Przekształcenia statyczne i dynamiczne w modelowaniu sygnałów i układów wykorzystującym dane eksperymentalne

Krystian Kubowicz

Abstract

The thesis deals with input preprocessing in signals and dynamic systems modelling which uses experimental data. What is considered is the need, neccessity and ways of input signal preprocessing in creation of neural models of dynamic systems. In the thesis it is shown that neural modelling of dynamic systems often requires the use of input signal preprocessing. Static as well as dynamic preprocessing is discussed. It is shown that in some situations it is not possible to create a model without an appropriate input signal preprocessing. The wiely used architecture with a series of unit delays may lead to the necessity of limitation of sampling period and/or the number of the delays. An architecture with all-pass filters instead of unit delays is proposed as a way of getting out of these limitations. An iterative algorithm is introduced in which the parameters of the all-pass filters are calculated. In every iteration the chosen parameter ensures that the relation of the obtained signal to the other signals is as close to the state of ortogonality as possible. The way how the algorithm works is illustrated by a numerical example. The notions of static and approximately static relationships between signals is defined. Based on these definitions four different indices of statics are introduced. They are used as a tool to determine whether a relationship between signals is approximately static. If the relationship is not approximately static then the indices show how far away from the state of statics is the given set of experimental data. Three of these indices are calculated based on the time values of the signals, while the fourth one uses mutliresolution analysis. Simulation examples are presented and discussed for all indices. It is shown how nonlinear transformations influence the spectrum of a signal. These results are used to develop a method which can be used to determine whether there exists a relationship between signals having a limited number of non-zero harmonic components.
Diploma typeDoctor of Philosophy
Author Krystian Kubowicz (FoEE)
Krystian Kubowicz,,
- Faculty of Electrical Engineering
Title in PolishPrzekształcenia statyczne i dynamiczne w modelowaniu sygnałów i układów wykorzystującym dane eksperymentalne
Languagepl polski
Certifying UnitFaculty of Electrical Engineering (FoEE)
Disciplineautomation and robotics / (technology domain) / (technological sciences)
Start date12-05-2004
Defense Date06-01-2010
End date27-01-2010
Supervisor Bartłomiej Beliczyński (FoEE / ICIE)
Bartłomiej Beliczyński,,
- The Institute of Control and Industrial Electronics

Internal reviewers Andrzej Dzieliński (FoEE / ICIE)
Andrzej Dzieliński,,
- The Institute of Control and Industrial Electronics
External reviewers Mirosław Świercz
Mirosław Świercz,,
-
Pages122
Keywords in Englishxxx
Abstract in EnglishThe thesis deals with input preprocessing in signals and dynamic systems modelling which uses experimental data. What is considered is the need, neccessity and ways of input signal preprocessing in creation of neural models of dynamic systems. In the thesis it is shown that neural modelling of dynamic systems often requires the use of input signal preprocessing. Static as well as dynamic preprocessing is discussed. It is shown that in some situations it is not possible to create a model without an appropriate input signal preprocessing. The wiely used architecture with a series of unit delays may lead to the necessity of limitation of sampling period and/or the number of the delays. An architecture with all-pass filters instead of unit delays is proposed as a way of getting out of these limitations. An iterative algorithm is introduced in which the parameters of the all-pass filters are calculated. In every iteration the chosen parameter ensures that the relation of the obtained signal to the other signals is as close to the state of ortogonality as possible. The way how the algorithm works is illustrated by a numerical example. The notions of static and approximately static relationships between signals is defined. Based on these definitions four different indices of statics are introduced. They are used as a tool to determine whether a relationship between signals is approximately static. If the relationship is not approximately static then the indices show how far away from the state of statics is the given set of experimental data. Three of these indices are calculated based on the time values of the signals, while the fourth one uses mutliresolution analysis. Simulation examples are presented and discussed for all indices. It is shown how nonlinear transformations influence the spectrum of a signal. These results are used to develop a method which can be used to determine whether there exists a relationship between signals having a limited number of non-zero harmonic components.
Thesis file
Kubowicz.pdf 1.56 MB
Citation count*5 (2020-09-18)

Get link to the record

Back
Confirmation
Are you sure?