JACoW is a publisher in Geneva, Switzerland that publishes the proceedings of accelerator conferences held around the world by an international collaboration of editors.
TY - CONF AU - Cai, M. AU - Feng, C. AU - Gu, D. AU - Tu, L. AU - Zhang, K.Q. AU - Zhao, Z.T. AU - Zhu, Z.H. ED - Zimmermann, Frank ED - Tanaka, Hitoshi ED - Sudmuang, Porntip ED - Klysubun, Prapong ED - Sunwong, Prapaiwan ED - Chanwattana, Thakonwat ED - Petit-Jean-Genaz, Christine ED - Schaa, Volker R.W. TI - Twin Delayed Deep Deterministic Policy Gradient for Free-electron Laser Online Optimization J2 - Proc. of IPAC2022, Bangkok, Thailand, 12-17 June 2022 CY - Bangkok, Thailand T2 - International Particle Accelerator Conference T3 - 13 LA - english AB - 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. PB - JACoW Publishing CP - Geneva, Switzerland SP - 1025 EP - 1028 KW - FEL KW - electron KW - laser KW - network KW - undulator DA - 2022/07 PY - 2022 SN - 2673-5490 SN - 978-3-95450-227-1 DO - doi:10.18429/JACoW-IPAC2022-TUPOPT013 UR - https://jacow.org/ipac2022/papers/tupopt013.pdf ER -