Evaluation of Source Rock Potentiality and Prediction of Total Organic Carbon Using Well Log Data and Integrated Methods of Multivariate Analysis, Machine Learning, and Geochemical Analysis

• Published in: Natural resources research (New York, N.Y.) Vol. 31; no. 1; pp. 619 - 641
• Main Authors: Nyakilla, Edwin E., Silingi, Selemani N., Shen, Chuanbo, Jun, Gu, Mulashani, Alvin K., Chibura, Patrick E.
• Format: Journal Article
• Language: English
• Published: New York Springer US 01.02.2022; Springer Nature B.V
• Subjects: Chemistry and Earth Sciences; Computer Science; Core analysis; Earth and Environmental Science; Earth Sciences; Economic importance; Fossil Fuels (incl. Carbon Capture); Gaussian process; Geochemistry; Geography; Hydrocarbons; Kerogen; Learning algorithms; Machine learning; Mathematical Modeling and Industrial Mathematics; Mineral Resources; Multivariate analysis; Oil exploration; Organic carbon; Original Paper; Physics; Predictions; Pyrolysis; Rocks; Statistical analysis; Statistics; Statistics for Engineering; Support vector machines; Sustainable Development; Total organic carbon; Triassic; Cluster analysis; Factor analysis; Machine learning; Geochemical analysis; ); Pearson's correlation coefficient; Source rock
• DOI: 10.1007/s11053-021-09988-1
• ISSN: 1520-7439

In this study, integrated approaches based on multivariate analysis (MVA), machine learning (ML), and geochemical analysis are proposed to investigate the potential of hydrocarbon reserves and total organic carbon (TOC) prediction. These approaches employed the MVA technique as a future selection method in source rock evaluation. We used geochemical data from 30 core samples taken equally from wells SS-5 and SS-7. Geochemical parameters, namely TOC, free hydrocarbon, thermal pyrolysis hydrocarbon, hydrogen index, production index, and oxygen index, were determined for statistical evaluation. IBM SPSS statistical software and MATLAB (R2020a) were used for MVA and ML, respectively. The performance of the models built using MVA and ML were evaluated by, among others, coefficient of determination (R 2 ) and mean square error (MSE). Findings revealed that fair through good to excellent source rock with TOC ranging from 0.85 to 2.95 wt% are hosted in the Triassic beds of Tanga. A high 1.61% Ro at a mature peak of 463 °C predominates with the existence of type III/II kerogen that can produce both oil and gas. Considering TOC prediction from conventional well log data, optimized Gaussian process regression showed the best performance followed by MVA and support vector machine, giving the MSEs of 0.5629, 0.6172, and 0.7023, respectively. In terms of prediction accuracy, their R 2 values of 0.952, 0.9346, and 0.835, respectively, were in good agreement with the geochemical results. The concurrence of geochemical analysis, ML, and MVA revealed that the Tanga basin has great hydrocarbon potential of great economic importance. The study revealed that combining MVA and other methods can be applied to assess the hydrocarbon resource potential of other prospects around the globe.