The KEK e⁺/e⁻ Linac supplies electron beams to SuperKEKB HER, PF, and PF-AR, and positron beams to SuperKEKB LER. We utilize machine learning for both online beam tuning and offline data analysis. Machine learning based on Bayesian optimization has been employed to improve and maintain beam quality, contributing to the enhancement and stabilization of beam injection efficiency into SuperKEKB HER and LER. In this report, we present monitors and beam tuning methods that incorporate machine learning. Identifying parameters that affect beam quality and stability is important, but finding them among the vast number of parameters is not easy. In machine learning-based beam tuning, selecting the appropriate parameters for tuning is crucial, and another important issue is identifying factors that lead to beam instability. To address this, we have applied explainable AI techniques to analyze archived data and attempted to extract parameters that have a significant impact on the beam. This report also covers our data archiving system and analysis efforts using explainable AI.
Miyahara et al. (Tue,) studied this question.