterization for Optimized Dispatch of Concentrating Solar Power Production Will Hamilton, Colorado School of Mines, Golden, CO, United States, whamilton@mymail.mines.edu Concentrated solar power technologies can store solar energy, decoupling electricity generation from time periods with solar resource. The decision to produce electricity or to store thermal energy for later use is complex and involves multiple inputs. We have developed a profit-maximizing mixed-integer linear program that determines start-up and shutdown times for the power cycle and solar receiver. Here, we present the results of sensitivity studies on these key plant performance parameters, which provide insight into the importance of the accuracy of each parameter to ensure the viability of operational decisions using optimal thermal energy storage dispatch. 2 - Power Curve Prediction through a Nonparametric Additive Multivariate Kernel Approach David Perez, Texas A&M.University, College Station, TX, United States, dmp4@tamu.edu This presentation intends to demonstrate how to estimate a wind turbine’s power curve based on environmental conditions through an additive multivariate kernel (AMK) approach. The power curve represents the sigmoidal relationship between power output of a wind turbine and velocity. However, other environmental variables like wind direction and air density also impact power production. The AMK method uses these other variables to formulate an estimator of power production, taking advantage of the multiplicative, nonlinear relationship between the response and multiple explanatory variables. Finally, the AMK method is compared to other popular approaches for prediction accuracy. 3 - A Polyhedral Study of Min-up/-down and Ramping Polytope Kai Pan, University of Florida, 411 Weil Hall, Gainesville, FL, 32611, United States, kpan@ufl.edu, Yongpei Guan We present a comprehensive polyhedral study on the integrated minimum-up/- down time and ramping polytope, which has broad applications in power generation scheduling (on both transmission/distribution levels), manufacturing scheduling, chemical engineering, etc. Both theoretical and numerical results are provided to demonstrate the strength and applications of the proposed cutting planes. 382A Data Driven Optimization Sponsored: Optimization, Optimization Under Uncertainty Sponsored Session Chair: Andrew Lim, National University of Singapore, National University of Singapore, Singapore, 119245, Singapore, andrewlim@nus.edu.sg 1 - Demand Forecasting for New Products in a Cosmetic Company Shanshan Huang, shanshanhuang@u.nus.edu We build a Bayesian structure model to study demand estimation for new products. Based on historical product covariates and sales, we are able to learn the sales pattern and forecast the future demand. MC80 MC81

2 - Smart “Predict, then Optimize” Adam Elmachtoub, Columbia University, 500 W. 120th St., New York, NY, 10027, United States, adam@ieor.columbia.edu, Paul Grigas We consider a class of optimization problems where the objective function is not explicitly provided, but contextual information can be used to predict the objective based on historical data. A traditional approach would be to simply predict the objective based on minimizing prediction error, and then solve the corresponding optimization problem. Instead, we propose a prediction framework that leverages the structure of the optimization problem that will be solved given the prediction. We provide theoretical, algorithmic, and computational results to show the validity and practicality of our framework. 3 - Failing to Learn via Exploration Boosts Michael Jong Kim, University of British Columbia, Vancouver, BC, Canada, mike.kim@sauder.ubc.ca, Ya-Tang Chuang We investigate a class of sequential Bayesian maintenance optimization problems where parameters of the lifetime distribution are learned using right-censored failure data. We show that the optimal preventive maintenance (PM) time can be expressed as the sum of a myopic-optimal PM time plus an ``exploration boost” that is proportional to the posterior variance. We conduct asymptotic structural and performance analysis of the optimal PM policy. 4 - A DC Optimization Approach to Sparse Spline Regression Jun-ya Gotoh, Professor, Chuo University, 1-13-27 Kasuga, Bunkyo-ku, Tokyo, 112-8551, Japan, jgoto@indsys.chuo-u.ac.jp, Yuichi Misawa In this paper we consider the estimation of a spline regression with a constraint on the number of splines to be employed. Reformulating the cardinality constraint into a DC (difference of two convex functions)-constraint, we further reformulate the problem to an unconstrained one, where the DC constraint is moved to the objective after multiplying a positive constant. Then a proximal DC algorithm (Gotoh, Takeda, Tono, 2015) is applied, so that a stationary point can be efficiently attained. Numerical results will be presented in comparison with an existing l1-regularized formulation. 382B Optimization under Multistage Uncertainty Sponsored: Optimization, Optimization Under Uncertainty Sponsored Session Chair: Kartikey Sharma, Northwestern University, Evanston, IL, 60208, United States, kartikeysharma2014@u.northwestern.edu 1 - Optimization under Connected Uncertainty Distributionally robust and standard robust optimization methods provide a tractable way to address uncertainties. In many applications, past observations influence future uncertainties. In this presentation, we leverage this dependence via connected uncertaintysets, where the set parameters at each period depend on previous realizations. To find optimal here- and-now solutions, we reformulate distributionally robust and standard robust constraints for connected uncertainty sets. We illustrate the advantages of this framework with an application in portfolio optimization. 2 - Multistage Robust Optimization with Dynamic Uncertainty Sets for Power System Operations Alvaro Lorca, Pontificia Universidad Catolica de Chile, Santiago, Chile, alvarolorca@uc.cl, Andy Sun Motivated by the deep integration of volatile renewable energy sources such as wind and solar power, we will present robust optimization models for real-time and day-ahead power system operations, considering uncertainty in the power output of there sources. An essential component of the developments presented is the proposal of dynamic uncertainty sets to capture temporal and spatial correlations of wind and solar power. We will also discuss the solution methods designed for these challenging problems and the insight obtained from computational experiments. MC82 Kartikey Sharma, Northwestern University, Room C223. 2145 Sheridan Road, Evanston, IL, 60208, United States, kartikeysharma2014@u.northwestern.edu, Omid Nohadani

233