مؤسسه ژئوفیزیک دانشگاه تهران فیزیک زمین و فضا 2538-371X 50 4 2025 03 15 Enhancing Porosity Prediction Accuracy in Oil Reservoirs: Evaluating Hybrid Machine Learning Approaches Integrating Well Log and Core Data Enhancing Porosity Prediction Accuracy in Oil Reservoirs: Evaluating Hybrid Machine Learning Approaches Integrating Well Log and Core Data 91 113 100681 10.22059/jesphys.2025.377351.1007611 FA Amir Reza Mehrabi Department of Petroleum Engineering, Science and Research Branch, Islamic Azad University, Tehran, Iran. Majid Bagheri Department of Earth Physics, Institute of Geophysics, University of Tehran, Tehran, Iran. Majid Nabi Bidhendi Department of Earth Physics, Institute of Geophysics, University of Tehran, Tehran, Iran. Ebrahim Biniaz Delijani Department of Petroleum Engineering, Science and Research Branch, Islamic Azad University, Tehran, Iran. Mohammad Behnood Department of Petroleum Engineering, Science and Research Branch, Islamic Azad University, Tehran, Iran. Journal Article 2024 06 08 Accurate prediction of porosity holds significant importance across various domains within the oil and gas sector, encompassing activities such as reservoir delineation, well design, and production enhancement. However, conventional methodologies often encounter difficulties in capturing the intricate relationships among diverse data streams and porosity metrics. This study introduces a novel hybrid model framework aimed at refining the precision and resilience of porosity forecasts by integrating multiple machine learning methodologies and exploiting complementary data modalities. This hybrid architecture enables flexible and intricate integration of diverse models and data sources, potentially leading to enhanced overall porosity prediction accuracy. Notably, the proposed model incorporates several innovative elements, including the amalgamation of ensemble techniques and deep learning models tailored for sequential data, as well as the utilization of complementary data sources, such as well log and core data, to facilitate automatic feature learning and representation, thereby bolstering robustness and generalization capabilities. Experimental outcomes underscore the hybrid model's potential to achieve notable prediction accuracies, with R-squared values surpassing 0.93 on log data and 0.88 on core data sets, outperforming individual models. The model also exhibits commendable robustness and training efficiency, leveraging advanced methodologies such as ensemble techniques. In conclusion, this study underscores the promise of hybrid machine learning models as dependable tools for porosity prediction from core data. The insights gleaned from this research hold the potential to advance the understanding and optimization of porosity forecasting, thereby facilitating the formulation of more efficient reservoir management strategies. Accurate prediction of porosity holds significant importance across various domains within the oil and gas sector, encompassing activities such as reservoir delineation, well design, and production enhancement. However, conventional methodologies often encounter difficulties in capturing the intricate relationships among diverse data streams and porosity metrics. This study introduces a novel hybrid model framework aimed at refining the precision and resilience of porosity forecasts by integrating multiple machine learning methodologies and exploiting complementary data modalities. This hybrid architecture enables flexible and intricate integration of diverse models and data sources, potentially leading to enhanced overall porosity prediction accuracy. Notably, the proposed model incorporates several innovative elements, including the amalgamation of ensemble techniques and deep learning models tailored for sequential data, as well as the utilization of complementary data sources, such as well log and core data, to facilitate automatic feature learning and representation, thereby bolstering robustness and generalization capabilities. Experimental outcomes underscore the hybrid model's potential to achieve notable prediction accuracies, with R-squared values surpassing 0.93 on log data and 0.88 on core data sets, outperforming individual models. The model also exhibits commendable robustness and training efficiency, leveraging advanced methodologies such as ensemble techniques. In conclusion, this study underscores the promise of hybrid machine learning models as dependable tools for porosity prediction from core data. The insights gleaned from this research hold the potential to advance the understanding and optimization of porosity forecasting, thereby facilitating the formulation of more efficient reservoir management strategies. https://jesphys.ut.ac.ir/article_100681_ab974c8fe4ae4c997ff48af8bdf160e9.pdf