Automated Intensity Optimisation Using Reinforcement Learning at LEIR

High intensities in the CERN Low Energy Ion Ring (LEIR) are achieved by stacking up to seven consecutive multi-turn injections from Linac3. Two inclined septa combined with a collapsing horizontal orbit bump allow a 6-D phase space painting via a linearly ramped mean momentum along the Linac3 pulse and injection at high dispersion. The beam is cooled and dragged longitudinally via electron cooling (e-cooling) into a stacking momentum. For optimal accumulation, the electron energy and trajectory need to match the ion energy and orbit at the e-cooler section. In this paper, a reinforcement learning (RL) agent is trained to adjust various e-cooler and Linac3 parameters to maximise the intensity at the end of the injection plateau. Variational Auto-Encoders (VAE) are used to compress longitudinal Schottky spectra into a compact representation as input for the RL agent. The RL agent is pre-trained on a surrogate model of the LEIR e-cooling dynamics, which in turn is learned from the data collected for the training of the VAE. The performance of the VAE, the surrogate model, and the RL agent is investigated in this paper. An overview of planned tests in the upcoming LEIR runs is given.