A new approach to predict failure parameters of carbonate rocks using artificial intelligence tools

Today's economic conditions emphasize the need for better engineering designs in drilling, well completion, and reservoir production operations. A knowledge of the mechanical behavior of reservoir rocks is important in connection with wellbore stability problems, fracturing operations, subsidence problems and sand production problems. To carry out any aforementioned operations, continuous profiles of rock mechanical parameters are needed. Retrieving reservoir rock samples throughout the depth of the reservoir section and performing laboratory tests on them are extremely expensive as well as time consuming. Consequently, rock failure parameters are often correlated with geophysical well logs such as bulk density, compressional and shear wave velocities, which may result in continuous profiles of failure parameters throughout the depth of the well section. However, these estimations not truly represent the reservoir in-situ stress-strain condition. Therefore, failure parameters are estimated from empirical correlations. Most of the previous correlations were developed using linear or non-linear regression techniques. Artificial intelligence tools once optimized for training can successfully model highly complex and nonlinear relationship between different well logs as inputs and failure parameters as output. Therefore, the objective of this paper is to accurately predict the failure parameters (Angle of internal friction and Cohesion) of the rock by using basic geophysical well logs namely; bulk density, neutron porosity, compressional, and shear wave travel times, from three artificial intelligence techniques namely; Adaptive Neuro Fuzzy Inference System (ANFIS), Support Vector Machine (SVM) and Functional Network (FN). The data used in this study were obtained from more than 10 wells located in giant carbonate reservoir. A comparison between the predicted failure parameters by the proposed models with actual laboratory measured data reveals that new proposed model predicted with significantly less average absolute percentage error (AAPE) and high coefficient of determination (R2). The developed technique can be useful for geo-mechanical engineers to determine continuous failure parameters profiles throughout the desired depth, when no laboratory tests are available.