PUMLE: A Framework for Data-Driven Applications in Underground Carbon Sequestration Venture

Abstract

PUMLE is a framework designed to seamlessly manage large datasets generated from numerical simulations of CO2 injection into underground reservoirs. Using an architecture inspired by the Medallion model, the workflow covers the key stages of the data lifecycle, from generation through batch simulations to the final storage of results in efficient file formats, prioritizing minimal computational effort. PUMLE is intended to generate high quality data to support machine learning experiments aimed at predicting the dynamics of CO2 plumes in future Brazilian carbon sequestration projects, particularly those based on techniques aware of the physical phenomenology behind the injection process. Performance metrics indicate that while the numerical solution accounts for the bulk of processing time, PUMLE’s subsequent data ingestion, transformation, and storage stages demonstrated high efficiency, collectively consuming only about 1.4% of the total wall-clock time in reference simulation tests. This tool is expected to facilitate the creation of consistent datasets, providing scalability and reproducibility for training, testing, and prototyping data-driven solutions focused on the development and monitoring stages of carbon sequestration ventures in heterogeneous domains.