Abstract
An accurate estimation of rock porosity is essential to determine the storage capacity of hydrocarbon reservoirs. As the most realistic technique, the porosity value can be measured directly for core samples. However, laboratory core analysis is costly and sometimes unfeasible, leading to developing other indirect estimation methods from geophysical well logs or seismic data. Besides, estimating porosity in carbonate reservoirs is often difficult due to inherent heterogeneity and complex geological environments. In this paper, we proposed a hybrid deep learning model that integrates the CNN and RNN architectures to predict the porosity in heterogeneous carbonate reservoirs based on well-log data. The novelty of this approach lies in the hybridization of CNN and RNN architectures, which allows the effective extraction and combining of relevant information from the well log data and an accurate estimate of porosity in challenging heterogeneous carbonate reservoirs. The ability of CRNNs to capture the complex spatial dependencies within these reservoirs can lead to more accurate porosity estimations and valuable insights into reservoir characterization. This hybrid deep model consists of several hidden layers, convolutional, pooling, recurrent, and dense layers. The dataset consists of well-log and core analysis from the Iranian carbonate oil field, which was used to build a predictive model. well-log data was used as model input and core data for model validation. We compare the performance of the CRNN model with the MLP model for predicting porosity values from well-log data. Results show the proposed model can accurately predict porosity values at new locations where only well-log data are available, and they outperform conventional neural network models in terms of accuracy and generalization. The correlation coefficient between the model and core data improves from 0.67 for the MLP model to 0.98 for the CRNN model.