_ This article, written by JPT Technology Editor Chris Carpenter, contains highlights of paper SPE 227866, “Real-Time CO2-Plume Monitoring and Visualization Considering Geologic Uncertainty at the Illinois Basin-Decatur Carbon-Sequestration Project, ” by Takuto Sakai, SPE, Masahiro Nagao, SPE, and Akhil Datta-Gupta, SPE, Texas A&M University. The paper has not been peer-reviewed. _ Monitoring CO2-plume evolution is essential for ensuring geologic storage security and integrity. Traditional numerical simulation-based data-assimilation workflows are computationally expensive, an aspect further complicated by the fact that geologic uncertainty must be incorporated for robust performance prediction. Therefore, reservoir simulation and model calibration accounting for geologic uncertainty are not amenable to real-time monitoring of CO2-plume evolution for large-scale applications. The authors propose a deep-learning (DL) -based approach that enables near-real-time CO2-plume visualization and rapid data assimilation that incorporates multiple geological realizations for predicting future plume evolution and area-of-review (AOR) determination. Methodology The proposed DL model takes available observed data from CO2 sequestration projects as input, including bottomhole pressure at the injection well, distributed pressure measurements at monitoring wells, and CO2-saturation-log data at monitoring wells. Based on this input, the model predicts a CO2-onset time map that serves as a real-time visualization of the CO2 plume. The CO2-onset time is defined as the calendar time when the gas saturation exceeds a pre-specified threshold at a given location. When using a small threshold value, the onset time map represents the CO2 saturation front effectively at each location. The proposed machine-learning (ML) model predicts only a single image of the onset time map, reducing the dimensionality of the training data significantly. The CO2-onset time map preserves the critical information about plume propagation while enabling a more-compact and computationally feasible representation suitable for real-time field-scale visualization. The entire process of the proposed ML workflow can be divided into three main components: selection of multiple geological realizations, training-data generation, and ML-model training and real-time CO2-plume visualization. Each of these steps is described in detail in the complete paper. Field Application Project Overview and Model Description. The Illinois Basin-Decatur Project (IBDP) is overseen by the Midwest Geological Sequestration Consortium (MGSC). The MGSC aimed to inject one million metric tons of CO2 into a deep saline aquifer. The project site includes one injection well (CCS1) and one monitoring well (VW1). CO2 injection began in November 2011 and continued for 3 years at a rate of approximately 1, 000 metric tons per day. The static reservoir model was developed using data from seismic surveys, geophysical logs, and core analyses. The static model consists of 11 stratigraphic zones, with the primary injection zone located in the Mt. Simon A lower zone. For this study, observed data included behind-casing pressure measurements at six depths (WB1 to WB6) along the monitoring well, bottomhole pressure at the injector (CCS1), and CO2-saturation-log data measured along the monitoring well (VW1).
Chris Carpenter (Wed,) studied this question.