Abstract

Expert systems are an artificial intelligence tool, which store expert opinions and methodologies of analysis. They have been successfully implemented in the medical field to make diagnoses given lists and degrees of symptoms. Expert systems have also been implemented in the oil industry, but generally only to examine narrow problems with a very small focus of expert knowledge required. This paper describes a new expert system for oil prospecting which is currently under final development using the lower Brushy Canyon and Devonian plays of SE New Mexico as case studies. This Fuzzy Expert Exploration Tool (FEE Tool) will eventually be generalized so that users anywhere will be able to add their own knowledge and data and make rapid evaluations of a large number of potential drilling sites in a systematic and consistent manner thus reducing time to generate prospects.

Brushy Canyon regional databases were formed on a 40-acre grid system, with each of 60,478 locations being a potential drilling site. Regional structure, isopachs, net pay sand, total organic carbon, gravity and aeromagnetic data were mapped and gridded to this scale. Production data was available at 911 grid points (wells), which were hand verified to have produced from the Brushy Canyon. An additional, 75 wells were verified as Brushy Canyon dry holes, wells that had been cased and in which a serious effort to generate production had been attempted.

A Brushy Canyon specific expert system was developed using input from successful explorationists and statistical analysis of the regional and local data. Preliminary estimates of final drilling risk were calculated at all 60,478 locations and a mean estimate of predicted well quality of 0.476 was calculated. Approximately 4600 prospects are in the upper range {0.65, 1.0} and would be considered favorable drilling sites based on this preliminary analysis. In order to validate this preliminary result 89 new wells that were drilled since the input of new data to the project was frozen, were evaluated by the system, and a mean estimate of well quality of 0.673 was found, indicating, that as a group the drill sites would have been recommended by the FEE Tool. While not differentiating the wells by degree of success (amount of production) this test demonstrates that the FEE Tool can make intelligent decisions. Further and more detailed verification tests will be employed as the system matures. Future work includes migrating the software to examine a markedly different play, the Devonian carbonates of southeast New Mexico, to test the portability of the FEE Tool design.

The ultimate goal is a tool that will aid in minimizing risk in wildcat exploration and reservoir infill and expansion projects through a convenient, easy to use interface capable of accepting, storing and using expert knowledge to aid the human explorationist in identifying new prospects.