A SYSTEMATIC MACHINE LEARNING APPROACH TO IMPROVE FACIES PREDICTION USING MULTISCALE WELL LOG AND IMAGE DATA

Since the advent of digital rock analysis, there has been a growing need for machine learning methods to analyze multi-scale and multi-modal image data and integrate with traditional formation evaluation methods. In this study, we applied feature augmentation machine learning models for facies identification using well log and multi-scale image data from whole cores. Our main goal is to determine geological and petrophysical facies, and fill in the missing information in their estimation using digitally derived parameters. Incorporation of digitally derived data from whole core CT and thin section petrographs improved the accuracy of the model by up to 80% compared to using conventional well data alone. We apply the model defined on a subset to the entire well, which can further be extended to multiple wells. This study thus provides a systematic workflow for facies prediction that can handle large datasets, including multi-scale image data and conventional well logs, to improve reservoir characterization studies.