Modeling extra-deep electromagnetic logs using a deep neural network

Modern geosteering is heavily dependent on real-time interpretation of deep electromagnetic (EM) measurements. We present a methodology to construct neural network (DNN) model trained reproduce full set extra-deep EM logs consisting 22 measurements per logging position. The in 1D layered environment up seven layers with different resistivity values. A commercial simulator provided by tool vendor used generate training dataset. dataset size limited because the optimized for sequential execution. Therefore, we design that embraces geological rules and specifics supported forward model. use this produce an based DNN without access proprietary information about configuration or original source code. Despite employing relatively small size, resulting quite accurate considered examples: multi-layer synthetic case section published historical operation from Goliat Field. observed average evaluation time 0.15 ms position makes it also suitable future as part evaluation-hungry statistical and/or Monte-Carlo inversion algorithms within workflows.