An artificial intelligence approach for predicting water-filled porosity and water saturation for carbonate reservoirs using conventional well logs

Water saturation is the most crucial parameter obtained by petrophysical analysis to estimate the potential reserve in oil and gas reservoirs. This parameter is usually obtained by either industry-known water saturation correlations, such as Archie, which carries inherently a wide room of uncertainties, or by the laboratory measurement through the Dean-Stark method. Dielectric logging tool is designed specifically to obtain accurate water saturation measurements that are independent of lithology, environmental effects, and water salinity. However, running the tool will add some complexity to the logging job and impose an extra cost. Therefore, this work aims to utilize artificial intelligence (AI) to obtain fast and reliable estimations for the dielectric water saturation in carbonate rocks utilizing conventional logs. The workflow of developing an AI model to obtain the dielectric water saturation from the conventional well logs went through two stages. Firstly, a new model was developed to predict the dielectric water-filled porosity (PWXO) from the conventional logs. Then, the dielectric’s invaded zone water saturation (SXO) was predicted based on the conventional logs and the PWXO. The conventional logs involved in this study are bulk density (RHOZ), photoelectric lithology factor (PEF), gamma-ray (GR), micro-resistivity (RXOZ), induction resistivities (AF10, AF20, AF30, AF60, and AF90), spontaneous potential (SP), and neutron porosity (NPHI). On the other hand, the actual values of the water-filled porosity (PWXO) and invaded zone water saturation (SXO) were obtained from the dielectric logs. Three AI techniques were used which are artificial neural networks (ANN), adaptive neuro-fuzzy inference system (ANFIS), and support vector machine (SVM). In addition, several evaluation indices were utilized to assess the prediction performance of the developed models, such as normalized root-mean-square error (NRMSE) and Pearson coefficient of correlation (CC). Also, the model inputs and structure were adjusted to achieve the best prediction performance. The water saturation (SXO) can be predicted from the conventional logs with an NRMSE of 0.04 and CC of 0.62; however, including the dielectric porosity (PWXO) as an input led to improve the prediction performance, CC is 0.98 and NRMSE is 0.003. Overall, this work will add a significant contribution to reservoir characterization by providing fast and reliable estimations of the water saturation using the conventional log without the need to run the dielectric tool. Furthermore, it can be used to obtain accurate values of cementation factors, tortuosity factors, and clay minerals’ cation exchange capacity (CEC) values which can be used later as the constants required for the widely known correlations to calculate the reservoir's water saturation. © The Author(s), under exclusive licence to Springer-Verlag London Ltd., part of Springer Nature 2025.