Use of Well Logging Data to Estimate Fluid Saturation Based on Artificial Neural Network Algorithms

Document Type : Original Article

Authors

1 Department of Petroleum Engineering, Kho.C., Islamic Azad University, Khomeinishahr, Iran

2 Department of Petroleum Engineering, Kho.C., Islamic Azad University, Khomeinishahr, Iran; Stone Research Center, Kho.C., Islamic Azad University, Khomeinishahr, Iran

3 Department of Mechanical Engineering, Kho.C., Islamic Azad University, Khomeinishahr, Iran

10.22077/jgm.2025.9704.1052

Abstract

     Among all the methods used for determining fluid saturation of the reservoir rock, the ability of neural networks to predict fluid saturation in reservoir rock is of great interest to researchers. This study gathers the necessary data for estimating this important reservoir parameter and the variables involved in the process. Afterward, artificial neural networks (ANNs) and particle swarm optimization (PSO) algorithms are combined to provide a proper and accurate model for estimating water saturation. This combination provides an outstanding model in which fluid saturation distribution at any point in one of Iran’s carbonate oil reservoirs can be obtained. To predict the water saturation value as the model output, several input parameters, including depth, gamma ray, resistance, neutron, micro-spherical resistance, and spontaneous potential logs, are employed. The multi-layer perceptron neural network (MLP) and radial basis function neural network (RBF) are the two models used, and the accuracy of each model is examined. Although the relationship between fluid saturation in the reservoir and logging information is completely nonlinear, these two artificial intelligence (AI) models can very well recognize this nonlinear relationship and provide great predictions with high correlation coefficient (R) values and low average absolute relative deviation (AARD) and root mean square error (RMSE) values. The values of R, AARD, and RMSE for the MLP model are obtained as 0.9739, 33.24, and 0.0824, respectively. Those for the RBF model are 0.9986, 7.47, and 0.0024, respectively, reflecting that the RBF model is superior to the MLP model due to its higher R value and lower AARD and RMSE values.

Keywords