ALPI-PIAVE performance at INFN-LNL with advanced optimization algorithms

The ALPI-PIAVE heavy-ion accelerator at INFN-LNL requires precise tuning due to its complex beam dynamics and high-dimensional parameter space. Although designed for a transmission efficiency of 60%, actual performance is typically limited to 15–20%, highlighting the need for more effective optimization strategies. In this work, we evaluate Bayesian Optimization (BO) and Trust Region Bayesian Optimization (TuRBO) to improve the tuning efficiency and robustness of the PIAVE injector line. To address the sensitivity of BO-based methods to parameter drifts, we incorporate Particle Swarm Optimization (PSO) and introduce controlled perturbations to assess algorithm performance under dynamic conditions. The results demonstrate the potential of these machine learning techniques for real-time accelerator optimization.