A Comparison of Reservoir Mineralogy Prediction Utilizing Logging Data and Elastic Wave Velocities by Virtue of an Artificial Neural Network in South Pars Field

Abstract

In this research, the advantages of an artificial neural network were taken into consideration to predict hydrocarbon reservoir mineralogy in South Pars field and the algorithm chosen was a back-propagation (BP) algorithm. Mineralogy prediction was performed using logging data and dipole sonic imager (DSI)-derived elastic wave velocities. The efficiency of utilizing elastic wave velocities in improving prediction accuracy was evaluated as well. Minerals' volumetric percentiles were predicted in a 100-m interval in one of South Pars field's wellbores. When gamma-ray, density, neutron, and photoelectric effect (PEF) logs were used as the inputs, the mean square error (MSE) magnitude was 0.078 and when shear wave velocities (Vs) were added to the above-mentioned inputs, the MSE magnitude decreased to 0.065; shear wave velocities were substituted by compressional wave velocities (Vp), the MSE was 0.054, which implies that using Vp magnitudes in association with other log data would provide more accurate results than utilizing Vs magnitudes in addition to the log data. Using both Vp and Vs together with the log data resulted in a decrease in MSE to 0.046, which indicates a higher degree of accuracy in the responses.