Use of Artificial Neural Networks to Invert 3D Electrical Resistivity Imaging Data

Abstract

The artificial neural network approach has presented to solve 3D resistivity inverse problems included in complex subsurface structure. The 3D synthetic data sets were generated to train the neural network using a finite element forward modeling code. Several artificial neural network algorithms have been tested on a basis of trial and error for training data set. The resilient back propagation paradigm was efficient in training stage. After 1340 epochs, the MSE error as a function of epochs for the combined training data reduced a threshold value (0.00034) and the network was converged. To test the trained network with real field data, the 3D electrical resistivity data was conducted on certain structure with known resistivity and was acquired with the Pole‐Pole configuration. The interpreted results showed, the artificial neural network approach was capable to invert 3D electrical resistivity data as well as synthetic data. The main advantage of this approach for resistivity inversion is that, once the neural network has been trained it can invert any 3D electrical resistivity imaging data rapidly.