A subsurface machine learning approach at hydrocarbon production recovery & resource estimates for unconventional reservoir systems: Making subsurface predictions from multimensional data analysis

Abstract

An innovative, practical, and successful subsurface machine learning workflow was introduced that utilizes any structured reservoir, geologic, engineering and production data. This workflow is colloquially called the Artificial Learning Integrated Characterization Environment (ALICE), and it has changed the way Chevron manages its tight rock and unconventional assets. The workflow guides users from framing and data gathering to geospatial assembly, quality control and ingestion, then on through machine-learning feature selection, modeling, validation, and acceptance for results reporting. The ultimate products of the workflow can be visualized in both map or log (depth) space to help identify key areas for well optimization or landing zones, respectfully. The results from ALICE have been used within Chevron to aid in exploration review assessments, type curve adjustments, landing strategies, well performance lookbacks and more. A real-data example of the workflow is presented from start to finished product for the Midland Basin Wolfcamp A, a maturely developed unconventional reservoir. The ALICE workflow and products were developed through close cross-functional collaboration between business units, data science, and research components of the corporation.