Research of Geoelectrical Model Identification for DC Resistivity Method Based on Deep Learning

As an important branch of geophysics, the nonlinear identification method has played a unique role, while deep learning is still in the initial stage of geophysical research. To identify the geoelectric model of the direct current (DC) resistivity method, this paper has established a convolutional neural network model under the TensorFlow framework in deep learning. Ten typical kinds of abnormal body model datasets were established by finite difference numerical calculation. To select the appropriate network parameters, the network is tested with different batch sizes, dropout rates, and optimization algorithms. After training the established convolutional neural network, the accuracy reached 97.1% and the loss degree was reduced to less than 0.087. To prove the accuracy of the identification network, six sets of abnormal body model data are selected to test the trained identification network, and the results show that the identification network can effectively identify the geoelectric model.