Inversion Method Of A Highly Generalized Neural Network Based On Rademacher Complexity For Rough Media GATEM Data

The ground-source airborne time-domain electromagnetic(GATEM)method is composed of a grounded long wire source and a receiver in the air,and the underground structural information is obtained by calculating parameters such as resistivity.The advantage of GATEM method being widely used is that it can quickly and widely detect in complex terrain conditions.This method is suitable for detection under complex terrain conditions.With the application and promotion of the GATEM system,the research in both modeling and inversion are also developing.The actual geological medium has rough characteristics,however,the current modeling and inversion methods for GATEM data are mostly based on homogeneous medium and inversion methods mostly extract only resistivity.Therefore,the accuracy of the inversion of GATEM is insufficient for rough medium,which affectsthe resolution of the actual underground electrical structure.To solve these problems,3-D modeling method for GATEM data in rough medium and inversion method of a highly generalized neural network based on Rademacher complexity are studied.The theoretical research is extended to the application of the GATEM field data,and a relatively complete research system is formed,and a highly generalized neural network method based on Rademacher complexity is applied to the GATEM field data.The main researches are as follows:(1)For the rough medium,the roughness parameter is introduced into Maxwell’s equations,the 3-D electromagnetic response for the grounded long wire source of rough media is solved.The 3-D modeling method for GATEM data of rough medium is realized.The algorithm is verified by the homogeneous half-space,rough half-space and rough layered models.The electromagnetic responses of rough tabular anomalies models and complex rough anomalous models are calculated,and the characteristics of GATEM responses are analyzed.(2)GATEM inversion method of a highly generalized neural network based on Rademacher complexity for rough medium is proposed.It is investigated to solve the dependence of neural network inversion methods on network structure.The GATEM response of rough medium is calculated as sample set for neural network.To realize high-precision inversion of GATEM data,Rademacher complexity is used to limit the generalization error,increase the generalization ability of the neural network.The highly generalized neural network is used to extract the two parameters of resistivity and roughness for rough media.(3)The high generalization neural network is verified by the rough half-space and rough layered models.It is further verified by applying it to the quasi-3D rough abnormal bodies model,which proves that the highly generalized neural network limited by Rademacher complexity can verify the theoretical model.The highly generalized neural network is applied to the GATEM field data,and the interpretation results are consistent with the geological data.This paper’s results effectively increase the accuracy of inversion results from GATEM data,and the inversion method can be extended to electromagnetic data obtained by different working methods,providing new ideas and applications for more accurate identification of underground electrical structures.