Research on Reconstruction Method of Plasma Parameters Based on P-U-Net Model

Plasma characteristic parameters are the basis for studying plasma properties and are important in engineering applications. A deep learning-based method, the P-U-net model, is proposed to meet the need for fast and efficient reconstruction of plasma parameters. The network is trained by improving the traditional U-net model and adding the previous information containing the target features into the scattered field. Several representative tests are carried out in the study, including both single-target plasma and multitarget plasma, to evaluate the performances of the proposed model. The simulation results demonstrate that the P-U-net model has obvious advantages in plasma parameter reconstruction compared with the classical iterative method and U-net model which only trains the real part of the scattered electric field. The mean absolute error of the reconstructed parameters of the P-U-net model is lower than other traditional methods. The proposed model is expected to provide more substantial theoretical support for further improving the real-time performance and detection accuracy of plasma parameters measurement.