Development of Coordinated Methodologies for Modeling CO2-Containing Systems in Petroleum Industry.
Ghiasi, Mohammad Mahdi.
MetadataShow full item record
Clathrate hydrates formation in natural gas processing facilities or transportation pipelines may lead to process and/or safety hazards. On the other hand, a number of applications are suggested on the basis of promoting the gas hydrate formation. Some researchers have investigated separation and purification processes through gas hydrate crystallization technology. Some works report that the hydrate formation is applicable to the gas transportation and storage. Gas hydrate concept is also studied as a potential method for CO2 capture and/or sequestration. Water desalination/sweetening, and refrigeration and air conditioning systems are other proposed uses of hydrates phenomenon. In the realm of food processing and engineering, several studies have been done investigating the application of gas hydrate technology as an alternative to the conventional processes. Accurate knowledge of phase equilibria of clathrate hydrates is crucial for preventing or utilizing the hydrates. It is believed that energy production or extraction from different fossil fuels is responsible for considerable emissions of CO2, as an important greenhouse gas, into the atmosphere. Furthermore, CO2 removal from the streams of natural gas is important for enhancing the gaseous streams’ heating value. Employment of solvent-based processes and technologies for removing the CO2 is a widely employed approach in practical applications. Amine-based or pure amine solutions are the most common choice to remove the produced CO2 in numerous carbon capture systems. Further to the above, ionic liquids (ILs) are capable to be utilized to capture CO2 from industrial streams. Other potential solvent are sodium piperazine (PZ) and glycinate (SG) solutions. Equilibrium absorption of carbon dioxide in the aqueous phase is a key parameter in any solvent-based CO2 capture process designing. The captured CO2, then, can be injected into the hydrocarbon reservoirs. In addition to the fact that injection of CO2 into potential sources is one of the most reliable methodologies for enhanced hydrocarbon recovery, utilizing this process in conjunction with the CO2 capture systems mitigates the greenhouse effects of CO2. One of the most significant variables determining the success of CO2 injection is known to be the minimum miscibility pressure (MMP) of CO2-reservoir oil. This research study concerns implementation of computer-based methodologies called artificial neural networks (ANNs), classification and regression trees (CARTs)/AdaBoost-CART, adaptive neuro-fuzzy inference systems (ANFISs) and least squares support vector machines (LSSVMs) for modeling: (a) phase equilibria of clathrate hydrates in: 1- pure water, 2- aqueous solutions of salts and/or alcohols, and 3- ILs, (b) phase equilibria (equilibrium) of hydrates of methane in ILs; (c) equilibrium absorption of CO2 in amine-based solutions, ILs, PZ solutions, and SG solutions; and (d) MMP of CO2-reservoir oil. To this end, related experimental data have been gathered from the literature. Performing error analysis, the performance of the developed models in representing/ estimating the independent parameter has been assessed. For the studied hydrate systems, the developed ANFIS, LSSVM, ANN and AdaBoost-CART models show the average absolute relative deviation percent (AARD%) of 0.04-1.09, 0.09-1.01, 0.05-0.81, and 0.03-0.07, respectively. In the case of hydrate+ILs, error analysis of the ANFIS, ANN, LSSVM, and CART models showed 0.31, 0.15, 0.08, and 0.10 AARD% of the results from the corresponding experimental values. Employing the collected experimental data for carbon dioxide (CO2) absorption in amine-based solutions, the presented models based on ANFIS, ANN, LSSVM, and AdaBoost-CART methods regenerated the targets with AARD%s between 2.06 and 3.69, 3.92 and 8.73, 4.95 and 6.52, and 0.51 and 2.76, respectively. For the investigated CO2+IL systems, the best results were obtained using CART method as the AARD% found to be 0.04. Amongst other developed models, i.e. ANN, ANFIS, and LSSVM, the LSSVM model provided better results (AARD%=17.17). The proposed AdaBoost-CART tool for the CO2+water+PZ system reproduced the targets with an AARD% of 0.93. On the other hand, LSSVM, ANN, and ANFIS models showed AARD% values equal to 16.23, 18.69, and 15.99, respectively. Considering the CO2+water+SG system, the proposed AdaBoost-CART tool correlated the targets with a low AARD% of 0.89. The developed ANN, ANFIS, and LSSVM showed AARD% of more than 13. For CO2-oil MMP, the proposed AdaBoost-CART model (AARD%=0.39) gives better estimations than the developed ANFIS (AARD%=1.63). These findings revealed the reliability and accuracy of the CART/AdaBoost-CART methodology over other intelligent modeling tools including ANN, ANFIS, and LSSVM.