Author ORCID Identifier

https://orcid.org/0009-0009-3361-941X

Semester

Fall

Date of Graduation

2023

Document Type

Thesis

Degree Type

MS

College

Statler College of Engineering and Mineral Resources

Department

Petroleum and Natural Gas Engineering

Committee Chair

Shahab Mohaghegh

Committee Member

Samuel Ameri

Committee Member

Kashy Aminian

Abstract

The application of artificial intelligence (AI) and machine learning (ML) in engineering holds great promise for addressing the challenge of climate change. One key focus in the fight against climate change is the geological storage of CO2, which has gained significant attention as a crucial strategy for mitigating greenhouse gas emissions. The primary objective is to ensure the safe and controlled containment of injected CO2 over an extended period, which has proven to be a major challenge in the journey toward carbon capture, and geologic storage (CCS). To achieve this goal, several critical tasks must be accomplished. These include ensuring the quality control of potential underground storage sites (candidate wells), monitoring the conditions of the CO2 saturation plume, and simulating the behavior of reservoir pressure distribution over time. Traditionally, the characterization of fluid flow in the oil and gas industry has heavily relied on numerical simulators, and this conventional approach remains the primary tool in CCUS for executing the tasks mentioned above. However, these numerical reservoir models are typically extensive, with tens of millions of grid blocks, and their potential remains largely untapped due to their high computational demands and time-consuming nature. Consequently, there is a pressing need for an efficient alternative tool that can facilitate swift and reliable decision-making processes. In this study an AI-based proxy model has been developed to replicate the pressure distribution of the injected CO2 as captured by the numerical simulation model (Eclipse) used to monitor the IBDP storage site. Employing Artificial Neural Networks (ANNs) and data-driven techniques, this study develops Smart Proxy models to reduce the high computational cost of reservoir simulation modeling. AI-based Smart proxy models are data-driven machine learning models that can accurately replicate the output of complex numerical reservoir simulation models at every layer and every time step in a fraction of the time (minutes). To develop a Smart Proxy Model, ANN algorithms are trained on large volumes of subsurface data to learn the complex patterns of fluid flow in a reservoir. 100 reservoir simulation realizations with varying geological properties, such as porosity, permeability, baffles and faults was provided for this project, with 2 injection wells; one active and the other inactive. The simulation results, including the geological reservoir properties of 15 of these realizations was selected, with developed features from the model was captured and used to train the Smart Proxy model. 10 realizations were left as blind validation dataset to perform final evaluation of the developed model. The results show that the developed Smart Proxy Model can successfully mimic the pressure distribution of the Eclipse outputs at every grid layer, and every time step of the model simulation.

Download

Share

COinS