RISK REDUCTION WITH A FUZZY EXPERT EXPLORATION TOOL

Incomplete or sparse information on geologic or formation characteristics introduces a high level of risk for oil exploration and development projects. Expert systems have been developed and used in several disciplines and industries, including medical diagnostics, with favorable results. A state-of-the-art exploration ''expert'' tool, relying on a computerized data base and computer maps generated by neural networks, is proposed through the use of ''fuzzy'' logic, a relatively new mathematical treatment of imprecise or non-explicit parameters and values. This project will develop an Artificial Intelligence system that will draw upon a wide variety of information to provide realistic estimates of risk. ''Fuzzy logic,'' a system of integrating large amounts of inexact, incomplete information with modern computational methods to derive usable conclusions, has been demonstrated as a cost-effective computational technology in many industrial applications. During project year 1, 90% of geologic, geophysical, production and price data were assimilated for installation into the database. Logs provided geologic data consisting of formation tops of the Brushy Canyon, Lower Brushy Canyon, and Bone Springs zones of 700 wells used to construct regional cross sections. Regional structure and isopach maps were constructed using kriging to interpolate between the measured points. One of the structure derivative maps (azimuth of curvature) visually correlates with Brushy Canyon fields on the maximum change contours. Derivatives of the regional geophysical data also visually correlate with the location of the fields. The azimuth of maximum dip approximately locates fields on the maximum change contours. In a similar manner the second derivative in the x-direction of the gravity map visually correlates with the alignment of the known fields. The visual correlations strongly suggest that neural network architectures will be found to correlate regional attributes with individual well production. On a local scale, given open-hole log information, a neural network was trained to predict the product of porosity and oil saturation as reported in whole core analysis. Thus a direct indicator of an oil show is available from log information. This is important in the thin-bedded Delaware sand reservoirs. Fuzzy ranking was used to prioritize 3D seismic attributes that were then correlated to formation depth with a neural network. The results were superior to those obtained using linear interpolation or low order polynomial interpolation as time-to-depth conversion tools. A radial basis function neural network was developed and used as a log evaluation tool. This new technology gives an additional tool to the more commonly used multilayer perceptron (MLP) neural network. An interactive web based MLP, PredictOnline, was coded in Java and made available to consortium members for beta testing. PredictOnline demonstrates the power of Java programming language for web-based applications. A draft design of the Fuzzy Expert Exploration (FEE) Tool system based on readily available software was completed. The recent development of a Java Expert System Shell, JESS, facilitates expert rule development