Automated stratigraphic interpretation of well‐log data

Abstract

A computerized well‐log stratigraphic interpretation system based on artificial intelligence can be seen as two steps, contact recognition and interval identification. Unlike previous approaches to mathematical zonation which are essentially statistical, consideration of the geologic environment is included for effective interpretation. Following the logic of human experts, the system first determines the log signature of the contacts. An algorithm incorporating linear regression with variable breakpoints is used to describe the structure of log curves. Log features are mapped to the assigned signature of contacts. Multiple logs are taken into account to set up the final set of contacts, which divides the test borehole into a string of intervals with attributes. A pattern string of intervals is based on the integrated geologic column for the section where the borehole is located. A string‐to‐string matching program then determines the optimal map of the test borehole as the pattern to generate a geologic column based on the computer suggestion. The cost formulas for matching represent knowledge about the stratigraphy of the area under study, and the string‐to‐string matching algorithm includes this knowledge of stratigraphy. In test boreholes in an oil‐sand deposit, 86 percent of the computer identifications of intervals were consistent with those determined by geologists from core descriptions, suggesting that the design concept and computer algorithm are consistent with the characteristics of the interpretation problem.