1 上海科技大学物质科学与技术学院，上海 201210
2 中国科学院上海高等研究院，上海 201210
3 中国科学院上海应用物理研究所，上海 201800
4 张江实验室，上海 201210
束流轨道优化是短波长自由电子激光调试放大过程的关键环节。在实际实验中，需要花费大量的时间来调整参数，以校正轨道。为简化该多参数调优过程，研究了基于深度强化学习的自动优化技术，在仿真环境中使用SAC、TD3和DDPG算法调整多个校正磁铁，以优化自由电子激光的输出功率。为模拟实际实验中非理想的轨道状态，在第一节波荡器入口处设置一磁铁以偏转束流轨道。随后利用深度强化学习算法自动调节后续7个磁铁以校正轨道。结果表明，通过引入偏差将输出功率降低一个数量级后，基于最大熵原理的SAC算法将功率恢复到初始值的98.7%，优于TD3与DDPG算法。此外，SAC算法表现出更强的鲁棒性，有望后续应用在我国X射线自由电子激光装置中实现自动调束。
激光光学 自由电子激光 轨道校正 输出功率 深度强化学习 多参数优化 光学学报
2023, 43(21): 2114002
Author Affiliations
Abstract
1 College of Physics and Electronics, Shandong Normal University, Jinan 250014, China
2 College of Physics and Electronic Engineering, Qilu Normal University, Jinan 250200, China
An algorithm is proposed for the fast reconstruction of off-axis digital holograms based on a combination of complex encoding (CE) and spatial multiplexing (SM). In this algorithm, every two off-axis holograms recorded in sequence are first assembled into a CE hologram using the CE method, and then four of the CE holograms are again encoded into one complex spatial multiplexing (CSM) hologram based on the SM algorithm. It is demonstrated that the eight holograms encoded into such one CSM hologram can be quickly reconstructed by performing a two-dimensional (2D) Fourier transform (FT) on the CSM hologram. Using this method, the eight 2D FTs required for the reconstruction of the eight holograms in the conventional spatial filtering methods can be simplified to a process with only one 2D FT, which can largely improve the computation efficiency with the resolution of the reconstructed images nearly unchanged.
090.1995 Digital holography 090.4220 Multiplex holography 200.3050 Information processing Chinese Optics Letters
2016, 14(6): 060902
