Detail publikace

Advantages of Neural Networks in Adaptive Control

Originální název

Advantages of Neural Networks in Adaptive Control

Anglický název

Advantages of Neural Networks in Adaptive Control

Jazyk

en

Originální abstrakt

This paper discusses the problems adaptive controllers have to face when working with a sampling period that is significantly shorter than the global time constant of the controlled system. A short sampling period is beneficial for disturbance cancellation, but it makes on-line identification of the system difficult in the presence of quantization effect, noise and disturbances. Neural networks present a promising approach to solving the problem. However, there remains a problem with extracting useful information about the system's dynamics in the form of training patterns for commonly used regressive models. Ways to enrich the training patterns with information about the system's behaviour are discussed.

Anglický abstrakt

This paper discusses the problems adaptive controllers have to face when working with a sampling period that is significantly shorter than the global time constant of the controlled system. A short sampling period is beneficial for disturbance cancellation, but it makes on-line identification of the system difficult in the presence of quantization effect, noise and disturbances. Neural networks present a promising approach to solving the problem. However, there remains a problem with extracting useful information about the system's dynamics in the form of training patterns for commonly used regressive models. Ways to enrich the training patterns with information about the system's behaviour are discussed.

BibTex


@inproceedings{BUT19696,
  author="Michal {Schmidt} and Petr {Pivoňka}",
  title="Advantages of Neural Networks in Adaptive Control",
  annote="This paper discusses the problems adaptive controllers have to face when working with a sampling period that is significantly shorter than the global time constant of the controlled system. A short sampling period is beneficial for disturbance cancellation, but it makes on-line identification of the system difficult in the presence of quantization effect, noise and disturbances. Neural networks present a promising approach to solving the problem. However, there remains a problem with extracting useful information about the system's dynamics in the form of training patterns for commonly used regressive models. Ways to enrich the training patterns with information about the system's behaviour are discussed.",
  address="Rektor der Hochschule Zittau/Gorlitz",
  booktitle="13th Zittau Fuzzy Coloquium",
  chapter="19696",
  institution="Rektor der Hochschule Zittau/Gorlitz",
  year="2006",
  month="october",
  pages="75",
  publisher="Rektor der Hochschule Zittau/Gorlitz",
  type="conference paper"
}