Application of multilayer feed-forward neural network to predict porosity and water saturation volumes in parts of onshore Niger Delta

Abstract

This study presents the successful implementation of a Multilayer Feed-Forward Neural Network to predict porosity and fluid saturation in two reservoirs using well log and 3D seismic data. By leveraging unknown nonlinear relationships in data between well logs and reservoir parameters, the technique accurately determines specific petrophysical features of reservoir rocks under various compaction conditions. Addressing the challenge of predicting petrophysical parameters, especially saturation, due to unclear correlations with seismic elastic characteristics, the research employed machine learning methods for seismic reservoir characterization. Prior to applying the neural network, seismic inversion was used to link geology, seismology, well-logs, and rock physics. The results identified hydrocarbon-bearing zones based on low acoustic impedance values. Training the neural network with six attributes for water saturation prediction and five for porosity prediction showed significant correlations, with actual and predicted water saturations and porosities having correlation coefficients of 72.6% and 80.6%, respectively. These parameters were then extended over reservoirs A and B to map their distribution in the field, proving the workflow's validity in accurately predicting water saturation and porosity.