Author Identifier

Seyedmohammadmehdi Nassabeh: http://orcid.org/0009-0003-6233-2908

Date of Award

2025

Document Type

Thesis

Publisher

Edith Cowan University

Degree Name

Doctor of Philosophy

School

School of Engineering

First Supervisor

Stefan Iglauer

Second Supervisor

Alireza Keshavarz

Third Supervisor

Zhenjiang You

Abstract

The transition to a low-carbon future necessitates innovative approaches to carbon management and hydrogen storage, particularly in the context of enhanced oil recovery (EOR) from hydrocarbon reservoirs. This study employs advanced analytics and machine learning techniques to optimize carbon management strategies. One key focus of this research is to evaluate the effectiveness of flue gas and CO2 in Water Alternating Gas (WAG) injection within a homogeneous fractured carbonate reservoir characterized by low porosity and permeability. A computational model was developed to depict the flow regime in the reservoir and simulate reservoir fluid behavior using Eclipse (E300) software, various hybrid EOR methods were evaluated, revealing that natural production accounted for only 29% of the total output. The optimized Hybrid EOR method achieved an impressive oil recovery factor of approximately 85%, demonstrating the critical need for EOR techniques to enhance overall production. In parallel, to mitigate greenhouse gas emissions caused by reliance on hydrocarbon resources, the integration simulation study of CO2 storage with EOR in fractured carbonate reservoirs is implemented to meet this pressing requirement. Utilizing the Eclipse simulator, various gas injection scenarios were modeled to assess the effectiveness of CO2 and flue gas geo-sequestration and EOR. Key findings revealed that flue gas demonstrated superior storage capacity (150 MMSCF) compared to CO2 (85 MMSCF) and maintained better reservoir pressure, while CO2 injection resulted in a higher oil recovery factor of 52% versus 36% for flue gas. Sensitivity analyses indicated that increased reservoir porosity, permeability, and injection rates enhanced gas storage capacity, although CO2 showed a normal distribution trend in permeability. In addition, further reservoir simulation study explores the synergistic relationship between flue gas compositions, reservoir characteristics, and injection rates, highlighting that flue gases with higher concentrations of CO2 and O2 significantly improve recovery factors. Key findings indicate that reservoir temperature, porosity, and permeability are vital factors influencing oil recovery, with CO2 injection consistently yielding the highest recovery rates. Notably, the study established that flue gas injection demonstrated greater sensitivity to increased injection rates, with specific flue gas compositions enhancing recovery efficiency.

On another facet of advanced analytical techniques, a data-driven framework for site screening of offshore CO2 storage is introduced, which integrates diverse geospatial data with expert-weighted criteria to identify optimal locations for Carbon Capture, Utilization, and Storage (CCUS) projects. Machine learning algorithms, particularly Deep Neural Networks (DNN), were employed to enhance predictive accuracy in site selection, achieving an Average Absolute Percentage Difference (AAPD) of 1.486% and a Variance Accounted For (VAF) of 0.9937, thus bridging the gap between scientific inquiry and practical application. This approach not only enhances the precision of site selection but also exemplifies the transformative potential of machine learning in advancing carbon management strategies. The Deep Neural Network (DNN) algorithm proved to be the most effective tool for predicting site suitability, achieving accuracy rates exceeding 90% across various performance metrics.

In response to the escalating global energy demands and the transition to a low-carbon future, this study presents an advanced analytics framework aimed at hydrogen storage potential through machine learning methodologies. Hydrogen is emerging as a vital clean energy vector, and efficient underground storage in geological formations is essential for maintaining energy security. However, challenges remain in understanding the interactions between gas and rock. To tackle these challenges, a comprehensive dataset of 1,045 entries and over 5,200 data points was utilized to develop predictive models for contact angles in water-hydrogen-rock systems. Various machine learning algorithms including Support Vector Machine (SVM), k-Nearest Neighbors (KNN), Feedforward Deep Neural Network (FNN), and Recurrent Deep Neural Network (RNN) were evaluated, with the FNN achieving the highest predictive accuracy. This capability is crucial for assessing hydrogen flow through porous media during underground storage.

Overall, this thesis highlights the potential of advanced analytics and machine learning in refining carbon management practices and optimizing hydrogen storage capabilities. It provides essential insights for the energy sector's sustainable transition, demonstrating that integrating CO2 storage with EOR is a viable strategy for reducing greenhouse gas emissions while enhancing oil recovery metrics. The findings underscore the importance of advanced analytics in addressing key challenges in the energy sector and promoting sustainable practices for a low-carbon future, paving the way for future research in these critical areas.

Comments

Author also known as Mehdi Nassabeh

DOI

10.25958/xpn3-6336

Share

 
COinS