Application of reinforcement learning in synchrotron power supply synchronization correction

Abstract

As a powerful tool, machine learning has promoted the development of natural science in many fields and it also has helped engineers make many remarkable achievements in industry. Recently, using machine learning has already emerge in control of heavy ion accelerators. By studying synchronization of heavy ion accelerators, a method based on reinforcement learning is proposed. It’s a method that can make automatic synchronization correction. The action of power supply is simulated to interact with agent. As a precondition of synchronization correction processing, a novel approach is proposed to identify the slower power supply. Experimental results have show that our method can automatically identify the slower power between power supplies and it can make power supply complete synchronization through interaction. Compared with the past method, our algorithm not only saves manpower but also increaing the accuracy of synchronization.