A Machine Learning Approach for Predicting Rock Brittleness fromConventional Well Logs

摘要

In modern logging practices,the Poisson ratio and the Young Modulus as measures of rock Brittleness maybe estimated from the dipole sonic and bulk density logs.This is important when subterranean formationsare considered for fracturing or when unintended fracturing can be a concern under high injection gradientsin fluid injection processes.For most oilfields,the bulk of well log suites run in the past has been limitedto basic lithology logs and occasionally some porosity logs.In particular,the sheer sonic velocity log,animportant component for estimation of geomechanical properties,has not been a standard measurement oncommon log suits.In this paper,we present the result of a study where shear travel time is correlated with measurementsfrom caliper and shale content.The training set we used consisted of well logs for wells that included theshear travel time.We experimented with various approaches and developed a process for in cooperatingDNN(deep neural network)to correlate the shear log data to other measurements.