Estimation of fracture network properties from FMI and conventional well logs data using artificial neural network

Abstract

This study presents a robust artificial neural network technique to estimate the fracture network properties including fracture density and fractal dimension to create the reservoir subsurface fracture map. Overcoming the limitations of the used data in characterizing the fracture properties is deeply investigated in this study by employing the neural network technique to establish a relationship between available data by developing a new correlation using conventional well logs and borehole images. Subsequently characterize fracture properties in terms of fracture density and fractal dimension. The neural network system in this study is developed based on FORTRAN language to establish in house code with the back-propagation algorithm as a learning procedure. The sigmoid function is used as well for output prediction. Two new correlations are generated, one for fractal dimension and other one for fracture density as function of conventional well logs. The developed correlations are used to generate a continuous 3D subsurface fracture map for the studied reservoir. The data are collected from five wells drilled in the reservoir include conventional well logs and Full bore micro-resistivity image data. The used data are distributed 80% for the training and 20% for the testing only from 4 wells. The results show that, the developed correlations able to predict the fracture properties precisely with mean square error = 0.05 and R square = 0.997 for the training process and with R square = 0.97 for testing. A validation is performed using a data from well#5 which are not used in the training process. The results of validation show that fracture properties are predicted with R square = 0.99. The subsurface fracture map for the studied reservoir is successfully generated using the obtained 3D fractal dimension and fracture density. In addition, the created subsurface fracture map is validated by using the available reservoir production data.