Abstract

A new technique based on structure analysis to identify and characterize spatial differences in natural fracture density is presented. Production from existing wells is assumed to be representative of fracture frequency, which is correlated to the structure. A neural network is used to establish correlations which can then be used to predict fracture frequency in areas where well coverage is sparse, but the structure can be mapped.

The slopes, curvature, bed thickness, and lithology of the structure influence the occurrence of natural fractures. The importance of structure on fracture frequency is well documented. Murray (1968) presented a correlation between the structural curvature taken in one direction, and the fracture-dominated production in the Williston Basin Sanish Pool. Ouenes (1994) introduced the concept of curvature in multiple directions as an important fracture frequency parameter. Structure can be obtained from 3D seismic information, which is preferred, or interpolated from well control data. The problem rests in correlating the multitude of structural parameters, plus thickness and lithology, with fracture frequency as defined by production.

Neural networks are well suited to handling multiple parameter correlations. Humans can visually correlate two parameters seen in an x-y plot of the data. A regression line can be fit through the data points and an equation developed to interpolate points anywhere on the line. The ability of neural networks to correlate multiple parameters can be adapted to associate structural parameters and thickness of a formation with inferred fracture intensity. The neural network can then be used to estimate fracture intensity anywhere the structure and thickness are known.

Examples are presented that show how this new technology was applied to two actual reservoirs located in slope basin and fluvial-deltaic environments.