Stochastic inverse modeling of transient laboratory-scale three-dimensional two-phase core flooding scenarios

Abstract

We develop a comprehensive and efficient workflow for a stochastic assessment of key parameters governing two-phase flow conditions associated with core-scale experiments. We rely on original and detailed datasets collected on a Berea sandstone sample. These capture the temporal evolution of pressure drop across the core and three-dimensional maps of phase saturations (determined via X-ray CT) in oil- and brine-displacement flooding scenarios characterized by diverse brine/oil viscosity contrasts. Such experiments are used as a test-bed for the proposed stochastic model calibration strategy. The latter is structured across three main steps: (i) a preliminary calibration, aimed at identifying a behavioral region of the model parameter space; (ii) a Global Sensitivity Analysis (GSA), geared towards identification of the relative importance of model parameters on observed model outputs and assessment of non-influential parameters to reduce dimensionality of the parameter space; and (iii) a stochastic inverse modeling procedure. The latter is based on a differential-evolution genetic algorithm to efficiently explore the reduced parameter space stemming from the GSA. It enables one to obtain a probabilistic description of the relevant model parameters through their frequency distributions conditional on the detailed type of information collected. Coupling GSA with a stochastic parameter estimation approach based on a genetic algorithm of the type we consider enables streamlining the procedure and effectively cope with the considerable computational efforts linked to the two-phase scenario considered. Results show a remarkable agreement with experimental data and imbue us with confidence on the potential of the approach to embed the type of rich datasets considered towards model parameter estimation fully including uncertainty.