Revolutionizing Particle Accelerator Control: Machine Learning Enables Autonomous Operation and Efficient Management

Machine learning is now being leveraged to control particle accelerators, traditionally reliant on manual intervention for optimal operation. A team of researchers from the Institute of Modern Physics, Chinese Academy of Sciences, and Xiamen University have made headway in this field, as detailed in a study published in Science China: Physics, Mechanics & Astronomy.

The research, led by He Yuan and Zhao Hong, proposes a feasible pathway for “autonomous driving” of accelerators, addressing theoretical and technical challenges that have long impeded the integration of AI technology in this domain.

By developing a control-process-based dynamic model for accelerators and introducing a time-series phase-space reconstruction technique, the team ensured their control system captured equivalent global information, enhancing reliability and controllability.

Technically, they designed a high-precision virtual accelerator and a machine learning controller, employing reinforcement learning algorithms to process massive data generated by the virtual accelerator for offline training and seamless transition to real-world applications.

This marks the first application of AI technology in complex accelerator systems within China, setting a significant milestone for machine learning applications in this field. The research lays a solid foundation for further advancements in intelligent control technologies for accelerators, with future studies expected to expand the applicability of these theories and methods while developing more efficient machine learning algorithms, driving particle accelerator technologies to new heights.

The dynamics of particle accelerators are exceptionally rapid, making traditional nonlinear dynamical control theories inadequate for direct application. Moreover, as systems with extremely high degrees of freedom and spatiotemporal evolution, only partial variable information is observable, complicating the design and tuning of controllers.

He Yuan’s team from the Institute of Modern Physics, Chinese Academy of Sciences, in collaboration with Zhao Hong’s team from Xiamen University, proposed a feasible pathway for “autonomous driving” of accelerators. They developed a control-process-based dynamic model and introduced a time-series phase-space reconstruction technique to ensure the control system captures equivalent global information.

Technically, they designed a high-precision virtual accelerator and a machine learning controller. By employing reinforcement learning algorithms, they efficiently processed massive data generated by the virtual accelerator, enabling offline training of the controller and its seamless transition to real-world applications.

This research marks the first application of AI technology in complex accelerator systems within China, setting a significant milestone for machine learning applications in this field. The team achieved the first-ever global trajectory adaptive control of the CAFe2 superconducting segment, encompassing 42 degrees of freedom, which is now integrated into routine operations.