Location of Underground Multilayer Media Based on BP Neural Network and Near-Field Electromagnetic Signal

Underground near-field electromagnetic positioning system is based on near-field electromagnetic ranging (NFER) technology. Obtaining the thickness of underground multilayer media during the positioning process introduces an ill-posed problem in the traditional electromagnetic model due to matrix inversion. Therefore, we propose a positioning method based on backpropagation (BP) neural network, which avoids the matrix inversion and contains the ranging model and positioning model. The positioning model is based on the ranging model and simultaneously predicts the ranging value and the thickness of each layer. The positioning is realized by utilizing the thickness value and range value combined with 2D-direction of arrival (DOA). This positioning method simplifies the positioning process compared with the trilateration algorithm. The results of the validation sets show that the positioning accuracy can reach 0.6 m at a depth of 40 m. Garson algorithm performs sensitivity analysis on the BP neural network, and it can be concluded that the signal angle contributes the most to the prediction of the results. Ultimately, the results reveal that the BP neural network-based positioning method performs well in nonhomogeneous media (NH) environments across various depth spaces and signal-to-noise ratios (SNRs).