Informs Annual Meeting 2017

WA80

INFORMS Houston – 2017

WA80

1 - On the Analysis of Accelerated Variable Sample-size Stochastic Approximation (VSSA) Uday Shanbhag, Pennsylvania State University, University Park, 310 Leonhard Bldg, University Park, PA, 16802, United States, udaybag@psu.edu Traditional stochastic approximation (SA) schemes employ a single gradient or a fixed batch of noisy gradients in computing a new iterate. To mitigate computational complexity, we consider increasing sample size when the total simulation budget is finite. We present amongst the first stochastic generalizations of Nesterov’s accelerated gradient scheme where we first show that when the objective is strongly convex, the expected sub-optimality error diminishes at a geometric rate while upon termination. Also, for convex case we can recover optimal convergence rate as well. 2 - A Unifying Selftuned Stepsize Rule for Stochastic Mirror Descent Methods Motivated by big data applications, we consider stochastic mirror descent methods for solving stochastic convex optimization problems. The performance of this class of methods is very sensitive to the choice of the stepsize sequence. We present a unifying self-tuned stepsize for smooth and nonsmooth problems such that: (i) it is characterized in terms of problem parameters and algorithm’s settings; and (ii) under this update rule, a suitably defined error metric is minimized. We present the performance of this update rule for the soft margin linear SVM problem over different large datasets. 3 - On Variable Sample-size Stochastic Mirror-descent and Fista-like Schemes for Nonsmooth Stochastic Optimization Afrooz Jalilzadeh, 424 Waupelani Drive, Apt Q34, State College, PA, 16801, United States, Azj5286@psu.edu, Uday Shanbhag Traditional stochastic approximation (SA) schemes with convex constraints employ a single gradient and a projection on a convex set in computing an update. To reduce computational complexity, we consider two schemes; nonsmooth Variable Sample-Size Stochastic Mirror-Descent (VS-SSMD) and Variable sample-size FISTA (VS-FISTA), in which N_k samples are utilized at step k and the total simulation budget is M. Under a strong convexity assumption, we prove linear convergence of iterates. Additionally, we show that the proposed VS- FISTA scheme for merely convex case displays the deterministic O(1/k^2) convergence rate. 4 - Sampling-based Schemes for the Solution for Two-stage Risk-limiting Economic Dispatch Problems Wendian Wan, Pennsylvania State University, University Park, PA, United States, wzw121@psu.edu Risk Limiting Dispatch (RLD) is a new operating paradigm that uses real-time information about supply and demand, integrating complex stochastic nature of power system. Tradeoff between computational complexity and accuracy need to be considered by system operator. In this talk, two-stage risk-limiting economic dispatch problem is formulated and a sampling-based schemes has been utilized in order to meet the scalability requirement of power system. Nahidsadat Majlesinasab, Oklahoma state university, nahid.majlesinasab@okstate.edu, Farzad Yousefian, Arash Pourhabib 382B Risk-Averse Stochastic Optimization Sponsored: Optimization, Optimization Under Uncertainty Sponsored Session Chair: Alexander Vinel, Auburn University, Auburn University, Auburn, AL, 36832, United States, alexander.vinel@auburn.edu 1 - Risk-averse Stochastic Optimization Model for Hazardous Material Transportation using Risk Parity and Modern Risk Measures Nasrin Mohabbati Kalejahi, PhD Candidate, Auburn University, Auburn, AL, United States, nasrin@auburn.edu, Alexander Vinel Hazardous material (hazmat) transportation problem aims to ship hazmats from an origin to one or several destinations through a road network. Using a single optimal path repeatedly over time increases the risk of accidents for the population in surrounding area. One of the biggest challenges in hazmat transportation is to reach risk equity over the network. In this research, risk parity concept is combined with modern risk measures and a risk-reward- diversification framework is developed. The objective of the optimization model is to find multiple routes and fairly distribute the exposure to risk in the transportation network to guarantee the equal risk contribution of each route to the total risk. WA82

381C Joint session PSOR/ENRE: Models for Energy Policy 1 Sponsored: Energy, Natural Res & the Environment, Energy Sponsored Session Chair: Erin Baker, Univ of Massachusetts-Amherst, University of Massachusetts-Amherst, Amherst, MA, 01003, United States, edbaker@ecs.umass.edu Co-Chair: Zana Cranmer, Bentley University, Waltham, MA, United States, zanacranmer@gmail.com 1 - Potential for Widespread Electrification of Vehicle Travel Jessika Trancik, Massachusetts Institute of Technology, 77 Massachusetts Avenue, E40-241, Cambridge, MA, 02139, United States, trancik@mit.edu Electric vehicles can contribute to climate change mitigation but only if vehicle range matches travelers’ needs. Evaluating electric vehicle range against a population’s needs is challenging because detailed driving behavior must be accounted for. Here we develop a model to combine information from expansive travel surveys with high-resolution GPS data to estimate the energy requirements of personal vehicle trips across the U.S. We find that the energy requirements of most driving days could be met by today’s affordable electric vehicles. For the highest-energy days, other vehicle technologies are likely needed, and predictive models combined with car sharing could play an important role. 2 - Identifying Portfolios of Energy Technologies that are Robust to Multiple Models Franklyn Kanyako, Research Assistant, University of Massachusetts, Amherst, Amherst, MA, 01003, United States, fkanyako@umass.edu, Erin Baker This project investigates the problem of allocating R&D funds across low-carbon energy technologies in the context of climate change. Multiple expert elicitation on the future cost of low energy technologies are used as input into three Integrated Assessment Models(IAMs), GCAM, WITCH, and MESSAGE. IAMs are designed with a different representation of processes and assumptions about the economy, energy, and policy, which create an uncertainty known as model uncertainty. Outputs from these models are used to conduct robust portfolio decision analysis to identify the non-dominated portfolio of alternatives that are robust to all models. 3 - Improving Land Use Efficiency of Utility-scale Solar Photovoltaics Jeremiah Johnson, University of Michigan, 440 Church Street, Dana Building, Ann Arbor, MI, 48109, United States, jxjohns@umich.edu, Noah Mitchell-Ward This study explores the ability and limitations to increase the land use efficiency of solar photovoltaic generation through decisions available to solar project developers. To achieve this, we developed a physics-based model to determine solar angles and shading impacts as a function of time, location, collector orientation, and collector tilt. Using Typical Meteorological Year data, we determine the maximum solar generation per unit of land area for various locations and module efficiencies. These findings show the potential reductions in land use for utility-scale solar, as well as the tradeoff between generation per module and generation per land area, as driven by shading losses. 4 - Valuing Offshore Wind Energy Zana Cranmer, Bentley University, Waltham, MA, United States, zanacranmer@gmail.com, Erin Baker Much attention has been paid to the environmental costs of offshore wind energy in terms of the impacts on ecosystems and species such as birds, bats, and marine wildlife. For comparison, we need to understand the value of permitting offshore wind farms in terms of mitigating global climate change. We develop a model for estimating the value of permitting offshore wind projects globally over the rest of the century and estimate that the value ranges between $25 billion to $48.8 trillion dollars in a business as usual case. The corresponding value of technological change ranges from about $474 billion to $45.2 trillion dollars in this case. 382A Enhancements of Stochastic Approximation Sponsored: Optimization, Optimization Under Uncertainty Sponsored Session Chair: Uday Shanbhag, Pennsylvania State University, University Park, 310 Leonhard Bldg, University Park, PA, 16802, United States, udaybag@psu.edu WA81

470