
THE FIELDS
INSTITUTE
FOR RESEARCH IN MATHEMATICAL SCIENCES 

Fields
Industrial Optimization Seminar
201112
held at the Fields
Institute, 222 College St., Toronto
Map
to Fields



The inaugural meeting of the Fields Industrial Optimization Seminar
took place on November 2, 2004. The seminar meets in the early evening
of the first Tuesday of each month. Each meeting is comprised of
two related lectures on a topic in optimization; typically, one
speaker is a universitybased researcher and the other is from the
private or government sector. The series welcomes the participation
of everyone in the academic or industrial community with an interest
in optimization – theory or practice, expert or student . Please
subscribe to the Fields mail list to be
informed of upcoming seminars.
201112
PAST SEMINARS 
May 1, 2012
5:00 p.m.
Fields Institute,
Room 230 
5:00 p.m.
Frédéric Meunier, CERMICS, Ecole Nationale
des Ponts et Chaussées, France
A routing problem raised by selfservice bicycle sharing systems
(slides)
Abstract: Operating bicyclesharing systems, such as the Vélib'
system in Paris or the Bixi system in Montréal or Toronto, raise
many challenging problems. The repositioning of the bicycles using one
or more truck is one of the most natural such problems. In this talk,
we focus on the case of a single truck. We are given a graph whose vertices
model the stations. Assuming that the current distribution of the bicycles
is known, we are to move the bicycles using the truck to reach a target
distribution at minimal cost. This problem corresponds to the situation
at the end of the night when very few bicycles are moving. The talk
will present special polynomial cases as well as approximation algorithms.
An efficient method solving reasonably sized practical instances will
be presented. The method combines the exact computation of a natural
lower bound and a local search exploiting theoretical properties of
the problem. Related open questions will be discussed.
Talk partially based on join works with Daniel Chemla and
Roberto Wolfler Calvo.

6:00 p.m.
Paul Raff and Prasad Kalyanaraman, Supply Chain Optimization
division of Amazon.com
Opportunity Cost Techniques and Fulfillment TieBreaking at Amazon.com
(slides)
For an online retailer that receives thousands of orders per
minute and focuses obsessively on customer satisfaction, Amazon.com
does not have the ability to optimize fulfillment decisions
over a time frame and must instead make fulfillment decisions
greedily, which is suboptimal. This talk will give a broad
detailed overview of the problem, and of the various ways
in which Amazon.com is attacking the problem. The majority
of the time will cover the fulfillment concept of tiebreaking,
which combines the theory of opportunity cost with the recognition
that in a large proportion of cases, no extra money need be
spent. Various examples will be given that exhibit the core
problem, and the theory of tiebreaking will be built up from
the basics. Various simulations – simple and complex
– will be shown that demonstrate the effectiveness of
tiebreaking.

April 3,
5:00 p.m.
Fields Institute,
Room 230 
5:00 p.m.
Ismael Regis de Farias, Texas Tech University
Cutting Planes for Some Nonconvex Combinatorial Optimization Problems
(slides)
In a wide variety of applications, a small number of nonconvex combinatorial
structures appear consistently, for example special ordered sets, cardinality,
and semicontinuous variables. These applications, besides being timely,
are strategic to our lifestyle and welfare. Examples include portfolio
optimization, sensing, and computational biology. I will discuss the use
of these constraints in practice and how integer programming can be used
to tackle them. I will emphasize modelling and the derivation of cutting
planes that make it possible solving to proven optimality industrystrength
instances of them, which otherwise would be unsolvable by the stateofthe
art methods.
_____________
6:00 p.m.
Yan Xu (SAS)
The MixedInteger Programming Solver and Solutions at SAS (slides)
SAS provides a suite of optimization tools which includes a
powerful algebraic modeling language, and a set of optimization
solvers for linear, mixedinteger, quadratic and nonlinear programs.
Based on these tools and leveraging the power of SAS in other
areas, a number of solutions have been successfully developed
for tackling industrial problems like optimizing retail pricing,
increasing marketing effectiveness, reducing inventory cost,
etc. In this talk, we first present some of the latest techniques
that we used to improve SAS mixedinteger programming solver.
Then, we show how to model and solve real world optimization
problem by using SAS optimization tools. Finally, we discuss
several challenges that we are facing in further improving optimization
products

March 6,
Fields Institute,
Room 230

5:00 p.m.
Michael Chen (York University)
A stochastic integer programming approach to the optimal thermal
and wind generator scheduling problem (slides)
In recent years, the increasing capacity of wind energy, together
with solar and other renewable energy, brings in a new challenge to
the electricity generator scheduling problem: the renewable energy is
stochastic by its nature and it is a daunting task to meet a stochastic
demand by a stochastic supply year around. In the near future, as we
approach the goal of 25% percent renewable energy by 2025, this challenge
will be more and more prominent. Sufficient reserve has been used to
achieve this goal in the past when the supply is fully controllable.
Will the same approach work with 25% stochastic supply? Do we need to
increase the reserve level? We will first model the complicated electricity
grid, physics of generators, stochastic demand and wind power, and the
dayahead decision process. Our model aims at a good balance of the
reality and computational complexity. Based on this stochastic integer
model, we develop an effective scenariocrossing deep cut, which accelerates
the stateofart CPLEX solver significantly.
_____________
Peter Hall (ArcelorMittal Dofasco)
The Practicality of Applying Optimization in the Steel
Industry
I will present 3 problems where optimization can be applied
in the Steel industry and specifically to areas within Supply
Chain Planning and Scheduling. Along with the presentation
of these problems, I will discuss the background behind these
problems and some of the pros and cons of using common optimization
modeling techniques to address these problems.
The first problem deals with the balancing of demand and supply
between Steel Making and Finishing operations within a fully
integratedSteel mill given that the manufacturing and business
objectives of both areas differ.
The second problem, campaign planning, addresses technological
constraints in an environment where these constraints hinder
the main KPI of ontime delivery and where optimization, although
very applicable, also conflicts with current work processes.
The third problem will be posed at a very high level that
is intuitive to applying common network optimization modeling
and solution techniques but which also poses some questions
about practicality of these techniques given knowledge base
of the work force, lack of cost and pricing information, and
the politics of using packaged software solutions vs. custom
inhouse applications.

February 7, 2012
Fields Institute,
Room 230

5:00 p.m.  7:00 p.m.
Kirsten Morris (University of Waterloo)
Optimal actuator location
Active noise control and control of structural vibrations
are examples of systems modelled by partial differential equations.
Because of the distribution of the system in space the location
of actuators, and sensors, in these systems is a variable
in the design of a control system. The talk will focus on
actuator location. The problem of locating sensors is mathematically
dual and will be mentioned briefly. The criterion for optimization
should be determined by the controller objective; common approaches
to controller design are linearquadratic, H2 and Hinfinity.
Since the controller is generally calculated using an approximation,
often of high order, determination of the optimal actuator/sensor
locations is not straightforward. Conditions for approximations
that can be used in determination off optimal actuator location
have been obtained. Recently developed algorithms for calculation
of the optimal actuator locations for several different control
objectives are discussed.
(audio and slides available here)
_____________
Oleksandr Romanko (Algorithmics Incorporated, an IBM Company,
Toronto)
ScenarioBased ValueatRisk Optimization
In financial risk management ValueatRisk (VaR) is a
popular tailbased risk measure which forms the basis for
regulatory capital according to Basel II Accord. Thus, optimizing
VaR can have benefits in terms of freeing up capital. The
problem is that VaR is a quantile of the loss distribution
(for a particular time horizon), which is a chanceconstrained
problem. Since the loss distribution is typically unknown
or computationally impractical, VaR optimization usually uses
a finite sample approximation to the distribution by means
of scenarios, so that an estimate of the VaR over a sample
scenario set is actually optimized. This, however, requires
mixedinteger programming, which makes the problem difficult
to solve.
To improve solution time, different heuristic techniques can
be used during optimization. We develop and test heuristic
algorithms for scenariobased VaR optimization. Due to high
computational complexity of VaR optimization, we utilize Conditional
ValueatRisk (CVaR)  based proxies for VaR objectives and
constraints. Our heuristic algorithm allows obtaining robust
results with low computational complexity.
This is joint work with Helmut Mausser from Algorithmics Incorporated,
an IBM Company.
(audio and slides available here)

December 6, 2011
Fields Institute,
Room 230

5:00  6:00 p.m.
Catalin Trenchea (University of Pittsburgh)
A stochastic collocation approach to constrained optimization
for random data estimation problems
We present a scalable, parallel mechanism for estimation
of statistical moments of input random data, given the probability
distribution of some response of a system of stochastic partial
differential equations. To characterize data with moderately
large amounts of uncertainty, we introduce a stochastic parameter
identification algorithm that integrates an adjointbased
deterministic algorithm with the sparse grid stochastic collocation
FEM. Rigorously derived error estimates are used to compare
the efficiency of the method with other techniques.
(audio and slides available here)
_____________
Szymon Buhajczuk (SimuTech Group Inc. Toronto)
Commercial Implementations of Optimization Software and
its Application to Fluid Dynamics Problems
ANSYS Computational Fluid Dynamics (CFD) tools such Fluent
and CFX have an established place in the engineering world
and are being used to help make design decisions early on
in product design cycles. Traditionally this has been a manual
process requiring engineering intuition and indepth understanding
of fluid dynamics. With the advent of powerful and inexpensive
computers, the optimization process can be increasingly automated
using commercial tools. Multi Disciplinary Optimization (MDO)
codes such as ANSYS Design Explorer and Red Cedar's HEEDS
are able to couple with existing CFD software and obtain optimal
designs faster. In comparison testing of Design Explorer and
HEEDS, both codes were benchmarked with an identical fluid
dynamics problem of external flow over a simplified car body.
The optimization exercise comprised of multiple geometric
design variables such as draft angle, body height above ground,
and roundness, with the ultimate design objective function
of minimizing drag. ANSYS Design Explorer created a design
response surface based on many arbitrary design point evaluations,
while HEEDS used an iterative proprietary search algorithm
to constantly refine the design based on previous evaluations.
In both cases, challenges arose in applying the methodologies
to a CFD problem due to long simulation run times required
to obtain a fluid flow solution. The performance of the two
codes appeared to greatly depend on the nature of the design
space. In the case of the car body, with many design variables,
the design space was complex enough that the HEEDS methodology
had an advantage over the use of a response surface approach.
In other problems, where multiple competing objective functions
are present, the ANSYS Design Explorer response surface approach
is superior, and can generate a Pareto Front tradeoff analysis
much quicker than the iterative approach. Overall, with the
significant computational cost of CFD simulations, both optimization
codes require preemptive and intelligent simplification of
the problem before starting the optimization process.(audio
and slides available here)

November 1, 2011
Fields Institute,
Room 230

5:00  6:00 p.m.
Hans J. H. Tuenter (University of Toronto)
The Modeling and Forecasting of Wind Energy
Wind energy is becoming a larger part of the electricity generation
mix, and with it the need for accurate forecasting of wind
energy. We describe a fundamental model that combines turbine
characteristics and meteorological forecasts. The model has
been used succesfully to provide shortterm forecasts for
a major utility.
(audio and slides available here)
_____________
6:00  7:00 p.m.
William (Bill) Smith (Siemens)
Wind Energy in Canada
Wind energy is playing an increasingly important role in the
generation mix in Canada. We describe the global market for
wind power, and the market in Canada on a provincial basis.
We outline the technological developments that are driving
the cost structure and efficiency of wind turbines. These
improved characteristics will bring wind energy closer to
parity with other generation sources.
(audio and slides available here)


October 4, 2011
Fields Institute,
Room 230

2:00  2:50 p.m.
Robert Stubbs (Axioma Inc.)
Factor Alignment Problems in Optimized Portfolio Construction
Construction of optimized portfolios entails the complex interaction
between three key entities: the risk factors, the alpha factors
and the constraints. The problems that arise due to mutual
misalignment between these three entities are collectively
referred to as Factor Alignment Problems (FAP). Examples of
FAP include riskunderestimation of optimized portfolios,
undesirable exposures to factors with hidden and unaccounted
systematic risk, consistent failure in achieving exante performance
targets, and inability to harvest high quality alphas into
aboveaverage IR. In this talk, we present a detailed investigation
of these alignment problems, discuss their sources, analyze
their effects on expost performance of optimized portfolios
and discuss a practical and effective remedy in the form of
augmented risk models.
(audio and slides available here)
_____________
2:50  3:40 p.m.
Thomas Coleman (University of Waterloo)
Risk Management of Portfolios by CVaR Optimization
The optimal portfolio selection problem is the fundamental
optimization problem in finance – optimally balancing
risk and return, with possible additional constraints. Unfortunately
the classical optimization approach is very sensitive to estimation
error, especially with respect to the estimated mean return,
and the resulting efficient frontier may be of little practical
value. Indeed it may be dangerous from a risk management point
of view. A popular alternative, usually under the banner of
“robust optimization” is ultraconservative and,
we argue, not really robust! In this sense it may also be
of questionable practical value. We propose an alternative
optimization approach – a CVaR optimization formulation
that is relatively insensitive to estimation error, yields
diversified optimal portfolios, and can be implemented efficiently.
We discuss this promising approach in this talk and present
strongly supportive numerical results.This is joint work with
Yuying Li and Lei Zhu
(audio and slides available here)

Back to top

