Understanding the Controlling Factors for CO2 Sequestration in Depleted Shale Reservoirs Using Data Analytics and Machine Learning

Abstract

Abstract Carbon capture and sequestration (CCS) will generate an industry comparable to, if not greater than, the existing oil and gas sector. Carbon capture and sequestration is the capture of carbon-dioxide from refineries, industrial facilities, and major point sources such as power plants and storing it in subsurface formations. Subsurface formations like unconventional reservoirs can be a good example for storing carbon-dioxide. Despite its importance in the oil and gas industry, our understanding of carbon-dioxide sequestration in unconventional reservoirs still needs to be developed. The objective of this paper is to identify the most important parameters that affect carbon-dioxide sequestration in depleted shale reservoirs using data-analytics and machine-learning. The dataset used was an extensive shale reservoir dataset which comprised a large set of numerical simulation scenarios. A quality check of the input data was performed for missing variables. Then, a data-analytics based investigation was followed to develop insights into the relationship between reservoir parameters and operational parameters, which were the main predictor variables, as well as between the predictor variables and the main response variable: cumulative CO2 injected. Machine-learning based predictive models such as multiple linear regression, regression tree, bagging, random forest, and boosting were built to predict the cumulative CO2 injected. Variable importance (screening) was carried out to determine the crucial parameters which drive CO2 sequestration performance in shale reservoirs. The results revealed that there is a relationship between the reservoir and operational parameters, together with the predictor variables and response variable. Operational parameters displayed a monotonic relationship with the cumulative CO2 injected. Random forest provided the best predictive ability among the machine-learning techniques consistent with the theoretical background of random forest. Regression tree had the worst predictive ability mainly because of overfitting. Screening results show that stimulated reservoir volume fracture permeability was the most important variable for the performance of CO2 sequestration. The findings and results reported in this study will allow the exploration and production companies to determine what causes low-performance or high-performance CO2 sequestration process in shale reservoirs.