Informs Annual Meeting 2017

MD79

INFORMS Houston – 2017

4 - On the Identification of Optimal Partition for Second Order Cone Optimization Tamás Terlaky, Lehigh University, 200 W. Packer Ave, Bethlehem, PA, 18015, United States, terlaky@lehigh.edu, Ali Mohammad-Nezhad The concept of optimal partition was originally introduced for LO and LCP and subsequently extended to SDO. Terlaky and Wang specialized the concept for SOCO. In this paper, we present a fast iterative procedure, which uses the optimal partition information to generate a pair of maximally complementary optimal solutions. We are interested in the local convergence of the iterative procedure near the optimal set. In this regard, we show that, under the primal and dual nondegeneracy conditions, the iterative procedure is quadratically convergent to a maximally complementary optimal solution when starting from a strictly interior pair of solutions close to the optimal set. 381A Electricity Markets Sponsored: Energy, Natural Res & the Environment Electricity Sponsored Session Chair: Todd Levin, Argonne National Laboratory, 9700 S. Cass Ave, Bldg 202, Lemont, IL, 60439, United States, tlevin@anl.gov 1 - A Multistage Stochastic Dynamic Programme for Major Consumers’ Demand Response in Electricity Markets Mahbubeh Habibian, University of Auckland, 439-325, 70 Symonds, Auckland, 1010, New Zealand, mhab735@aucklanduni.ac.nz, Golbon Zakeri, Anthony Downward We optimize the bidding policy for a major consumer of energy in an energy and reserve co-optimized electricity market. In our model, the major consumer has dispatchable demand, and it is able to offer interruptible load reserve. The large consumer is a strategic player, maximizing its profit, while meeting its required level of consumption over a long-time horizon. By reformulating the equilibrium constraints, we present a multi-stage stochastic dynamic programming. In order to solve our mixed integer program in an adequate time-frame we use heuristics while utilizing the intuition of stochastic dual dynamic programming. 2 - Demand Response to Electricity Prices in Flexible Manufacturing Kazem Abbaszadeh, The University of Auckland, Room F, level 1,11 Mount St, Auckland, 1010, New Zealand, kabb955@aucklanduni.ac.nz, Golbon Zakeri, Geoffrey Pritchard Large manufacturers are often also large users of electricity. This includes Aluminium and Steel manufacturers as well as paper and dairy processing plants in New Zealand. We consider a major consumer of electricity whose decision to contract for production could depend on electricity prices. We describe our stochastic dynamic programming approach for a steel mill in New Zealand with some flexibility in their operations, and utilize this flexibility to provide an optimal production plan. As a part of this model, we need to develop electricity price models over the short and medium time horizon. We will describe these models as well. 3 - Key Drivers of Prices for Ancillary Services Todd Levin, Energy Systems Engineer, Argonne National Laboratory, 9700 S. Cass Ave, Bldg 362, Lemont, IL, 60439, United States, tlevin@anl.gov We apply AURORAxmp, a commercial power systems model, to forecast how prices for ancillary services (AS) in PJM are affected under various future scenarios. We find that doubling the natural gas price increases the average frequency regulation price by 41% and decreases the average price for synchronized and primary reserves by 75% and 23% respectively. Increasing wind generation capacity by a factor of 15 decreases average AS prices by 8%, 25% and 90% for regulation, synchronized and primary reserves. Several additional sensitivities including increased AS requirements, unit retirements and natural gas supply constraints are also considered and revenue sufficiency implications are discussed. 380B CS and OR Interface Contributed Session Chair: Hesam Shams, University of Tennessee, Knoxville, TN, United States, hesam@utk.edu 1 - Bias in Balance Optimization Subset Selection Hee Youn Kwon, PhD Student, University of Illinois at Urbana- Champaign, 104 S.Mathews Avenue, Transportation Building, Urbana, IL, 61801, United States, hkwon14@illinois.edu, Jason Sauppe, Sheldon H. Jacobson MD78 MD78B

When estimating a treatment effect from observational data, researchers encounter bias regardless of estimation methods. By focusing on a particular method of estimation called Balance Optimization Subset Selection (BOSS), we investigate all the possible cases that may lead to bias in the context of BOSS and try to mitigate the bias. While doing so, we define a balance hierarchy and a correct imbalance measure which corresponds to a functional form of the responses. 2 - Service Placement Optimization in Mobile Edge Computing Hossein Badri, Graduate Teaching Assistant, Wayne State University, 4185 4th Street, Manufacturing Building, Detroit, MI, 48202, United States, fq2529@wayne.edu, Kai Yang Service placement is a very important and at the same time a super complex problem in Mobile Edge Computing (MEC). Dynamisms of users’ location and uncertainties in their demands are the main source of complexity in this problem. In this research, we develop a stochastic programming model for the service problem in MEC and conduct an extensive computational analysis to evaluate the performance of the proposed approach. 3 - Two-stage Stochastic Mathematical Model for Inventory Management of Red Blood Cells Accounting for Age-differentiated Demand Bayan Hamdan, Masdar Institute, Abu Dhabi, United Arab Emirates, bhamdan@masdar.ac.ae, Ali Diabat Blood Supply Chain inventory management has recently become one of the most important fields in operations research due to the perishability of blood. This paper forms a mathematical model that determines an optimal inventory ordering policy for red blood cells (RBCs). The model accounts for the perishability of the red blood cells and the age-differentiation and stochasticity of the demand. This paper proposes an (R,S) inventory policy where R is the review frequency and S is the target level, to minimize system outdate, shortage, substitution and holding costs. 4 - Surrogate Modeling Recommendations Reza Alizadeh, Graduate Research Assistant, University of Oklahoma, Norman, OK, 73019, United States, reza.alizadeh-2@ou.edu, Janet K. Allen, Farrokh Mistree Models describing engineering systems are often simplified, thus creating uncertainty. Surrogate models with their reduced computational costs allow us to manage uncertainty by substantially reducing computational expense. Much has been done to compare different methods; however, there is a gap in providing guidance for the choice of appropriate surrogate modeling methods. We ask these questions: 1) What are the critical characteristics of a model to be approximated? 2) What are desirable characteristics of the DOE, surrogate model and model fitting methods to achieve useful surrogate models? Answering these questions leads to a toolbox to choose appropriate surrogate modeling methods. 5 - Enhancing Optimization Algorithm via Machine Learning Hesam Shams, University of Tennessee, 851 Neyland Drive, 525 John D. Tickle Engineering Building, Knoxville, TN, 37996, United States, hesam@utk.edu, Oleg Shylo We explore an application of classification techniques to binary optimization. Given a set of solutions and the corresponding objective values, the goal is to classify each solution component either as zero or one. We implement statistical models based on random forests, convolutional neural networks, and logistic regression and compare their performance on a set of scheduling problems. A learning module based on an efficient logistic regression model is embedded into the tabu search method. The comparison with standard tabu algorithm implementations reveals significant gains in efficiency. 381B Computational Challenges in RTO Markets Sponsored: Energy, Natural Res & the Environment Electricity Sponsored Session Chair: Yonghong Chen, Midwest ISO, Carmel, IN, 46032, United States, ychen@misoenergy.org 1 - Computational Challenges in RTO Markets Yonghong Chen, Midwest ISO, 720 City Center Drive, Carmel, IN, 46032, United States, ychen@misoenergy.org Optimization techniques and algorithms are at the core of electricity market operations with the goal of minimizing system operation cost while ensuring system security. The industry evolution has posed unprecedented challenges to the existing optimization and computation techniques. This panel will discuss the latest advances in formulating optimization problems and developing new algorithm and solution techniques. It includes the development of tighter Mixed Integer Programming (MIP) models and exploration of novel algorithms in solving MIP problems on traditional hardware as well as with distributed and high performance computers. MD79

265