DATA-DRIVEN MODEL PREDICTIVE CONTROL FOR AUTOMATED OPTIMIZATION OF INJECTION INTO THE SIS18 SYNCHROTRON: DATA-DRIVEN MODEL PREDICTIVE CONTROL FOR AUTOMATED OPTIMIZATION OF INJECTION INTO THE SIS18 SYNCHROTRON

In accelerator labs such as GSI / FAIR, automating com-plex systems is the key to maximize the time spent on physicsexperiments. This study explores the application of a data-driven model predictive control (MPC) to refne the multi-turn injection (MTI) process into the SIS18 synchrotron, de-parting from conventional numerical optimization methods.MPC is distinguished by its reduced number of optimiza-tion steps and its superior ability to control performancecriteria, addressing issues like delayed outcomes and safetyconcerns – in this case septum protection. The study focuseson a highly sample-efficient MPC approach based on Gaus-sian processes, which lies at the intersection of model-basedreinforcement learning and control theory. This approachmerges the strengths of both fields, offering a unified andoptimized solution and yielding a safe and fast state-basedoptimization approach beyond classical reinforcement learn-ing and Bayesian optimization. Our study lays the ground-work for enabling safe online training for the SIS18 MTIissue, showing great potential to apply data-driven controlin similar scenarios.