Volume change of CO2 + long-chain liquid n-alkane (with n ≥ 14) mixtures under geological storage conditions: An observed abrupt change near the critical point of CO2

• Published in: Journal of molecular liquids Vol. 413; p. 125827
• Main Authors: Bei, Ke, Wang, Youyan, Yang, Dan, Shao, Qizhi, Zheng, Xiangyong, Zhao, Min, Chou, I-Ming, Kikuchi, Jun, Pan, Zhiyan
• Format: Journal Article
• Language: English
• Published: Elsevier B.V 01.11.2024
• Subjects: CO2 + long-chain alkane; High pressure optical cell; Machine learning; Raman spectroscopy; Supercritical fluid; Volume expansion; High pressure optical cell; Supercritical fluid; Volume expansion; CO2 + long-chain alkane; Raman spectroscopy; Machine learning
• ISSN: 0167-7322
• DOI: 10.1016/j.molliq.2024.125827

[Display omitted] •Volume change of CO2 + long-chain liquid n-alkane (with n ≥ 14) mixtures were studied.•Abrupt changes in volume expansion were observed near the CO2 critical point.•The abrupt due to the critical anomaly of the thermodynamic behavior of the mixture.•Machine learning (ML) were used to mining the intrinsic trend of the volume expansion.•An ML predictive framework could provide rapid predictions of the volume change. This study used a high pressure optical cell (HPOC) combined with a heating/cooling stage, a pressure device, and a laser Raman spectrometer to investigate the volume expansion (VE) of CO2 + tetradecane and CO2 + hexadecane mixtures under geological storage conditions. Abrupt changes in VE were observed near the CO2 critical point (CP), which is assumed to be ubiquitous in CO2 + long-chain alkane mixtures with ≥14 carbon alkanes. This is due to the critical anomaly of the thermodynamic behavior of the mixture near the CP of CO2, which inhibits the VE of CO2 + long-chain alkanes at lower temperatures. In addition, four machine learning (ML) algorithms, including backpropagation neural network (BPNN), K-nearest neighbors (KNN), random forest (RF), and support vector regression (SVR), were applied to predict the volume expansion factor (VEF). The BPNN model provided much higher robustness and accuracy with correlation coefficient (R2) >0.98 and mean square error (MSE) <0.03, which indicates a strong agreement between the predicted result and matching experimental data. We concluded that a predictive framework built using ML could provide rapid predictions of the volume change curve or curved surface of CO2 + organic mixtures under various temperature and pressure conditions.