Speaker
Alexander Zhukov
(Oak Ridge National Laboratory)
Description
To quickly and continuously minimize the beam loss at the spallation neutron source, we implement a reinforcement learning (RL) algorithm to control the tens of magnet settings based on the readbacks of tens of loss monitors.
To make this an operational procedure the can be safely used without damaging the accelerator, the RL is tested on a virtual accelerator and its settings go through a proxy gateway to keep the settings within predefined limits.
We describe the setup and testing of the RL algorithm running on a GPU cluster inside the accelerator network.
Funding Agency
DOE/BES
| I have read and accept the Privacy Policy Statement | Yes |
|---|
Primary author
Alexander Zhukov
(Oak Ridge National Laboratory)
