Numerical Simulation of a Kinetic Cell for in-situ Combustion. A Parametric Study and History Matching Aspects

Abstract

Abstract The in-situ combustion method is an enhanced oil recovery technique based on the injection of air in petroleum reservoirs with the aim to burn a portion of hydrocarbons. This reduces the oil viscosity improving substantially the oil mobility. Simultaneously other phenomena take place as: distillation, segregation, oil upgrading, among others. In this work, a mathematical model to simulate oil combustion for kinetic cell experiments is presented. The model includes four-phases, nine components and four chemical reactions: coke formation, heavy oil fraction combustion, light oil fraction combustion and coke combustion. This formulation is commonly used to simulate in-situ combustion projects at combustion tubes- and petroleum reservoir-scales. The mass and energy balances were formulated leading to one set of highly coupled ordinary differential equations, which was numerically solved. The predictive model capabilities were tested by comparison with lab data, and it was found that CO and CO2 productions, oxygen uptake and cell temperature evolution agree well with experimental results. At one preliminary stage, the parameters fitting experimental results were inferred by individual manipulation until the best results were found. These parameters were perturbed in order to identify those parameters dominating the global dynamic of process. We found that energy activations and the mass density of oil components are the dominant parameters. We suggest that history matching processes must be focused over these parameters, and for this end, the implementation of advanced computational routines to solve multivariable inverse problems is recommended. In this work, we developed two automatic history matching techniques: one process based on Newton’s method and the second one based on evolutionary algorithms. The Newton’s method showed problems to find the minimum error, meanwhile the evolutionary algorithm was able to optimize the dominant parameters, but at the expense of slow convergence.