History matching is a key task for field reviews. Over the past decades, assisted history matching techniques have effectively automated parts of the process. Methods commonly applied include genetic algorithms, adjoint methods, ensemble Kalman filters, streamline-based inversion, and conventional Newton-type methods. All these methods, however, are numerically expensive, often to a prohibitive degree for large and complex reservoirs.
A novel method aims to overcome this limitation. It combines the advantages of two well-established technologies: a numerically efficient Newton-type optimization scheme and full-physics proxy-type modeling using grid coarsening. It focuses on unknowns that can be readily re-parameterized, such as relative permeability and capillary pressure functions, fluid and rock properties, or fluid contacts. In addition, the statistical analysis of the parameter identification allows the establishment of confidence intervals, parameter correlations, and sensitivities.
Newton-type optimization methods can numerically be less expensive and potentially show superior convergence behavior under certain conditions. Although they are prone to converge to local minima and are less suitable for adjusting permeability or porosity distributions, they can be advantageous for smaller parameter spaces when using a re-parameterization approach. At the same time, grid coarsening allows drastically reduced runtimes while maintaining realistic geological and physical assumptions. Using the Newton-type calibration technique, the workflow suggested comprises initial screening, ranking of parameter sensitivities, and subsequent detailed parameter calibration to obtain the optimal parameter set.
A large reservoir dataset in which simulation runtimes would not allow for traditional assisted history matching approaches shows the advantages of the novel method. Convergence is rapid and minimal time is required to optimize sets of relative permeabilities, including endpoints and curvature of each individual function. As illustrated, the re-parameterization also allows imposition of a physically sensible solution. Most importantly, the statistical analysis of the results identifies key sensitivities in the reservoir, ultimately contributing to a better understanding of data gaps and uncertainties.