A Fast Inversion of Induction Logging Data in Anisotropic Formation Based on Deep Learning

With the development of large gradient wells and horizontal wells, resistivity anisotropy has become an important concern. Most of the inversion problems encountered in geophysical applications are time-consuming. In this letter, we propose a novel approach based on deep learning (DL) to solve the time-consuming problem of the inversion of induction logging data in layered anisotropic formation. A deep neural network (DNN) architecture combined with an adaptive moment estimation (Adam) optimization algorithm is utilized. Batch normalization is used to accelerate the training convergence. Data sets are obtained from forward modeling in different anisotropic formations, by which the DNN is trained, validated, and tested. A series of experiments are carried out to test the accuracy, robustness, and speed of the inversion based on DL in the anisotropic formation. The experimental results indicate that the approach based on DL can achieve the inversion with high accuracy. Furthermore, the trained DNN model for the inversion is two orders of magnitude faster than the most commonly used nonlinear iterative algorithm that achieves the same inversion precision. The trained DNN model considerably reduces the inversion computation time without compromising accuracy. Moreover, the approach overcomes the defect of the dependence of the nonlinear iterative algorithm on the initial values. Our study has shown the promising potential of the approach based on DL for inverting the logging data in anisotropic formations that are difficult to quickly invert by the conventional method.