Improved Predictability of In-Situ-Combustion Enhanced Oil Recovery

In-situ combustion (ISC) possesses advantages over surface-generated steam injection for deep reservoirs in terms of wellbore heat losses and generation of heat above the critical point of water. ISC also has dramatically lower requirements for water and natural gas, and potentially a smaller surface footprint, in comparison with steam. In spite of its apparent advantages, prediction of the likelihood of successful ISC is unclear. Conventionally, combustion tube tests of a crude oil and rock are used to infer that ISC works at reservoir scale and estimate the oxygen requirements. Combustion tube results may lead to field-scale simulation on a coarse grid with upscaled Arrhenius reaction kinetics. As an alternative, we suggest a comprehensive workflow to predict successful combustion at the reservoir scale. The method is derived from experimental laboratory data and simulation models at all scales. In our workflow, a sample of crushed reservoir rock or an equivalent synthetic sample is mixed with water/brine and the crude-oil sample. The mixture is placed in a kinetics cell reactor and oxidized at different heating rates. An isoconversional method is used to estimate kinetic parameters vs. temperature and combustion characteristics of the sample. Results from the isoconversional interpretation provide a first screen of the likelihood that a combustion front is propagated successfully. Then, a full-physics simulation of the kinetics cell experiment is used to predict the flue gas composition. The model combines a detailed pressure/volume/temperature (PVT) analysis of the multiphase system and a multistep reaction model. A mixture identical to that tested in the kinetics cell is also burned in a combustion tube experiment. Temperature profiles along the tube, as well as the flue gas compositions, are measured during the experiment. A high-resolution simulation model of the combustion tube test is developed and validated. Finally, the high-resolution model is used as a basis for scaling up the reaction model to field dimensions. Field-scale simulations do not use Arrhenius kinetics. As a result, significant stiffness is removed from the finite-difference simulation of the governing equations. Preliminary field-scale simulation shows little sensitivity to gridblock size, and the computational work per timestep is much reduced.