2015 Informs Annual Meeting

MA15

INFORMS Philadelphia – 2015

2 - Sparse but Efficient Operation: A Conic Programming Approach Gao Yini, National University of Singapore, 1 Business Link, Singapore, Singapore, yini.gao@u.nus.edu, Chung Piaw Teo, Zhenzhen Yan Standardization and flexibility are two competing paradigms in designing efficient operations. We ask whether there is a sparse but flexible operation mode to reap the benefits of both. Using copositive conic programming, we develop a new mechanism which gives sparse but efficient network structures. It recovers k chain in the context of process flexibility. We further apply it to Singapore Changi Airport “roving team” deployment problem and obtain a sparse yet efficient deployment network. 3 - SDP Reformulation of CP Programs: Best-worst Choice and Range Estimation Applications Karthik Natarajan, Singapore University of Technology and Design, Singapore, 487372, Singapore, karthik_natarajan@sutd.edu.sg, Chung Piaw Teo We show that the worst case moment bound on the expected optimal value of a mixed integer linear program with a random objective c is obtained from a SDP reformulation of a completely positive program. We illustrate the usefulness of the distributionally robust bounds in estimating the expected range of random variables with two applications arising in random walks and best-worst choice models. 4 - Robust Inventory Models with Demand Partitioning Information Joline Uichanco, Asst. Professor, University of Michigan, Ross School of Business, 701 Tappan Ave, Ann Arbor, MI, 48109, United States of America, jolineu@umich.edu, Karthik Natarajan, Melvyn Sim We present the distributionally robust newsvendor with demand asymmetry information through partition statistics. We derive a closed-form for the robust order quantity under the special case of semivariance, implying a simple rule-of- thumb for setting order quantities under limited information. The distribution can be calibrated from primitive demand data. We demonstrate the performance of the method in computational experiments on data from an automotive spare parts company. MA14 14-Franklin 4, Marriott Stochastic Optimization Applications to Renewable Energy Integration Sponsor: Optimization/Optimization Under Uncertainty Sponsored Session Chair: Lindsay Anderson, Assistant Professor, Cornell University, 316 Riley Robb Hall, Ithaca, NY, 14853, United States of America, landerson@cornell.edu 1 - Multi-Objective Optimal Sensor Deployment under Uncertainty for Advanced Power Systems Urmila Diwekar, President, Vishwamitra Research Institute, 2714 Crystal Way, Crystal Lake, IL, 60012, United States of America, urmila@vri-custom.org, Pallabi Sen, Kinnar Sen Advanced power plants using an integrated gasification combined cycle (IGCC) offer a competitive and economical means to produce electricity with reduced emission levels. An efficient, safe, and reliable operation of an IGCC plant requires effective strategies for monitoring and control. The results of this multi- objective framework for optimizing observability, efficiency, and cost for an IGCC system are presented in this work. 2 - Optimal Microgrid Design under Load and Photovoltaic Power Uncertainty Alex Zolan, University of Texas at Austin, 204 E. Dean Keeton Street, Stop C2200, Austin, TX, 78712, United States of America, alex.zolan@utexas.edu, Alexandra Newman, David Morton We present a model for establishing the design and energy dispatch for a microgrid that minimizes cost and fuel requirements given the set of technologies (diesel generators, solar arrays and batteries), photovoltaic power (PV) availability on location, and probability model that governs the load and PV availability of a forward operating base. We introduce a policy-based restriction of the problem that allows for the solution of a multiple scenario problem while preserving solution quality. 3 - Tracking a Stochastic Generate-pump Schedule for a Pumped-storage Hydroelectric Unit Bismark Singh, The University of Texas at Austin, 204 E. Dean Keeton Street, Stop C2200 ET, Austin, TX, 78705, United States of America, bismark.singh@utexas.edu, Surya Santoso Using a stochastic dynamic program, we first optimize the generate-pump schedule for a pumped-storage hydroelectric unit to maximize profit. Since energy prices are stochastic, we find an adaptive policy for the schedule. And,

since we must submit bids to an ISO, we seek a bidding strategy that will allow us to track the desired generate-pump schedule. Thus, we solve a model that yields an optimal block-bidding policy in the sense of tracking the desired stochastic generate-pump policy. 4 - A Stochastic Model to Determine Probabilistic Reserves Requirements for Unit Commitment Problems Gabriela Martínez, Cornell University, Ithaca, NY, United States of America, gabriela.martinez@cornell.edu, Lindsay Anderson In this work, we propose a stochastic unit commitment model to decide appropriate spinning and non-spinning reserve requirements for a power system with high penetration of renewable energy. The day-ahead scheduling of the systems is formulated as a chance-constrained model in which the network power balance of the systems is ensured with a high-probability level and the system reserves are allocated in a risk-averse fashion by selection of quantiles of the uncertain generation. MA15 15-Franklin 5, Marriott Radiation Therapy Optimization Sponsor: Optimization in Healthcare Sponsored Session Chair: Arka Roy, Bowling Green State University, 440 W. Barry Ave., Chicago, IL, United States of America, arkaroy1@gmail.com 1 - A Robust Optimization Method for Homogeneous Magnet Design in MR-guided Radiation Therapy Iman Dayarian, University of Toronto, 5 King’s College Road, Toronto, ON, M4Y 2P9, Canada, iman@mie.utoronto.ca, Timothy Chan, Teodor Stanescu Magnetic resonance imaging uses a magnetic field generated by a configuration of coils to image patients. An optimization-generated coil configuration can be sensitive to small perturbations that affect the homogeneity of the magnetic field. This sensitivity is especially important when the coils are mounted on a treatment device that rotates during treatment, which is the case in MR-guided radiation therapy (MRgRT). This talk presents a robust optimization approach to magnet design for MRgRT. 2 - Incorporating Lung Ventilation Function in Intensity-modulated Radiation Treatment Planning 4DCT-derived ventilation images were used for pencil-beam intensity modulation to achieve functional sparing of lung on a voxel-by-voxel basis. This functional approach was compared to the conventional anatomical planning on 10 patients retrospectively. Significant reductions (p-values < 0.001) of V20 (lung volume receiving >=20Gy) (11%), functional V20 (18%), mean lung dose (MLD) (7%) and functional MLD (11%) were achieved without significantly increasing doses to the other organs-at-risk. 3 - Optimizing Global Liver Function in Stereotactic Body Radiotherapy Treatment Planning Fujun Lan, Postdoctoral Fellow, University of Maryland, Baltimore, 22 S. Greene St., GGJ02, Baltimore, MD, 21201, United States of America, flan@email.arizona.edu, Warren D’Souza, Hao Zhang Victor Wu, PhD Student, University of Michigan, 1205 Beal Avenue, Ann Arbor, MI, 48109, United States of America, vwwu@umich.edu, H. Edwin Romeijn, Marina Epelman, Martha Matuszak, Yue Cao, Mary Feng, Hesheng Wang, Randall Ten Haken We propose a radiotherapy treatment planning optimization model for liver cancer cases. In this work, we plan treatment using voxel-based liver dose- response model: post-treatment liver function depends on its pre-treatment function and the dose received. We maximize predicted post-treatment global liver function. We approximately solve the resulting non-linear non-convex problem with a customized mixed-integer linear programming-based algorithm. 2D synthetic and 3D clinical cases were studied. 4 - Robust Adaptive Optimization in Radiation Therapy Arka Roy, Bowling Green State University, 440 W. Barry Ave., Chicago, IL, United States of America, arkaroy1@gmail.com, Omid Nohadani Radiotherapy treatments degrade over time in the presence of uncertainties. Robust models leap beyond such limitations. However, traditional robust models solve for the worst-case realization of the uncertainty prior to the start of the treatment, which may be too conservative at later fractions. We propose a robust two-stage approach that adapts to the first-stage decisions during treatment. The results are demonstrated through a clinical prostate case.

149