Reduced Order Modeling Of Multiphase Flow In Fractured Formation

Abstract:

The success of a thermal water flood for enhanced oil recovery (EOR) depends on (1) accurate modeling of the nonlinear multiphase flow processes, and (2) detailed representation of the geometrical and hydraulic details of the fracture network. The resulting high-resolution numerical model is typically computationally demanding. Here, we compare two methods for approximating high-resolution solutions: the Proper Orthogonal Decomposition (POD) Mapping method, and the POD-Gaussian Process Regression (GPR) method. The POD Mapping method utilizes an efficient low-resolution model for prediction after training a reduced order model (ROM) using high- and low-resolution solutions. On the other hand, the POD-GPR method constructs a statistical ROM that directly maps the input parameters to the high-resolution solutions. The approximation error can be quantified either through an error estimator (POD Mapping method) or a variance estimate (POD-GPR method). Initial results indicate that the POD Mapping method is more accurate than the POD-GPR method when the same set of training data is used. This work was supported, in part, by the U.S. Dept. of Energy under Contract No. DE-AC02-05CH11231.