Machine learning for beam correction study of the injection beamline at Wuhan Advanced Light Source

摘要:         As a fourth-generation synchrotron radiation light source working at 1.5 GeV, Wuhan Advanced Light Source (WALS) is being designed, which uses a full-energy linear accelerator (LINAC) as its electron beam injector. The injection beamline adopts a three-stage scheme: firstly, the beam from the LINAC that is 6 m under the storage ring is horizontally deflected below the storage ring, then it gradually climbs from underground to the same altitude as the storage ring, and finally the beam is delivered horizontally into the injection straight section inside the storage ring. Meanwhile, the Twiss parameter matching between the LINAC and storage ring is completed. During the construction of the beamline, magnet manufacturing errors, installation errors and beam injection errors from the LINAC will cause beam deviations from predetermined ideal orbits, and even particle losses. As a result, the electron beam correction is required during beam commissioning. Different from the single-direction beam correction of general transfer lines, the horizontal and vertical directions of the beam are coupled in the WALS injection transfer line, which greatly increases the complexity and difficulty of beam correction. Machine learning technology has been developed extensively in recent years, and its powerful algorithm of invertible neural network model is expected to be able to solve the beam commissioning difficulty of the beam injection transfer line at the WALS. Therefore, an invertible neural network model has been designed and trained to simulate the beam transport and beam correction of the WALS injection beamline. By optimizing the number and location of beam profile diagnostics, the accuracy of bidirectional prediction and beam correction effect can be greatly improved. The method is of great practical significance for the commissioning and operation of similar complex beam transport systems.