Informs Annual Meeting 2017

TC72

INFORMS Houston – 2017

TC72

6 - Distribution Network Reconfiguration Considering Decentralized Autonomous Electric Vehicles Zhi Zhou, Argonne National Laboratory, 9700 South Cass Avenue, Bldg 221, Argonne, IL, 60439, United States, zzhou@anl.gov, Zhaomiao Guo Autonomous electric vehicles (AEVs) provide unique opportunities to cope with the uncertainties in electricity distribution network (DN), but the effects are limited by the inherent radial topology. We investigate the potential benefits (e.g. reducing line losses, battery degradation, etc.) of dynamic DN reconfiguration (DDNR) based on AEVs spatial-temporal availability. A mixed integer programming model is proposed to optimally coordinate the charging/discharging of AEVs with day-ahead DDNR plan, while satisfying AEVs’ daily travel demand in a two-settlement setting. Numerical findings will be presented through the IEEE 33-node test feeder and Sioux Falls transportation network. 372B Meta-Algorithmics Invited: Operations Research and the Future of Computing Invited Session Chair: Holger Hoos, University of British Columbia, Vancouver, NB, N6T I24, Canada, hoos@cs.ubc.ca 1 - Algorithm Selector and Prescheduler in the ICON Challenge François Gonard, INRIA, Paris, France, francois.gonard@inria.fr Algorithm portfolios are known to offer robust performances, efficiently overcoming the weakness of every single algorithm on some particular problem instances. In order to get the best out of an algorithm portfolio, one can either use an algorithm selection (AS) approach, or define a scheduler, sequentially launching a few algorithms on a limited computational budget each. The ASAP system presented here relies on the joint optimization of a pre-scheduler and a per instance AS, selecting an algorithm well-suited to the problem instance at hand. ASAP has been thoroughly evaluated against the state-of-the-art during the ICON challenge for algorithm selection, receiving an honourable mention. 2 - A Machine Learning Approach to Re-designing Branch-and-Bound Search Bistra Dilkina, Georgia Institute of Technology, 1304 Klaus Bldg, 266 Ferst Dr, Atlanta, GA, 30332, United States, bdilkina@gmail.com, Elias Khalil Branch-and-Bound approaches have been improved over many years by a series of ideas from operations research and computing. However, many of the components of this complex complete search strategy used in state of the art solvers are highly engineered and to a large extent static. We investigate the use of machine learning ideas to re-vamp diverse aspects of the branch-and-bound search in order to discover new strategies and heuristics in a data-driven and adaptive way that can be useful both broadly as well as fine-tuned to domain- specific families of instances. 3 - Generation Techniques for Linear and Mixed Integer Programming Instances with Controllable Properties Simon Bowly, Monash University, Clayton, 3800, Australia, simon.bowly@monash.edu, Kate Smith-Miles, Davaatseren Baatar, Hans Mittelmann The results of empirical analysis of optimisation algorithm performance are highly dependent on the quality of test instances available. Real world data sets may lack diversity in feature and performance data spaces, while random generation can produce instances which are not applicable to the problem being studied. We propose a new framework which shifts test instance generation to an encoded space. Through its application to the generation of linear and mixed integer programming test cases we demonstrate the potential to improve the quality of metadata used in algorithm analysis. 4 - How Simple is Too Simple: Evaluating Increasingly Sophisticated Techniques for Algorithm Portfolios Chris Cameron, University of British Columbia, Vancouver, BC, Canada, chris.cameron.queens@gmail.com, Holger Hoos, Frank Hutter, Kevin Leyton-Brown State-of-the-art methods for building algorithm portfolios have achieved notable success on computationally-hard problems such as SAT. While these methods are often quite complex, we are not aware of any previous, systematic study of the extent to which this complexity is warranted. We created a wide design space of algorithm portfolio techniques and evaluated the resulting designs on a comprehensive set of benchmarks (ASlib 2.0), which includes industrial problems in SAT and CSP. In particular, we investigated three design axes: (1) model sophistication (from model-free to model-based cost-sensitive classification), (2) feature sophistication TC73

372A Optimization, Nonlinear Programming Contributed Session Chair: Zhi Zhou, Argonne National Laboratory, Argonne, IL, United States, zzhou@anl.gov 1 - Rate of Convergence of the Bundle Method and Selective Linearization Algorithm for Large Scale Convex Optimization Yu Du, Assistant Professor, University of Colorado Denver, 1475 Lawrence St, Denver, CO, 80202, United States, duyu197@gmail.com, Andrzej Ruszczynski The number of iterations needed by the bundle method for nonsmooth optimization to achieve a specified solution accuracy can be bounded by the product of the inverse of the accuracy and its logarithm, if the function is strongly convex. The result is true for the versions of the method with multiple cuts and with cut aggregation. We also illustrate the related efficient selective linearization algorithm on the large scale statistical learning problems. 2 - Maximizing Resilience of Transmission Power Grids through a Bi-level Restoration Model Saeedeh Abbasi, University of Houston, 2821 Greenridge Dr, Apt 100, Houston, TX, 77057, United States, sabbasi5@uh.edu, Masoud Barati, Gino J.Lim Power grid’s outdoor facilities are vulnerable to extreme events, and their failure could trigger the failure of the whole network. The imposed complete blackout lasts for a long period while terribly costs. Therefore, it is so crucial to make the grid resilient, and an appropriate fast restoration is demanded to mitigate these impacts. In this study, the power grid resiliency is quantified and maximized in a bi-level optimization framework and solved through an iterative optimization algorithm (IOA) and a mathematical program with equilibrium constraint (MPEC). The model’s performance is analyzed with 6- and 118-bus IEEE test systems. 3 - Survivable All-optical Networks: Extensions to the Routing and Wavelength Assignment Problem Mehdi Hemmati, Sharif University of Technology, Tehran, Iran, Islamic Republic of, soheilmn@gmail.com, Cole Smith All-Optical Networks (AONs) form the foundation of the Internet backbone. They transmit the data over light rays, which require dedicated wavelengths. Hence, the routing problems in AONs are coupled with wavelength assignment tasks, which result in the Routing and Wavelength Assignment (RWA) problem. Due to the massive amount of data transmission over AONs, their vulnerability to disruptions are of high importance. We discuss variations of the RWA problem to study the resilience of AONs in the presence of intentional or unintentional disruptions. We propose multi-level integer programs and investigate the solution techniques by studying the solution space of the related optimization problems. 4 - Probabilistic Interdiction for Power Systems Kaarthik Sundar, Post-doctoral Researcher, Los Alamos National Laboratory, 3000 Trinity Drive, Apt 40, Los Alamos, NM, 87544, United States, kaarthik01sundar@gmail.com, Carleton Coffrin, Harsha Nagarajan, Russell Bent A probabilistic version of the N-k interdiction problem in power transmission networks is considered. The probability of failure of each component in the network is known a priori and the goal of the problem is to find a set of k components that maximizes disruption to the system loads weighted by the probability of simultaneous failure of the k components. The resulting problem is formulated as a bilevel mixed-integer nonlinear program. Convex relaxations, linear approximations, and heuristics are developed to obtain feasible solutions that are close to the optimum via a cutting-plane algorithm. Extensive numerical results corroborate the effectiveness of the proposed algorithms. 5 - An Auction-driven Mechanism for Electric Transmission Network Design This paper proposes a combinatorial auction formulation for electric transmission network design, where market participants can bid for the geographic location of their points of interconnection and its capabilities, according to their willingness to pay. The formulation proposed identifies the projects that maximize the social welfare, and obtains the network topology necessary to support those projects. The proposed approach allows to consider quantitatively the perspective of different market participants at the early beginning of transmission planning. This may reduce the negative consequences of transmission shortfall, without over building transmission infrastructure. Juan Andrade, PhD Student, University of Texas-Austin, 1616 Guadalupe Street, Austin, TX, 78701, United States, jandraderam@utexas.edu, Ross Baldick

362