Machine Learning–Enhanced Analytical Modeling of In-Situ Combustion with Experimental Evaluation of Carbonate Decomposition and Mineral/Gas CO2 Interactions

Summary

In-situ combustion (ISC) is a mature enhanced oil recovery technique, historically applied to heavy- and high-viscosity oil reservoirs. Its application to light-oil systems remains limited due to high ignition-temperature requirements, reduced fuel availability, and increased operational sensitivity. In this study, we evaluate the laboratory-scale feasibility of ISC in a US light-oil reservoir through three combustion-tube experiments (two wet and one dry) conducted using reservoir rock samples from three different wells to examine mineralogical variability.

X-ray diffraction (XRD) analysis revealed that approximately 45 wt% of the reservoir rock consists of calcite and dolomite, leading to significant carbonate decomposition at ISC combustion-front temperatures. This mineral-derived carbon dioxide (CO2) generation complicates conventional ISC stoichiometric analysis, which typically attributes all produced CO2 to crude-oil oxidation. To address this limitation, an integrated framework combining combustion-tube experiments, thermogravimetric analysis (TGA), and machine learning (ML) was developed to quantify carbonate-derived CO2 and correct analytical ISC models. A total of 5,748 temperature/mass-loss data points obtained from 35 TGA experiments on calcite and dolomite were used to train and test random forest and extreme gradient boosting (XGBoost) models. TGA results confirmed that carbonate decomposition initiates at temperatures as low as 350–400°C, particularly under higher heating rates (15–20°C/min), well below commonly assumed thresholds. XGBoost outperformed random forest, achieving a test coefficient of determination (R2) of 0.986 and root mean squared error (RMSE) of 0.504, enabling more accurate correction of CO2 source attribution. Incorporating these corrections improved combustion-parameter estimates, including crude-oil heat release (~15,000 BTU/lbm) and energy consumption associated with carbonate decomposition (~3,000 BTU/lbm).

To explore CO2 mitigation mechanisms under realistic ISC conditions, natural mineral filters composed of olivine, dolomite, and ultramafic rock were installed at the combustion-tube outlet and exposed to actual ISC flue gases for 6.7 hours. The filters exhibited measurable carbon uptake of 2.39–7.81 mg per 100 mg of mineral. Although no crystalline carbonate phases were detected by XRD, converging evidence from scanning electron microscopy (SEM)-energy-dispersive X-ray spectroscopy (EDS), Fourier-transform infrared (FTIR) spectroscopy, and post-exposure TGA suggests surface-mediated formation of amorphous magnesium (Mg) carbonate or related oxygenated carbon species. These results demonstrate a laboratory-scale CO2 uptake mechanism based on gasmineral surface interactions rather than bulk mineral carbonation.

Overall, this study highlights the critical role of carbonate decomposition in ISC CO2 accounting, demonstrates the necessity of correcting analytical combustion models for rock-derived CO2, and provides experimental evidence of mineral/gas interactions that may contribute to partial CO2 mitigation at the surface. While not indicative of field-scale deployment, the findings establish a mechanistic basis for improving ISC analytical interpretation and guiding future studies on emission reduction strategies in carbonate-rich light-oil reservoirs.