We describe a method for precisely regulating the gradient magnet power supply at the Fermilab Booster accelerator complex using a neural network trained via reinforcement learning. We demonstrate preliminary results by training a surrogate machine-learning model on real accelerator data to emulate the Booster environment, and using this surrogate model in turn to train the neural network for its regulation task. We additionally show how the neural networks to be deployed for control purposes may be compiled to execute on field-programmable gate arrays. This capability is important for operational stability in complicated environments such as an accelerator facility.
@article{arxiv.2011.07371,
title = {Real-time Artificial Intelligence for Accelerator Control: A Study at the Fermilab Booster},
author = {Jason St. John and Christian Herwig and Diana Kafkes and Jovan Mitrevski and William A. Pellico and Gabriel N. Perdue and Andres Quintero-Parra and Brian A. Schupbach and Kiyomi Seiya and Nhan Tran and Malachi Schram and Javier M. Duarte and Yunzhi Huang and Rachael Keller},
journal= {arXiv preprint arXiv:2011.07371},
year = {2021}
}
Comments
16 pages, 10 figures. Phys. Rev. Accel. Beams vol 24, issue 10. Published 18 October 2021. For associated dataset and data sheet see http://doi.org/10.5281/zenodo.4088982