Improved porosity estimation in complex carbonate reservoirs using hybrid CRNN deep learning model

• Published in: Earth science informatics 2024-10, Vol.17 (5), p.4773-4790
• Main Authors: Mehrabi, Amirreza, Bagheri, Majid, Bidhendi, Majid Nabi, Delijani, Ebrahim Biniaz, Behnoud, Mohammad
• Format: Article
• Language: English
• Subjects: Accuracy; Artificial neural networks; Carbonates; Core analysis; Correlation coefficient; Correlation coefficients; Deep learning; Earth and Environmental Science; Earth Sciences; Earth System Sciences; Information Systems Applications (incl.Internet); Machine learning; Neural networks; Oil and gas fields; Oil fields; Ontology; Performance evaluation; Performance prediction; Petroleum; Petroleum industry; Porosity; Recurrent neural networks; Reservoirs; Simulation and Modeling; Space Exploration and Astronautics; Space Sciences (including Extraterrestrial Physics; Spatial dependencies; Well logs
• ISSN: 1865-0473; 1865-0481
• DOI: 10.1007/s12145-024-01419-y

This paper aims to improve porosity estimation in complex carbonate reservoirs by proposing a hybrid CRNN deep learning model. The objectives include addressing the challenges associated with porosity estimation in heterogeneous carbonate reservoirs and evaluating the performance of the CRNN model in accurately predicting porosity based on well-log data. The overall approach involves integrating CNN and RNN architectures within the CRNN model to effectively extract and combine relevant information from well logs. The model is trained using a dataset consisting of well-log and core analysis data from an Iranian carbonate oil field. Well-log data is used as the input including GR, DT, RHOB, LLD, and NPHI for model training, while core data is utilized for model validation. The model's performance is compared with the traditional MLP model in terms of accuracy and generalization. The proposed hybrid CRNN model demonstrates superior performance in predicting porosity values at new locations where only well-log data are available. It outperforms conventional neural network models, as evidenced by the significant improvement in the correlation coefficient between the model predictions and core data (from 0.67 for the MLP model to 0.98 for the CRNN model). The CRNN model's ability to capture complex spatial dependencies within heterogeneous carbonate reservoirs leads to more accurate porosity estimations and valuable insights into reservoir characterization. This paper presents novel and additive information to the existing body of literature in the petroleum industry. The hybrid CRNN model, combining CNN and RNN architectures, offers a unique approach to porosity estimation in complex carbonate reservoirs. By effectively integrating spatial and temporal patterns from well-log data, the model demonstrates higher accuracy rates and improved generalization capabilities. The findings contribute to the state of knowledge by providing a robust and efficient tool for accurate porosity prediction, which can assist in reservoir characterization and enhance decision-making in the petroleum industry.