JACoW is a publisher in Geneva, Switzerland that publishes the proceedings of accelerator conferences held around the world by an international collaboration of editors.
@inproceedings{cai:ipac2022-tupopt013,
author = {M. Cai and C. Feng and D. Gu and L. Tu and K.Q. Zhang and Z.T. Zhao and Z.H. Zhu},
% author = {M. Cai and C. Feng and D. Gu and L. Tu and K.Q. Zhang and Z.T. Zhao and others},
% author = {M. Cai and others},
title = {{Twin Delayed Deep Deterministic Policy Gradient for Free-electron Laser Online Optimization}},
booktitle = {Proc. IPAC'22},
% booktitle = {Proc. 13th International Particle Accelerator Conference (IPAC'22)},
pages = {1025--1028},
eid = {TUPOPT013},
language = {english},
keywords = {FEL, electron, laser, network, undulator},
venue = {Bangkok, Thailand},
series = {International Particle Accelerator Conference},
number = {13},
publisher = {JACoW Publishing, Geneva, Switzerland},
month = {07},
year = {2022},
issn = {2673-5490},
isbn = {978-3-95450-227-1},
doi = {10.18429/JACoW-IPAC2022-TUPOPT013},
url = {https://jacow.org/ipac2022/papers/tupopt013.pdf},
abstract = {{X-ray free-electron lasers (FEL) have contributed to many frontier applications of nanoscale science which benefit from its extraordinary properties. During FEL commissioning, the beam status optimization especially orbit correction is particularly significant for FEL amplification. For example, the deviation between beam orbit and the magnetic center of undulator can affect the interaction between the electron beam and the FEL pulse. Usually, FEL commissioning requires a lot of effort for multi-dimensional parameters optimization in a time-varying system. Therefore, advanced algorithms are needed to facilitate the commissioning procedure. In this paper, we propose an online method to optimize the FEL power and transverse coherence by using a twin delayed deep deterministic policy gradient (TD3) algorithm. The algorithm exhibits more stable learning convergence and improves learning performance because the overestimation bias of policy gradient methods is suppressed.}},
}