Optimizing seismic-based reservoir property prediction: a synthetic data-driven approach using convolutional neural networks and transfer learning with real data integration
Muhammad Ali,He Changxingyue,Ning Wei,Ren Jiang,Peimin Zhu,Zhang Hao,Wakeel Hussain,Umar Ashraf
DOI: https://doi.org/10.1007/s10462-024-11030-8
IF: 9.588
2024-12-01
Artificial Intelligence Review
Abstract:Reservoir characterization through seismic data analysis is essential for exploration and production in the petroleum industry. However, seismic-to-well tie discrepancies, limited availability of high-quality well data, and resolution constraints pose a reliability challenge. While previous studies offer valuable insights, they still struggle to achieve high-resolution predictions in a complex geologically environment given high reliance on well data. This study integrates synthetic data-driven techniques with real data, including convolutional neural networks (CNN) and transfer learning, to improve seismic reservoir characterization. We utilize nearby well statistics and a rock physics model (RPM) to simulate pseudo wells representing various geological scenarios. Synthetic seismic gathers are generated from these pseudo wells, which are based on RPM and local well control, to train the CNN. Transfer learning is then applied to adapt the CNN to better distinguish between real and synthetic data, enhancing reservoir predictions. A comparative analysis of P-impedance predictions from three methodologies: theory-driven Pre-Stack-Seismic-Inversion (TDSI), Deep-Neural-Network (DNN), and our CNN approach, showed that CNN achieved nearly 97% prediction accuracy with low error rates, compared to relatively lower prediction accuracy rates of DNN (86.2%) and TDSI (81.5%) with high error rates, according to robust metrics including R-square, RMSE, MSE, and MAE. These results indicate that CNN not only enhanced resolution but also closely aligned with well data and superior lateral continuity, even in blind well scenarios. This study effectively integrates synthetic data-driven techniques with CNNs and transfer learning to advance seismic reservoir property prediction, offering a robust solution to overcome limitations in traditional and DNN-based approaches.
computer science, artificial intelligence