HB2025 - the 71st ICFA Advanced Beam Dynamics workshop on High-Intensity and High-Brightness Hadron Beams

Research on Optimization of Beam Fault Compensation in CiADS Superconducting Section Based on Reinforcement Learning

THPT54

Oct 23, 2025, 3:30 PM

2h

third floor (conference center)

Poster Presentation WGB:Beam Dynamics in Linacs THPT poster session

Speaker

Tielong Wang (Institute of Modern Physics, Chinese Academy of Sciences)

Description

High reliability is a major challenge of high-current linear accelerators. This is particularly problematic for Accelerator Driven Systems (ADS) such as the China initiative Accelerator Driven System (CiADS). In order to achieve rapid beam recovery, it is necessary to adjust and compensate the superconducting solenoids and cavities adjacent to the failed components in superconducting linear accelerators. In this study, we employ the Soft Actor-Critic (SAC) algorithm, a reinforcement learning technique, to train a compensation model within a simulated environment of the CiADS superconducting section. Compared to previous methods utilizing genetic algorithms, the reinforcement learning approach demonstrates superior performance in delivering more stable and consistent results for beam dynamics control.