Data-Driven Approach for Resistivity Prediction Using Artificial Intelligence

Abstract

Formation resistivity is crucial for petrophysics and formation evaluation. Laboratory measurements and/or well logging can be used to provide resistivity data. Laboratory measurements are time-consuming and costly, limiting their use. Furthermore, certain log records may be missing in some segments for a variety of reasons, including instrument failure, poor hole conditions, and data loss due to storage and incomplete recording. The purpose of this study is to apply support vector machines (SVM), and functional networks (FN) to introduce intelligent models for formation resistivity prediction using other available logging parameters. The well logs include gamma ray, density, neutron, and sonic data. The predictive models were built using a data collection of roughly 4300 data points collected from vertical sections of complex reservoirs. For model training and testing, the data set was split at random in a 70:30 ratio. The predictive models were validated using a different set of data (around 1300 points) that had not been seen by the model. The models predicted the target with a good correlation coefficient (R) of around 0.93 and accepted root-mean-squared error (RMSE) of 0.3 for training and testing. The suggested methods for estimating formation resistivity from available logging parameters are shown to be reliable in this study. Resistivity prediction can fill the missing gaps in log tracks and may save money by removing resistivity logs running in all offset wells in the same field.