Publication detail
ACTIVE MAGNETIC BEARING CONTROL THROUGH Q-LEARNING
BŘEZINA, T. KREJSA, J. KRATOCHVÍL, C.
Czech title
ACTIVE MAGNETIC BEARING CONTROL THROUGH Q-LEARNING
English title
ACTIVE MAGNETIC BEARING CONTROL THROUGH Q-LEARNING
Type
conference paper
Language
cs
Original abstract
The paper is focused on the control of active magnetic bearing using improved version of Q-learning. The improvement subsists in separating the Q-learning into two phases – efficient prelearning phase, which uses mathematical model of real system, and tutorage phase working with the real system and used for further improvement. Q-learning based controller is compared with PID controller and shows better results regarding the percentage of successful trials. When tutorage is applied the Q-learning based controllers show better results also in terms of control quality criterion. The policy found by learning express high robustness against errors of system variables observations.
Czech abstract
The paper is focused on the control of active magnetic bearing using improved version of Q-learning. The improvement subsists in separating the Q-learning into two phases – efficient prelearning phase, which uses mathematical model of real system, and tutorage phase working with the real system and used for further improvement. Q-learning based controller is compared with PID controller and shows better results regarding the percentage of successful trials. When tutorage is applied the Q-learning based controllers show better results also in terms of control quality criterion. The policy found by learning express high robustness against errors of system variables observations.
English abstract
The paper is focused on the control of active magnetic bearing using improved version of Q-learning. The improvement subsists in separating the Q-learning into two phases – efficient prelearning phase, which uses mathematical model of real system, and tutorage phase working with the real system and used for further improvement. Q-learning based controller is compared with PID controller and shows better results regarding the percentage of successful trials. When tutorage is applied the Q-learning based controllers show better results also in terms of control quality criterion. The policy found by learning express high robustness against errors of system variables observations.
Keywords in Czech
Q-learning, control, active magnetic bearing
Keywords in English
Q-learning, control, active magnetic bearing
RIV year
2003
Released
11.02.2003
Publisher
Ústav termomechaniky AV ČR
Location
Praha
ISBN
80-85918-81-1
Book
Dynamics of Machines 2003
Edition number
1
Pages count
8
BIBTEX
@inproceedings{BUT9747,
author="Tomáš {Březina} and Jiří {Krejsa} and Ctirad {Kratochvíl},
title="ACTIVE MAGNETIC BEARING CONTROL THROUGH Q-LEARNING",
booktitle="Dynamics of Machines 2003",
year="2003",
month="February",
publisher="Ústav termomechaniky AV ČR",
address="Praha",
isbn="80-85918-81-1"
}