Modelling of CO2 and green-house gases (GHG) miscibility and interactions with oil to enhance the oil recovery in gas flooding processes

Abstract

1. Objective: The objective of this research has been to develop more reliable models to predict the miscibility and interactions between CO₂ or green-house gas (GHG) and oil (dead and live oils) over a wider range of conditions, based on data from different site sources, considering all the major variables affecting each modelled parameter, and for different injected gas compositions. The Genetic algorithm (GA), an artificial intelligence technique based on the Darwinian theory of evolution that mimics some of the natural processes in living organisms, was used to develop these models, based on GA software that has been developed in this work (as a modelling technique). While applications of GA have been used recently in the mathematical and computer sciences, its applications in the petroleum engineering, especially EOR research, have been limited. 2. Motivation to Investigate the Potential of GA-based Models: The detrimental effects of CO₂ and/or GHG emissions from various industrial and human/activity sources on the environment are a major concern worldwide. This has resulted in an intensive global R&D effort to lower or mitigate the damaging impact of GHG on the environment. One potentially attractive and effective means of lowering the GHG emissions could be to capture them from their major sources of emissions and then sequester them in depleted oil and gas reservoirs while also enhancing oil recovery. Typically, a GHG stream, also referred to as "flue gas", contains high percentages of CO₂ in addition to other gases, notably, N₂, NOₓ and SOₓ. The presence of high CO₂ content in the flue gas, in particular, could make this option potentially viable, provided the miscibility and interaction properties between the injected gas and reservoir fluids are favorable. Therefore, it is critical to ascertain the likely miscibility and interactions parameters between the injected gas (CO₂ or flue gas) and oil at different conditions to determine the optimal miscibility and interaction conditions that contribute to oil viscosity reduction and oil swelling. They in turn enhance oil recovery through improved gas flooding process performance due to higher oil mobility, volumetric sweep efficiency, and relative permeability to oil. Often miscibility and interactions between injected gases and oils are established through "experimental methods", "new mathematical models" based on phase equilibria data and equations of state (EOS), and available "published models". Experimental methods are time-consuming and costly. Moreover, they can handle only limited conditions. Mathematical models require availability of a considerable amount of reservoir fluid composition data, which may not be available most of the time. Although, the published models are simpler and faster to use, one must however recognise that most of these models were developed and validated based on limited data ranges from site-specific conditions. Therefore, their applications cannot be generic. Another noteworthy point is that most of the interactions models have been developed using dead oil data and pure CO₂ as an injected gas. Hence, they do not perform well for a wider range of live oils, as well as injected flue gases, which contain different components besides CO₂. Consequently, there is a need to have more reliable miscibility and interaction models, which can handle a much wider range of conditions and different data sources. Also, these models should be able to consider all the major variables, different injected gas compositions, and live oil in addition to dead oil. 3. GA-based Models Developed in This Research: -- GA-based model for more reliable prediction of minimum miscibility pressure (MMP) between reservoir oil and CO₂: This model recognised the major variables affecting MMP (reservoir temperature, MWc₅₊ , and volatiles and intermediates compositions). It has been successfully validated with published experimental data and compared to common models in the literature. It is noted that GA-based CO₂-oil MMP offered the best match with the lowest error and standard deviation. -- GA-based flue gas-oil MMP model: For this model, the MMP was regarded as a function of the injected gas solubility into oil, which in turn is related to the injected gas critical properties (pseudocritical temperature and pressure) besides reservoir temperature and oil composition. A critical temperature modification factor was also used in developing this model. The GA-based model has also been successfully validated against published experimental data and compared to several models in the literature. It yielded the best match with the lowest average error and standard deviation. Moreover, unlike other models, it can be used more reliably for gases with higher N₂ (up to 20 mole%) and different non-CO₂ components (e.g., H₂S, N₂, SOₓ, O₂, and C₁-C₄) with higher ratio (up to 78 mole%). -- GA-based CO₂-oil physical properties models: These models have been developed to predict CO₂ solubility, impact on the oil swelling factor, CO₂-oil density, and CO₂-oil viscosity for both dead and live oils. These models recognised the major variables that affect each physical property and also properly address the effects of CO2 liquefaction pressure and oil molecular weight (MW). These models have been successfully validated with published experimental data and have been compared against several widely used models. The GA-based CO₂-oil properties models yielded more accurate predictions with lower errors than other models that have been tested. Furthermore, unlike the other tested models, which are applicable to only limited data ranges and conditions, GA-based models can be applied over a wider data range and conditions. -- GA-based flue gas-oil physical properties models: These models predict flue gas-oil properties such as, flue gas solubility, impact on the oil swelling factor, and flue gas-oil density and viscosity while recognising all the major variables affecting each property. Also, the GA-based models recognised the different injected flue gas compositions. These models have been successfully validated with published experimental data and have also been compared against other commonly reported CO₂-oil models, which are often used for flue gas-oil physical properties prediction. The GA-based models consistently yielded a lower prediction error than the models that have been tested. Furthermore, unlike other models, which are applicable only over limited data ranges and conditions, GA-based models can be valid over a wider range of data under various conditions. All the above-mentioned models, developed in this research, are particularly useful when experimental data are lacking and the project financial situation is a concern. In addition, these models can be useful as a fast track gas flooding project screening guide. Also, they can easily be incorporated into a reservoir simulator for CO₂ or flue gas flooding design and simulation. Furthermore, they can serve as yet another useful tool to design optimal and economical experimental test protocols to etermine the miscibility and interactions between the injected CO₂ or flue gas and oils in gas flooding processes.