Shale Gas Rock Properties Prediction using Artificial Neural Network Technique and Multi Regression Analysis, an example from a North American Shale Gas Reservoir

Abstract

Estimation of reservoir parameters has always been a challenge for shale gas reservoirs. This study has concentrated on neural network technique and multiple regression analysis to predict reservoir properties including porosity, permeability, fluid saturation and total organic carbon content from conventional wireline log data for a large North American shale gas reservoir. More than 262 core analysis data from 3 wells were used as ``target' and ``response' for neural network and multiple regression analysis. Common log data available in three wells including GR, SP, RHOB, NPHI, DT and deep resistivity were used as ``input' and ``predictor'. This study shows that reservoir parameters could be better estimated using the neural network technique than through multiple regression. The neural network method had a correlation coefficient greater than 80% for most of the parameters. Although providing a set of algorithms, multiple regression analysis was less successful for predicting reservoir parameters.