Using machine learning methods to analyze data related to CSNS RCS beam transmission efficiency

Purpose As a high-intensity hadron accelerator-based user facility, minimizing the radiation dose induced by uncontrollable beam loss at the China Spallation Neutron Source (CSNS) is crucial for manual maintenance by operators. A correlation has been observed between the beam transmission efficiency of the Rapid Cycling Synchrotron (RCS) and outdoor temperature. Given that the only RCS components located outdoors are the resonant power supply systems, further investigation into this relationship is necessary to identify opportunities for improving beam transmission efficiency during operation.
Methods To address the nonlinear relationship between the resonant power supply variables (amplitude and phase) and beam transmission efficiency, this study employs machine learning techniques. Decision Tree-based algorithms, including Extra Trees, Gradient Boosting Decision Trees (GBDT), Random Forest, and LightGBM, are used for the analysis. These methods enable a statistical examination of the most influential factors affecting beam transmission efficiency in the RCS, with a focus on the amplitude and phase of the resonant power supply.
Results The analysis highlights the significance of two amplitudes from higher-order harmonic components in affecting transmission efficiency. It is suggested that increasing the RCS transmission efficiency can be achieved by adjusting these critical amplitudes.
Conclusion Adjusting the K3 amplitude of QPS02 and the K2 amplitude of BPS01 has the potential to significantly improve beam transmission efficiency in the RCS, contributing to more efficient operations.