Speaker
Description
During operation of free-electron laser (FEL) facilities, the beam trajectory is highly susceptible to parameter sensitivities, real-time drifts, and environmental disturbances, leading to significant degradation of beam quality. Conventional control methods rely on accurate system models and struggle to handle uncertainties, nonlinear couplings, and dynamic perturbations encountered in real-time operation. To address these challenges, this work proposes an online beam optimization method based on Multi-Agent Proximal Policy Optimization (MAPPO). By enabling collaborative policy learning among multiple agents within a shared environment, the approach achieves distributed, adaptive control of FEL subsystems, effectively captures inter-subsystem couplings, and sustains high performance under dynamic conditions. Experiments conducted at the Shanghai Soft X-ray Free-Electron Laser (SXFEL) facility demonstrate that, compared to PID control and single-agent reinforcement learning, the proposed MAPPO-based method achieves significantly improved long-term stability and exhibits superior robustness under high-noise and strong-perturbation scenarios. These results validate its practicality, reliability, and technical advantage for efficient online deployment in advanced FEL systems.
| In which format do you inted to submit your paper? | LaTeX |
|---|