ABSTRACT
Determination of uncertainty in the description of reservoir properties is an important aspect of enhanced oil recovery projects. This paper compares four geostatistical techniques: simulated annealing method (SAM), source point method (SPM), fast fourier transform method (FFTM), and turning band method (TBM) in the generation of property distribution. The four methods are tested and compared by using an exhaustive minipermeameter permeability measured data set, and a tracer flow experiment on a rock sample. The methods are used in conditional simulation, assuming some known permeability data. The conditioned fields are then used in the simulation of miscible tracer flow through the sample rock and compared with the actual tracer concentration curves. The methods are compared on the basis of 1) spatial distribution and the variograms of the generated fields, 2) global error variance, 3) computational cost of generated fields, 4) effect on flow dynamics, and 5) tracer production curves.
Comparison of the four methods has shown that SAM produces the best fit of the experimental variograms without regard to the type of variogram (e.g., trend, etc.). The FFTM and TBM can utilize many types of theoretical variograms. All four methods can produce anisotropic fields. Although the spatial distribution obtained by SAM preserves the main features of the original data better than the other three methods, the global error variance in SAM is 25% more than in SPM, FFTM, and TBM. When considering computer run time and storage, SPM is the most efficient method, followed by TBM, FFTM, and finally SAM.
Finally, the conditioned permeability fields from these methods are used in the simulation of a miscible tracer flow through the rock sample and compared with the actual tracer output curves. It is observed that a very good fit of the simulated variogram to the experimental ones, will not necessarily give a good production history match. Rather, based upon similarity between the simulated and experimental tracer output concentration histories, the best results were obtained with SPM, then TBM, SAM, and finally FFTM.
