THE CONTROL OF ACTIVE MAGNETIC BEARING USING TWO-PHASE Q-LEARNING

THE CONTROL OF ACTIVE MAGNETIC BEARING USING TWO-PHASE Q-LEARNING

Paper in proceedings (conference paper)

Q-learning method proved to be usable in active magnetic bearing (AMB) control task, however the learning speed remains the main problem. Two-phase variant of the Q-learning speeds up the learning process by dividing the method into two stages – prelearning stage which uses mathematical model of AMB and is highly efficient and tutorage stage which uses conventional Q learning and is used for further fining of the Q-function. Comparison with PID controller is included to demonstrate algorithms control abilities. As table (step function) is used as Q-function implementation, the optimum configuration of the table (number and relative positions of node points) was also investigated.

Q-learning method proved to be usable in active magnetic bearing (AMB) control task, however the learning speed remains the main problem. Two-phase variant of the Q-learning speeds up the learning process by dividing the method into two stages – prelearning stage which uses mathematical model of AMB and is highly efficient and tutorage stage which uses conventional Q learning and is used for further fining of the Q-function. Comparison with PID controller is included to demonstrate algorithms control abilities. As table (step function) is used as Q-function implementation, the optimum configuration of the table (number and relative positions of node points) was also investigated.

Q-learning, active magnetic bearing (AMB), control

BŘEZINA, T., KREJSA, J.

12.05.2003

Institute of Theoretical and Applied Mechanics, Academy of Sciences of the Czech Republic, Prague

Praha

80-86246-18-3

Engeneering Mechanics 2003

38

2