Q-LEARNING USED FOR CONTROL OF AMB: REDUCED STATE DEFINITION

Q-LEARNING USED FOR CONTROL OF AMB: REDUCED STATE DEFINITION

Paper in proceedings (conference paper)

A great intention is lately focused on Reinforcement Learning (RL) methods. Previous work showed that stochastic strategy improved model free RL method known as Q-learning used on active magnetic bearing (AMB) model. So far the position, velocity and acceleration were used to describe the state of the system. This paper shows simplified version of controller which uses reduced state definition - position and velocity only. Furthermore the controlled initial conditions area and its development during learning are shown. Numerical experiments proved that simplified controller version is fully capable of AMB control.

A great intention is lately focused on Reinforcement Learning (RL) methods. Previous work showed that stochastic strategy improved model free RL method known as Q-learning used on active magnetic bearing (AMB) model. So far the position, velocity and acceleration were used to describe the state of the system. This paper shows simplified version of controller which uses reduced state definition - position and velocity only. Furthermore the controlled initial conditions area and its development during learning are shown. Numerical experiments proved that simplified controller version is fully capable of AMB control.

Reinforcement Learning, Q-learning, Active Magnetic Bearing

Reinforcement Learning, Q-learning, Active Magnetic Bearing

BŘEZINA, T., KREJSA, J.

05.06.2002

Brno University of Technology, Faculty of Mechanical Engineering

Brno

80-214-2135-5

Mendel 2002

347

6