DNN Enabled Real-Time Modeling of EM LWD Tool Responses in Complex Subsurface Formations

In this paper, a fast forward modeling framework for predicting the logging-while-drilling (LWD) tool responses is presented. Unlike the traditional numerical methods based on the finite difference or finite element method, the proposed approach makes use of a deep neural network (DNN). During the offline training stage, the DNN is trained to extract the mapping from the formation parameters to tool responses. Then it can be used to predict the tool responses immediately given the formation parameters. The target application is the directional LWD operations, where the formation resistivity model has to be updated on the fly. This requires an extremely efficient forward modeling of LWD tool responses. Numerical tests are presented and results indicate the proposed method is outperforming the conventional methods.