Use of nonparametric regression and inverse modeling for reservoir characterization

This dissertation focuses on improved techniques for data correlation and integration for reservoir characterization. First, we have utilized a non-parametric transformation and regression technique called ACE (alternating conditional expectation) to estimate permeability from well logs at the Salt Creek Field Unit (SCFU), Texas, a heterogeneous reef carbonate reservoir. The results have led to an enhanced reservoir characterization based on flow (permeability) rather than storage (porosity). This first, full-field application of ACE in a carbonate reservoir has demonstrated the strength and potential wide-scale applicability of non-parametric methods to predict permeability in heterogeneous reservoirs.; Second, development and application of a three-dimensional streamline simulator has been presented. The streamline simulator has been generalized to handle pressure update and changing well configurations in a more efficient manner. Other developments include incorporation of horizontal wells, gravity effects and miscible flooding. Comparison with a commercial finite-difference simulator has been made for a variety of examples. The speed of the streamline simulator can sometimes be orders of magnitude faster than standard numerical simulators making it ideal for dynamic data integration during reservoir characterization.; Third, we have developed a production data integration technique for estimating reservoir parameters such as porosity, permeability and residual oil saturation. We have discussed a simulated annealing based inversion approach and uncertainty analysis for estimating spatial distribution of permeability, porosity and NAPL saturation (Non Aqueous Phase Liquid) using tracer data. For characterizing NAPL saturation we follow a two-step procedure. First, we match a conservative tracer response to generate a permeability and porosity distribution. Next, a partitioning tracer response is matched by varying the NAPL saturation distribution. The non-uniqueness in the solution is addressed through the use of regularization techniques and/or prior information. An uncertainty analysis of the estimated parameters has also been performed to quantify resolution and averaging kernels. The proposed technique has been applied to synthetic and field examples. The field example is from the Hill Airforce Base, Utah where tracer tests were conducted in an isolated test cell. Tracer responses from 51 sampling locations are analyzed to determine permeability variations and NAPL saturation distribution in the test cell.