| Methodologies, September 13,
1999, 14:00 - 16:00 h
E: Preprocessing Techniques
Marc Thuillard, Siemens Building Technologies, Cerberus
Division, Switzerland
In most real world projects using computational
intelligence, preprocessing is certainly one of the most important steps
towards good and efficient algorithms. The main goal of preprocessing is to put
data under a form suitable for standard processing techniques. In this
tutorial, we discuss some of the most important preprocessing methods for
reducing the dimensionality and/or the complexity of a dataset (data
transforms, filtering, principal component analysis, neural networks,
discriminant analysis, projection pursuits, exploratory data analysis,
classification trees,...). In the second part, we will explain how
multiresolution analysis and wavelet theory is about to revolutionize
preprocessing. After an introduction on wavelet theory and multiresolution
analysis, we show how mutiresolution analysis can be used to preprocess the
data as a stand-alone technique or in conjonction to soft computing.
F: Active Decision Support
Christer Carlsson, IAMSR, Abo Akademi University, Finland
It is said that "in today's business you have to be big
or you have to react quickly". To react quickly and successfully is a
matter of knowledge, and the task to provide relevant, updated and useful
knowledge for management is the arena for developing, building and implementing
intelligent systems.
Modern intelligent systems will be based on fuzzy logic and
include the soft computing technology. Fuzzy logic has several benefits for
management purposes: (i) exact reasoning is viewed as a limiting case of
approximate reasoning; (ii) knowledge is interpreted as a collection of elastic
or fuzzy constraints on a collection of variables; (iii) inference is viewed as
a process of propagation of elastic constraints, and (iv) any logical system
has a fuzzy logic based counterpart.
There are two main characteristics of fuzzy systems that
give them better performance for specific applications: (i) fuzzy systems are
suitable for uncertain or approximate reasoning, especially when the systems
are difficult to describe with a mathematical model; (ii) fuzzy logic allows
problem solving and decision making with estimated values on the basis of
incomplete or uncertain information.
The design, development, building and implementation of
intelligent systems in management is aimed at improving the productivity of
working time for both individuals, teams and groups of knowledge workers and
managers.
Intelligent systems are also part of the emerging paradigm
for Active DSS (as opposed to "passive" / reactive DSS) - which is
probably more visible in Europe than in the U.S. This gradual paradigm shift
has been triggered by several factors: (i) the growing need for relevant and
effective decision support to deal with a dynamic, uncertain and increasingly
complex management environment (in both business and public sectors); (ii) the
effective DSS are typically context-tailored, not general purpose systems;
(iii) standard support technology is becoming obsolete and is making less and
less impact on real-life problem solving, and - therefore - (iv) support
technology has been increasingly losing enthusiasm among senior decision makers
as a way to improve decision quality and work productivity. The paradigm shift
towards Active DSS aims at tackling the above four issues.
G: Fuzzy constraint-based problem
solving and its application to scheduling
Didier Dubois, University Paul Sabatier, France
The most popular approach to decision-making in the setting
of fuzzy sets is the maximin ranking of solutions proposed by Bellman and Zadeh
. This method is natural when fuzzy sets model flexible constraints that cannot
compensate with one another. Then a paradigm generalizing the framework of
constraint satisfaction problems is obtained, that allows for the
representation of soft as well as prioritized constraints and for preference
propagation in constraint networks. The representation of prioritized
constraints is based on Sugeno integral. However the obtained ranking of
solutions is very coarse. Two refinements to this ordering have been
introduced: one focusing on the least satisfied discriminating constraint, and
the other involving a lexicographical ranking. The latter refines the former
and combines utilitarist and egalitarist points of view on the aggregation of
feasibility degrees. A generic approach to the computation of such solutions
will be described. While the maxmin paradigm makes sense when only flexible
constraints (represented by fuzzy sets) are involved, it no longer always
applies when some of the fuzzy sets represent uncertainty. In this situation a
specific decision theory has been developed where criteria make sense for
one-shot decisons and reflect the attitude of the decision-maker in front of
uncertainty.
The approach is particularly adapted to problems such as
sheduling where due-dates are flexible and duration of tasks are either
flexible or uncertain.
This tutorial introduces to these recent developments of
Bellman and Zadeh's setting.
H: Process Modeling and Control
Robert Babuska, Technical University of Delft, The
Netherlands
Modern production and manufacturing methods have increased
the performance requirements expected from the control systems. Over a broad
range of operating conditions, most processes exhibit strongly nonlinear
behavior, and cannot be adequately controlled by using conventional techniques.
This tutorial gives a state-of-the-art overview of fuzzy
modeling and control techniques in the context of the identification and
control of nonlinear processes. Different types of fuzzy models are presented
and compared. Methods to construct fuzzy models by integrating prior knowledge
and process data are
reviewed, including neuro-fuzzy identification techniques,
fuzzy clustering, non-linear optimization and parameter-estimation methods.
Semi-mechanistic approaches based on a combination of ``first principles'' and
fuzzy modeling and neural network techniques are discussed. Methods to design
nonlinear control systems based on available fuzzy models are addressed,
including local linear design, internal model control and model-based
predictive control. The described methods are illustrated by examples. Case
studies and applications are presented.
|
