NEURAL NETWORK-BASED APPROACH TO RESISTIVITY LOGS EXPRESS SIMULATION IN REALISTIC MODELS OF COMPLEX TERRIGENOUS SEDIMENTS

Abstract

The article proposes a new algorithmic approach to resistivity logs simulation based on convolutional neural networks wich allows constructing algorithms for solving forward problems for specific logging tools in detailed models of near-wellbore space with thin layers, accounting for radial resistivity changes, borehole wall irregularities and drilling fluid displacement by the logging tool. Experimental algorithms for expressmodeling for three common Russian galvanic and induсtion logging methods in two-dimensional models of the near-wellbore space have been implemented based on the proposed approach. Logs simulation using the developed neural network algorithms is multi pletimes faster than using numerical solvers. The proposed solutions open up possibilities to use more sophisticated basic geoelectric models of the near-wellbore space. The use of models adequate in complexity to the actual target geological objects will increase the reliability of interpretation results of resistivity logs measured in complex geological conditions.