Abstract
In this paper we present a methodology to verify and update geostatistically based reservoir models using numerical simulation and automatic history matching of well-test data. A Bayesian estimation technique provides the framework for the inversion procedure used to update reservoir models. A restricted-step Gauss-Newton method and an extension of Carter's method for sensitivity coefficients make the methodology efficient and practical. Reservoir models (permeability, porosity and skin factor) can be updated based on individual simulation cells, geological feathers, or on constant multipliers applied in the well-test radius of investigation. When short-term well-test data is available from many wells, we found it useful to calculate property multipliers around each well and then interpolate the multipliers to unsampled areas in the reservoir.
Field examples demonstrate the utility of conditioning reservoir models with well-test data. In the examples, flowing bottomhole pressure mismatches are as high as 2000 psi and core based permeabilities deviate by two orders of magnitude from the well-test results. After conditioning to well-test data, pressure mismatches are significantly reduced, permeabilities are updated to the correct magnitude, and the skin factor closely matches the classic well-test analysis result.
We applied our updating methodology to a full-field simulation study on a large carbonate reservoir. Comparisons of historical pressure and production data to the updated full- field model demonstrate good agreement.
