Reservoir characterization requires accurate estimates of permeability. The commonly used porosity-permeability transforms are often inadequate as permeability is also a function of clay distribution, sorting, pore connectivity, tortuosity and variations in other petrophysical properties. More robust permeability estimation can be made by integrating multiple logs in the transform instead of just porosity.
This paper deals with a novel technique for deriving permeability by correlating multiple well logs with core permeability using non-parametric regression methods. First, we classify the well log data into electrofacies based on the ‘similarity’ of their response. This electrofacies classification does not require any artificial subdivision of the data population but follows naturally based on the unique data values reflecting minerals and lithofacies. A combination of principal component and model-based cluster analysis are used to characterize the electrofacies. Secondly, we apply a non-parametric regression technique to predict permeability using well logs within each electrofacies. The main advantage of this technique is that it is primarily data-driven as opposed to model driven and does not require a priori specification of functional forms, which makes conventional multiple regressions difficult and often biased.
The proposed technique was used in a deepwater reservoir system consisting of thick massive sand beds along with thinly bedded sand-shale sequences. The work flow consists of electrofacies identification, facies-wise permeability transform generation, calibration with DST data, and finally permeability population in the geological model using cloud transforms and Sequential Gaussian Simulation (SGS). The permeability transforms generated from core and well log data were validated via blind tests whereby we predicted permeability in another cored well that was not included in the correlation. For further validation, we also used the transform-derived permeabilities in analyzing the DST results in a blind well where a laminated sand section was tested. A reasonable match of the permeability-thickness product was observed. Finally, the permeability transforms were integrated into a 3-D geologic modeling using SGS and cloud transform. It was observed that our approach correctly captured the porosity-permeability scatter in the geologic model for various facies groups.