Identification of natural fractures using resistive image logs, fractal dimension and support vector machines

Abstract

The purpose of this research is to apply a new approach to identify natural fractures in wells in a hydrocarbon reservoir using resistive image logs, fractal dimension and support vector machines (SVMs). The stratigraphic sequence investigated by each well is composed of Cretaceous calcareous rocks from the Catatumbo Basin, Colombia. The box counting method was applied to image logs in order to generate a curve representing variations of fractal dimension in these images throughout each well. The arithmetic mean of fractal dimension showed values ranging from 1,70 to 1,72 at the mineralized fracture intervals, and from 1,72 to 1,76 at the open fracture intervals. Morphological classification between open and mineralized natural fractures is performed using corelogs integration in a pilot well. Fractal dimension of images along with gamma rays and resistivity logs were employed as the input dataset of a SVM model identifying intervals with natural open fractures automatically, shortly after logs acquisition and previous to its interpretation by specialists. Although final results were affected by borehole conditions and logs quality, the SVM model showed accuracy between 72,3% and 82,2% in 5 wells evaluated in the studied field.

Keywords

• Fractal dimension
• resistive image logs
• box counting method
• natural fractures
• hydrocarbon reservoir
• Catatumbo basin
• support vector machines (svms)