Prediction of Gamma Ray data from pre-stack seismic reflection partial angle stacks using Continuous Wavelet Transform and convolutional neural network approach

Abstract

Seismic inversion is one of the critical issues for geophysicists in the oil and gas industry. It is commonly used to estimate the distribution of facies-types and fluids in reservoirs across the value chain from exploration to development. However, the low vertical and spatial resolution of seismic data leads to large uncertainties in these estimations, which makes direct use of inversion results challenging. The objective of this study was the estimation of a 3D volume of Gamma Ray data from pre-stack seismic reflectivity partial angle stacks based on the Convolutional Neural Network architecture by representing data through the Continuous Wavelet Transform. The proposed methodology used 132 deviated wells from the Azeri and Chirag parts of the Azeri – Chirag – Gunashli field in the South Caspian Basin as training data. Blind tests show promising Gamma Ray predictions. To optimize the predictions, sensitivity analysis was performed on the Continuous Wavelet Transform parameters (wavelet types, range of pseudo-frequencies, etc.), the Convolutional Neural Network architecture (number of convolutional and pooling layers, dropout, activation functions) and the network parameters (batch size, number of epochs, optimizers, etc.). Due to the vast amount of data required to speed up the learning process, simulations were performed on a Graphics Processing Unit based on the Compute Unified Device Architecture platform.