WEOT02 　加速器制御　 7月31日 テルサホール 15:30-15:50

Multi-Objective Bayesian Optimization of Electron Cyclotron Resonance Ion source.

○Andrea De Franco, Tomoya Akagi (QST), Benoit Bolzon (CEA), Fabio Cismondi (F4E), Nicolas Chauvin (CEA), Herve Dzitko (F4E), Tomonobu Itagaki, Keitaro Kondo, Kai Masuda (QST)

ECR ion sources require tuning by an expert to achieve best performance. We developed a Multi-Objective Bayesian optimization for the ECR of LIPAc. The free parameters are: RF power, gas flow, position of 2 RF tuners and current of 2 solenoid coils. The machine learning approach demonstrated a fast convergence to a working point where not only the extracted beam current is >130mA, but also the emittance is successfully constrained to <0.25 π mm mrad and the intra-pulse and inter-pulse current fluctuations are <3mA. We present the detailed algorithm, testing methodology, results achieved and encountered challenges posed by the dimensionality of the problem, hysteresis and evolving state of the system.