Chen, ZhangxingMcCoy, Sean T.Ma, Haoming2023-10-302023-10-302023-10-25Ma, H. (2023). Data-driven carbon dioxide enhanced oil recovery models and their applications (Doctoral thesis, University of Calgary, Calgary, Canada). Retrieved from https://prism.ucalgary.ca.https://hdl.handle.net/1880/117445https://doi.org/10.11575/PRISM/42288In the context of the 2050 net zero carbon emission goal to mitigate climate change, the demand for CO2 geological sequestration is substantially increasing. In the oil and gas industry, injecting CO2 has been proven as a mature technology to improve the oil recovery associated with permanent geologic carbon storage to meet the dual demands of energy consumption and emission reduction. Additionally, the revolutionary development of artificial intelligence provides a time-effective and accurate approach to forecast reservoir performance. The paradox of maximizing profitability while minimizing greenhouse gas (GHG) emissions has been heavily debated in recent years. This thesis attempts to address three challenges in the field of CO2-enhanced oil recovery (CO2-EOR) associated with carbon storage. In the first part, a generalized reduced-form model is developed based on numerical simulation and statistical nonlinear regressions. Applied to the Weyburn oil field production data, the proposed model demonstrates its capability to forecast the reservoir performance within 5% tolerance in terms of an incremental oil recovery factor, net CO2 retention, and net CO2 utilization. Three major contributions are made to the existing literature: (1) the predictive capabilities have been proven by combining the early-stage field production data and the proposed generalized reduced-form model; (2) the statistical correlations are illustrated with respect to CO2 injection rather than the total injected fluid compared to the previous investigation; (3) the model is developed based on numerical results with the field data validation, which fills the research gap that statistical models cannot be validated owing to the lack of empirical data. In the second part, three machine learning algorithms are studied to further improve the data-driven models by (1) establishing a large dataset based on reservoir simulations; (2) capturing the field variabilities and reducing the real-time production data as inputs, which are two major limitations of applying reduced-form models to predict the reservoir performance. Results from this part demonstrate that the computational time can be reduced by 700 to 5000 times with an optimal artificial neural network (ANN) model compared to the reservoir simulation approach. Although machine learning has been deployed to the oil and gas industry in recent years, two major challenges remain: (1) most studies are developing site-specific data-driven models based on simulation, which makes it difficult to apply them to other fields; (2) owing to confidentiality reasons, little production data is accessible. This part differs from other studies as it resolved the limitations of statistical models and introduced a framework to develop a generalized data-driven model with the field production data validation. As such, it can be applied to a variety of oil fields to quantify the CO2-EOR potential. Furthermore, the proposed data-driven model has been applied to understand the economic and environmental concerns of CO2-EOR from the entire field scale. In the third part, the potential economic and environmental outcomes are analyzed by combining the predicted reservoir performance with the data-driven models. It is found that when utilizing CO2 from industrial sources, varying the WAG ratio can result in a minimum levelized cost and net emission because increasing and decreasing the WAG ratio can both result in an increasing of a levelized cost of oil production (LCOP) and net emission E_net, but through different mechanisms. When the WAG ratio increases, LCOP increases because of water usage and CO2 recycling and E_net increases owing to the GtG (gate-to-gate) emission. When the WAG ratio decreases, LCOP increases because of CO2 purchase and E_net increases owing to the upstream emission. The trade-off can be estimated for the scenario of CO2 from DAC (direct air capture), which is -1.40 kgCO2eq/$ for the Weyburn oil field because the WAG ratio drives LCOP and E_net in opposite directions. In addition, the weighted factors are used to determine the driven mechanisms as well as the prospective economic and environmental outcomes resulting from the varying market conditions. As a result, CO2-EOR initiatives can be divided into four categories based on the operators' choice of a CO2 source, as current climate policies are insufficient to deploy DAC. Negative emission potential in CO2-EOR has been demonstrated to require a high net utilization factor, CO2 from DAC, and sufficient carbon revenues for operators. Using captured CO2 from the air may result in negative emissions, but insufficient financial incentives exist to encourage this practice.enUniversity of Calgary graduate students retain copyright ownership and moral rights for their thesis. You may use this material in any way that is permitted by the Copyright Act or through licensing that has been assigned to the document. For uses that are not allowable under copyright legislation or licensing, you are required to seek permission.EnergyData-Driven Carbon Dioxide Enhanced Oil Recovery Models and their Applicationsdoctoral thesis