Palestras da Escola de Altos Estudos
System Identification
Lennart Ljung Automatic Control Linköpings universitet ABSTRACT System identification is the art and science of building mathematical models of dynamical systems based on observed inputs and outputs. In this presentation, the basic concepts and tools for this process will be described. The talk will focus on the basic three entities: the model structure, the data set and principles for fitting models to data.
Both linear and nonlinear models will be discussed. Of particular importance is how to gain confidence in the estimated model: Model validation.
The talk will also deal with the practical side of system identification: What are the software tools to efficiently construct models from data, and how does one make sure that the experimental data contains relevant information?
Discrete
Event Systems
ChristosG. Cassandras
Division of Systems
Engineering,
Department of Electrical
and Computer Engineering, and
Center for Information
and Systems Engineering (CISE)
Boston University
ABSTRACT
We will contrast time-driven to event-driven systems
so as to motivate new modeling frameworks and associated techniques developed
for the analysis, control, and optimization of Discrete Event Systems (DES).
Formal models for DES will be presented with emphasis on automata, timed
automata, and stochastic timed automata, which are best suited to tackle
resource contention problems commonly encountered in computer, communication,
and sensor networks, in manufacturing, in transportation, and in new problems
related to energy management. Discrete event simulation will be presented as a
direct implementation of stochastic timed automata. The uses and limitations of
simulation-based methods will be discussed.
The absence of
analytical, closed-form expressions quantifying the performance of most DES of
interest has motivated the development of techniques that take maximal
advantage of the properties of DES state dynamics. This includes the ability to
carry out efficient sensitivity estimation for performance metrics with respect
to critical design and control parameters. Resulting techniques lead to “rapid
learning” of the behavior of such systems under different parameter settings
without having to actually simulate or implement all such settings. We will
present the theoretical foundations of Infinitesimal Perturbation Analysis
(IPA) and Concurrent Estimation (CE) and their use for control and optimization
purposes.
Despite many
advances, DES remain highly complex to analyze and recent efforts have focused
on developing abstraction models and methods that extract the salient features
of a DES through hybrid dynamic system models. Recent developments will be
presented on an “IPA calculus” for stochastic hybrid systems and its robustness
properties that allow us to now study highly complex DES through proper
modeling abstractions and new analysis tools. Applications to various control
and optimization problems for DES will be included.
Nonlinear Systems
Hassan K. Khalil
University Distinguished Professor
Department of Electrical and Computer Engineering Michigan
State University
ABSTRACT
This tutorial talk will give an introduction to nonlinear systems and control. It will describe essentially nonlinear phenomena and Lyapunov’s method for studying stability. It will present techniques for nonlinear control design, including feedback linearization, backstepping, passivity-based control, and sliding mode Control. Finally it will present techniques for nonlinear observer design, including observers with linear error dynamics and high-gain observers, and will illustrate the use of such observers in output feedback control.
It is assumed that attendees have undergraduate knowledge of control systems and differential equations. Knowledge of state models of linear systems is helpful.
Power System Automation: Challenges and
Opportunities
Om P. Malik, Ph.D., LFIEEE
University
of Calgary
ABSTRACT
Modern electric power
systems have developed into very large and complex systems. The fundamental
principle in the effective operation of interconnected power systems requires
that electric power generation and load remain balanced at all time.
When a disturbance
occurs, protection and control actions are required to minimize the impact of
the disturbance and restore the system to a normal state. Control and
protection actions can be taken at various levels of the power system hierarchy
depending on the type and severity of the disturbance.
At the upper levels,
the control center operators must deal with large scale power system problems
in a very complex situation. Many times they rely on heuristic solutions and
policies. Systems with automated solutions designed to detect predetermined
system conditions having a high probability of causing stress on the power
system and provide system-wide control and protection solutions are being
developed.
Local protection and
control devices at the generation and sub-station levels arrest the propagation
of emergencies through automatic actions by addressing local system or
equipment specific problems.
A broad understanding
of the power system specific problems, and protection, control and automation
measures being employed and under development to improve power system integrity
and reliability at both the system and local levels, will be provided in this
presentation.
Intelligent Systems: Analysis and Design-A Perspective
of Computational Intelligence
Witold Pedrycz
Department of Electrical & Computer Engineering
University of Alberta, Edmonton Canada
and
Systems Research Institute, Polish Academy of Sciences
Warsaw, Poland
ABSTRACT
The talk
offers a comprehensive introduction to intelligent systems along with their
analysis, design methodologies and practice. We start with a concept of
intelligent systems and review existing definitions (descriptions) of such
architectures. Then we look at the key features of intelligent systems as well
as discuss several representative examples of practical relevance. We cast the analysis and design of
intelligent systems in the framework of Computational Intelligence (CI) and
demonstrate that CI offers a comprehensive environment supporting both system
analysis and design. A special attention is paid to various ways of knowledge
representation along with an evaluation of the features of the corresponding
knowledge representation schemes to facilitate or enhance learning abilities in
intelligent systems. A number of selected commonly encountered learning
mechanisms are elaborated on and assessed vis-à-vis a nature of the problem at
hand. Intelligent classifiers regarded as an important category of intelligent
systems are investigated. A class of collaborative intelligent systems is
discussed as well.
The
talk is self-contained and all required prerequisites will be covered
Fluid & Hybrid Petri Nets
Manuel Silva
Universidad de
Zaragoza
ABSTRACT
The talk will start providing a presentation of the
fluidization of discrete event dynamic systems (DEDS), as a relaxation (or over-approximation) technique for dealing with the classical state
explosion problem. This is particularly interesting when dealing with DEDS
provided with large populations such in some manufacturing, traffic, logistics
or population systems.
Even if named as continuous or fluid,
the obtained relaxed models are frequently hybrid
in a technical sense. Thus, techniques used for logical verification,
performance evaluation or control studies in discrete, hybrid and continuous
models can be adapted in some sense. Moreover, the possibilities for
transferring concepts and techniques from one modelling paradigm to others are
very significant, so there is much space for synergy. As a central modelling
paradigm for concurrent and synchronized DEDS, Petri nets (PNs) will be
considered.
Being a relaxation of DEDS, the analysis
and synthesis problems on fluid models are frequently much more tractable at
the computational level. Nevertheless, timed fluid PNs (under so called infinite server semantics) are able to
simulate Turing Machines (!), so
great “expressive power” and “undecidabilities” appear in the horizon. Being an
approximation, problems like the loss of some discrete properties by means of
fluidization will be considered. Otherwise stated, not all Petri Nets allows a “reasonable approximation” by
fluidization (like not all ordinary
differential equations, even provided with constant coefficients, allows a
reasonable linearization, for example). Equally important, “non-monotonic
behaviours” in untimed and timed models will be stressed, what raises the
importance of control strategies. After discussing observability and controllability
issues, we will conclude with some remarks and possible directions for future
research.
Among the aspects that distinguish the
adopted approach are: the focus on the relationships between discrete and
continuous PN models, both for untimed, i.e., fully
non-deterministic abstractions, and timed versions; the use of structure
theory of (discrete) PNs, algebraic and graph based concepts and results;
and the bridge to Automatic Control Theory.
|