Physics-Informed Deep Neural Network for Inhomogeneous Magnetized Plasma Parameter Inversion
Yangyang Zhang, Haiyang Fu, Yilan Qin, Kangning Wang, Jiayu Ma
Abstract
Plasma parameter inversion is important for space plasma physics and applications, particularly for inhomogeneous magnetized plasmas. A physics-informed deep neural network for Maxwell’s plasma coupling system is proposed in this letter. The network architecture consists of inhomogeneous plasma parameter inversion and electromagnetic field reconstruction. We verified our physics-informed neural network method for one-dimensional (1-D) Maxwell’s plasma coupling system with inhomogeneous magnetized plasma parameters. The simulation results show that this meshless method can effectively achieve simultaneous inversion of inhomogeneous plasma parameter and global field based on sparse sampling. The physics-informed deep neural network for Maxwell’s plasma coupling system has a certain generalization ability, which may be applied for more complex plasma applications.