2016 INFORMS Annual Meeting Program

127 / 561

INFORMS Nashville – 2016
127
4 - Robust Empirical Optimization
Andrew Lim, National University of Singapore, Singapore,
Singapore, , Jun-ya Gotoh,
Michael Jong Kim
We analyze the out-of-sample performance of robust empirical optimization and
show that it asymptotically optimizes the mean-variance reward under the data
generating model. We also introduce the notion of robust cross-validation as a
method of calibrating the empirical robust optimization problem. Theoretical and
experimental comparison to empirical optimization is also provided.
MA17
105B-MCC
Optimizing Power System Operations
Under Uncertainty
Sponsored: Optimization, Optimization Under Uncertainty
Sponsored Session
Chair: Yu Zhang, University of California, Berkeley, 200 California Hall,
Albany, CA, 94720, United States,
1 - Unit Commitment Under Gas-supply Uncertainty And
Gas-price Variability
Antonio J. Conejo, The Ohio State University, 1971 Neil Av.,
Columbus, OH, 43210, United States,
Ramteen Sioshansi, Bining Zhao
We propose a two-stage stochastic optimization model to analyze the scheduling
of power units under natural gas-supply uncertainty and natural gas-price
variability. The first stage of this model represents the day-ahead scheduling stage,
while the second stage represents real-time operations via scenarios. We use this
model to analyze the effect of two types of gas-supply conditions. First, we
analyze a case involving low-cost gas supply with gas-transmission issues. We
then examine a case involving higher-cost gas supply, which is used solely to
attain feasibility with fast ramping events.
2 - Stochastic Market Clearing With High-penetration Wind Power
Yu Zhang, University of California, Berkeley,
, Georgios Giannakis
Integrating renewable energy into the modern power grid requires risk-cognizant
dispatch of resources to account for the stochastic availability of renewables.
Toward this goal, day-ahead stochastic market clearing with high-penetration
wind energy is pursued in this work. The objective is to minimize the social cost,
which consists of conventional generation costs, end-user disutility, and a risk
measure of the system re-dispatching cost based on the conditional value-at-risk.
The resulting convex optimization problem is solved via the sample average and
the alternating direction method of multipliers. Numerical results corroborate the
merits of the proposed approaches.
3 - Integrated Generator Maintenance And Operations Scheduling
Under Uncertain Failure Times
Beste Basciftci, Georgia Institute of Technology, Atlanta, GA,
United States,
Murat Yildirim, Shabbir Ahmed, Nagi Gebraeel
In this study, we formulate an integrated generator maintenance and operations
scheduling problem as a stochastic mixed integer program by considering
unexpected failure times. We generate failure scenarios based on the remaining
life time distributions of the generators. We adopt a scenario decomposition
approach to solve this problem in a distributed framework that identifies and
evaluates solutions by solving scenario subproblems. Finally, we present
computational experiments demonstrating the effectiveness of the approach.
MA18
106A-MCC
Theory of Integer Optimization
Sponsored: Optimization, Integer and Discrete Optimization
Sponsored Session
Chair: Alberto Del Pia, University of Wisconsin, Madison,
Madison, WI, United States,
1 - Facet Separation With One Linear Program
Laurence Wolsey, Université catholique de Louvain,
, Michele Conforti
Given polyhedron P and and a point x*, the separation problem for polyhedra
asks to certify that x* in P and if not, to determine an inequality that is satisfied
by P and violated by x*. This problem is central in cutting plane methods for
Integer Programming and the “quality” of the violated inequality is an essential
feature in the performance of such methods. In this talk we address the problem
of finding efficiently an inequality that is violated by x* and either defines an
improper face or a facet of P. We provide some evidence that our method works
on structured and unstructured problems.
2 - Ellipsoidal Mixed-Integer Representability
Jeffrey Poskin, University of Wisconsin - Madison, Wisconsin, WI,
United States, , Alberto Del Pia
Representability results for mixed-integer linear systems play a fundamental role
in optimization since they give geometric characterizations of the feasible sets that
can be formulated by mixed-integer linear programming. We consider a natural
extension of mixed-integer linear systems obtained by adding just one ellipsoidal
inequality. The set of points that can be described, possibly using additional
variables, by these systems are called ellipsoidal mixed-integer representable. In
this work, we give geometric conditions that characterize ellipsoidal mixed-
integer representable sets. “
MA19
106B-MCC
Evaluating the Performance of Optimization Solvers
Sponsored: Computing
Sponsored Session
Chair: Paul Brooks, Virginia Commonwealth Univ, Virginia
Commonwealth Univ, Richmond, VA, 00000, United States,
1 - IPET: Interactive Performance Evaluation Tools For Benchmarking
Optimization Software
Gregor Hendel, ZIB,
The optimization community has recently seen an increasing number of non-
standard benchmark measures for evaluating solver performance, which some
data processing tools do not deliver. The benchmark evaluation starting from the
parsing of raw solver log data until the preparation of publication-ready tables can
be time-consuming and error-prone if done by hand. In this talk, we will first
review some of the specialized benchmark measures used for optimization
software. We will then present a Python tool named IPET to speed-up repetitive
tasks during benchmarking. We will give an overview of the tool and show
examples how automated benchmark scripts can be created using IPET.
2 - Surrogate Models For Algorithm Configuration
Meinolf Sellmann, IBM Research, Yorktown Heights, NY,
United States,
Automatic algorithm configurators are important tools for improving program
performance. Local search approaches in particular have proven very effective for
tuning. We study the use of non-parametric models in the context of population-
based algorithm configurators. We introduce a model designed specifically for the
task of predicting high-performance regions in the parameter space. Moreover,
we introduce the ideas of genetic engineering of offspring. Numerical results show
that model-based genetic algorithms significantly improve our ability to
effectively configure algorithms automatically.
3 - Tuning Of Optimization Software Parameters For Mixed Integer
Programming Problems
Toni P Sorrell, Virginia Commonwealth University, Richmond, VA,
United States, , Paul Brooks, David Edwards
The tuning of optimization software is of key interested to researchers solving
mixed integer programming (MIP) problems because the efficiency of the
optimization software is can be greatly impacted by the solver’s parameter settings
and the structure of the MIP. A designed experiment approach is used to fit a
model that would suggest settings of the parameters that provided the greatest
impact on the primal integral. Primarily, this research focuses on using classes of
MIPs to not only obtain good parameter settings for a practitioner to use on
future instances of the same class of MIPs, but to also gain understanding of why
the settings work well for that class of MIPs.
4 - Deploying MPL Optimization Models With Google
Web Services API’s
Bjarni Kristjansson, President, Maximal Software Inc., 2111
Wilson Boulevard, Suite 700, Arlington, VA, 22201, United States,
Over the past decade the IT has been moving steadfastly towards utilizing
software on clouds using Web Services API’s. The old standard way of deploying
software on standalone computers is slowly going away. Google has been one of
the leading software vendors in this area and publishes several web API’s which
can be quite useful for deploying optimization applications. In this presentation
we will demonstrate several Google API’s, including the Google Sheets API,
Google Maps API, and Google Visualization API and show how they can be
integrated with the MPL OptiMax Library for deploying optimization to service
both web and mobile clients.
MA19