Summary
A new genetic algorithm (GA)-based correlation has been developed to estimate the change in MMP when CO2 is diluted with other gases, termed "impure CO2" in the context of this paper. The advantage of this correlation over others is that it can be used for gas mixtures with higher N2 concentrations (tested up to 20 mol%) and with non-CO2 component concentrations up to 78 mol% (e.g., H2S, N2, SOx, O2, and C1-C4) with a higher accuracy. Equally important, it could be a useful screening tool when experimental data are not available and when developing an optimal and economical laboratory program to estimate the MMP.
In developing this correlation, the GA software developed in our earlier work (Emera and Sarma 2005a) has been modified to account for various components in the injected-gas stream. The correlation estimates the change in MMP as a function of injected-gas solvency in the oil. The solvency, in turn, is related to critical properties of the injected gas (critical temperature and pressure). In addition, pure CO2/oil MMP is used as an input in this correlation. The correlation has been validated successfully against published experimental data and several correlations in the literature. It yielded a better match with an average error of 4.7% and a standard deviation of 6.3%, followed by the Sebastian et al. (1985) correlation with a 13.1% average error and a 22.0% standard deviation and the Alston et al.(1985) correlation with a 14.1% average error and a 43.2% standard deviation.
Introduction
CO2 miscible flooding is among the most widely applied nonthermal enhanced-oil-recovery (EOR) techniques. Among gas-injection processes, CO2 is preferred to hydrocarbon gases because of its lower cost and high displacement efficiency. Furthermore, the increasing global awareness of the detrimental effects on the environment of industrial gases containing high CO2 concentrations has also contributed to an added impetus to harness these gases and sequester them into petroleum reservoirs while also enhancing oil recovery.
An a priori understanding of the effect of various impurities on the CO2/oil MMP is critical to the design and implementation of a CO2 gas-injection project. Key factors that affect CO2 flooding are reservoir temperature, oil characteristics, reservoir pressure, and the purity of injected CO2 itself. Field case histories from CO2 floods in the Permian Basin, west Texas, suggest that CO2 purity should not be viewed as too rigid a constraint because the use of a low-purity CO2 stream could also be economic and effective in enhancing oil recovery. In fact, certain impurities, such as H2S and SOx, could contribute toward attaining CO2/oil miscibility at lower pressures. The presence of C1 and N2, however, could increase the MMP. From an operational perspective, it is often the remaining low percentages of non-CO2 gases that are more difficult and costly to remove, requiring expensive gas-separation facilities. Safety and compression cost considerations also justify near-miscible CO2 flood applications for some reservoirs. Therefore, the potential of injecting impure gases containing both CO2 and non-CO2 components (H2S, N2, SOx, O2, and C1-C4) could be an attractive option, provided the impure gas composition does not affect the process performance adversely and its overall impact on miscibility with the oil, separation/purification at the surface, and subsequent reinjection is evaluated and well understood a priori.
This paper presents a reliable GA-based correlation to estimate the change in MMP when CO2 is diluted with other gases, together with a comprehensive comparison of its efficiency against other commonly used correlations (listed in Table 1). The software designed in our earlier work (Emera and Sarma 2005a) to develop an MMP correlation for pure CO2 and oil has been modified to account for impure CO2 gases with non-CO2 components.
The GA software used in this study has been presented in the flow chart provided in Fig. 1. This figure also presents the stopping criterion under which the fitness of the solution is decided and accepted. The GA software uses real numbers coded as chromosomes (problem solutions comparable to chromosomes of the biological system) to encode the correlation in an initial random population (group of solutions) of 100 chromosomes size. Such an encoding technique enhances the GA robustness. Each chromosome is evaluated on the basis of a fitness value, which is designed on the basis of the objective function (minimizing the misfit between observed and predicted values). For the selection technique, the roulette wheel method was used. Also, to produce a new offspring (new solutions), reproduction operators such as one-point crossover and mutation were used. Moreover, the correlation errors could be minimized further through a series of iterative optimization runs using the previous software results as a new initial population.