A New Model to Predict the Dielectric Properties by Employing Artificial Neural Networks

Abstract

Abstract Studying the dielectric properties of reservoir systems can provide crucial information such as water saturation, total porosity, and brine salinity. Different models have been developed to estimate the dielectric parameters such as the conductivity and relative permittivity. However, multiple assumptions are used, and considerable computational processes are employed. This work aims to develop a new model to predict dielectric properties utilizing artificial neural networks (ANN). In this study, the relative permittivity (dielectric constant) and conductivity for shale rocks were predicted based on the raw measured data. Multi-frequency dielectric measurements were carried out using a coaxial dielectric probe. The model input is mainly the reflection caused by the samples at different frequencies. The relative permittivity and conductivity of the samples were predicted using the ANN model at a frequency range of 1MHz to 3GHz, in order to capture multiple polarization mechanisms. Different types of error indexes were determined to indicate the prediction performance. Also, a comprehensive analysis was conducted to optimize the performance of the developed ANN model. The results showed that the ANN-based model can effectively predict the dielectric parameters. A correlation coefficient of more than 0.94 was obtained. The optimized ANN model consists of one hidden layer with 7 neurons and uses Levenberg-Marquardt (trainlm) function during the training phase. Thereafter, a new equation was extracted from the developed ANN model to allow fast and accurate estimations for the dielectric properties. The developed equation can be applied to a wide range to estimate the relative permittivity and conductivity of shale rocks. For the first time, an AI-based model is developed to estimate the dielectric properties of shale rocks. The proposed model can predict the dielectric constant and conductivity in less time and with high reliability.