Machine Learning-Based CO2 Saturation Tracking in Saline Aquifers Using Bottomhole Pressure for Carbon Capture and Storage CCS Projects

Abstract

Abstract In the face of escalating global climate concerns, the imperative to mitigate CO2 emissions has never been more pressing. A pivotal question that arises pertains to the responsible disposal of captured CO2. Deep saline aquifers have emerged as a promising solution, owing to their inherent attributes of high permeability and porosity, enabling efficient CO2 injection and long-term storage. Nevertheless, the successful implementation of CO2 reservoir injection presents multifaceted challenges, notably the need for an impermeable cap rock to prevent leakage while preserving reservoir permeability for injection ease. This study delves into the realm of data-driven decision-making, where the oil and gas industry is progressively harnessing the capabilities of Machine Learning (ML) and Deep Learning (DL) technologies. Specifically, we investigate the application of ML and DL techniques in monitoring CO2 saturation levels within saline aquifers, employing bottomhole pressure as the primary predictive parameter. A range of algorithms, including Random Forest (RF), Support Vector Regressor (SVR), Recurrent Neural Networks (RNN), and Long Short-Term Memory (LSTM), were rigorously tested to ascertain their efficacy in this endeavor. The training data for these models were meticulously generated using a well-known reservoir simulator. Our comprehensive investigation culminated in insightful findings. We present a detailed analysis of how emerging technologies, such as ML and DL, can be harnessed to accurately track CO2 saturation levels. The performance evaluation of the employed algorithms provides valuable insights into their proficiency for predicting CO2 saturation. These results offer a nuanced understanding of the potential applications of these technologies in the management of CO2 reservoirs, paving the way for more effective and sustainable carbon capture and storage solutions. This research underscores the integration of cutting-edge machine learning and deep learning technologies within the oil and gas sector to tackle the intricate challenges associated with CO2 disposal. Furthermore, it highlights the pivotal role of data-centric decision-making in the context of CO2 injection and storage, contributing significantly to the ongoing discourse on sustainable carbon capture and storage (CCS) solutions. In a world grappling with the urgent climate crisis, our study's novelty lies in its potential to drive forward more efficient and environmentally responsible CO2 management strategies.