Application of deep learning methods for online beam optics and coupling feedback in synchrotron light sources

This paper presents a beam optics and coupling feedback system for storage rings during operation, integrating deep learning methods with turn-by-turn (TBT) beam position monitor data. A deep learning network has been applied to extract phase advance and betatron coupling information implicit in TBT data. Subsequently, these data were fed into a multilayer neural network for rapid modeling and correction. The simulation data and sampled spectrum data from the Shanghai Synchrotron Radiation Facility (SSRF) were used to create training datasets for the network. The efficacy of this feedback system has been successfully demonstrated at the SSRF, enabling continuous monitoring and correction during each injection cycle. This method reduces optical distortion to less than 1% after correction, effectively suppresses optical and coupling distortion caused by insertion device movements during storage ring operation, and ensures robust performance under various operational conditions.