Predictive control modeling of ADS's MEBT using BPNN to reduce the impact of noise on the control system.

• This paper is the first attempt of Deep Neural Network in China ADS injector II. • Explore the relationship the beam position and quadrupole magnet under noise. • Analyse the applicability of BPNN to the quadrupole magnet control model. • This paper provide a new method for beam debugging of the system. In the beam debugging of China ADS Injector II, the mathematical matrix operation, software simulation or physical derivation methods are mainly used. These methods do not take into account noise conditions, so it is difficult to accurately control the deflection position of the beam. For this reason, this paper takes MEBT as an example to establish a function model of quadrupole magnet control parameter and beam position, and based on the BPNN to fit the function model to reduce the influence of noise on the control process, and then realizes control prediction model in beam debugging. The experiments results show that the prediction accuracy of the quadrupole magnet control voltages Q2 and Q3 are 0.9072 and 0.9352, respectively, and the MSE is 0.0064. This paper is the first attempt of deep neural network in China ADS injector II, which provides a new method for beam debugging of the system. [ABSTRACT FROM AUTHOR]