1 Key Laboratory of Exploration Technologies for Oil and Gas Resources (Yangtze University), Ministry of Education, Wuhan 430100, China. 2 East China University of Technology, Nanchang 330013, China.
Accurate identification of lacustrine carbonate lithofacies is of critical importance for reservoir evaluation. However, conventional deep learning methods face significant bottlenecks in such complex settings, constrained by extreme class imbalance, the absence of uncertainty quantification in deterministic models, and insufficient geological prior knowledge. To address these limitations, this study proposes a Geologically Constrained Bayesian Prototypical Network (GC-BPN) that systematically overcomes these challenges through three synergistic mechanisms. First, the method introduces prototypical network metric learning, which effectively mitigates data scarcity and overfitting tendencies for extreme few-shot classes by learning class prototypes in an embedding space. Second, a Bayesian probabilistic inference framework is constructed to transform deterministic weights into probability distributions, enabling quantitative assessment of predictive uncertainty and providing reliable confidence indicators for lithofacies transition zones. Third, an enhanced Markov transition matrix is constructed by integrating mineral compositional continuity with label transition probabilities, and a Viterbi dynamic programming decoder is employed for global optimal sequence search, significantly suppressing lithofacies discontinuities that violate depositional principles. Systematic validation was conducted using 16,648 labeled samples from five cored wells in the Qianjiang Depression of the Jianghan Basin. Results demonstrate that the GC-BPN achieves an accuracy of 95.47% and a Macro-F1 score of 93.23% on the full-well validation set. The proposed architecture not only achieves breakthroughs in recognizing extreme few-shot classes (e.g., salt rock, granular mixed sedimentary rock) but also exhibits robust performance in blind-well cross-well generalization tests, establishing a new paradigm for intelligent well-log lithofacies identification in complex geological settings.
Download the full text PDF for viewing and using it according to the license of this paper.
Funding
This work was supported by the Open Fund of the SINOPEC Key Laboratory of Geophysics (Grant No. 36750000-24-FW0399-0002) and the Natural Science Foundation of Hubei Province, China (Grant No. 2024AFB271).
References
Abdel-Fattah, M. I., Mahdi, A. Q., Theyab, M. A., Pigott, J. D., Abd-Allah, Z. M., & Radwan, A. E. (2022). Lithofacies classification and sequence stratigraphic description as a guide for the prediction and distribution of carbonate reservoir quality: A case study of the Upper Cretaceous Khasib Formation (East Baghdad oilfield, central Iraq). Journal of Petroleum Science and Engineering, 209, 109835.
Aboubacar, M. S. I., Zhang, H., Ousmane, B. I., Li, J., & Cai, Z. (n.d.). An integrated approach for improved permeability and reservoir quality prediction in multiporosity systems, Tahe Ordovician naturally fractured vuggy carbonates.
Al-Mudhafar, W. J., Hasan, A. A., Abbas, M. A., & Wood, D. A. (2025). Machine learning with hyperparameter optimization applied in facies-supported permeability modeling in carbonate oil reservoirs. Scientific Reports, 15(1), 12939.
Bao, L.-L., Zhang, J.-S., Zhang, C.-X., Guo, R., Wei, X.-L., & Jiang, Z.-L. (2023). A reliable Bayesian neural network for the prediction of reservoir thickness with quantified uncertainty. Computers & Geosciences, 178, 105409.
Chen, J., Zhang, X., Chen, Z., Pang, X., Yang, H., Zhao, Z., Pang, B., & Ma, K. (2021). Hydrocarbon expulsion evaluation based on pyrolysis Rock-Eval data: Implications for Ordovician carbonates exploration in the Tabei Uplift, Tarim. Journal of Petroleum Science and Engineering, 196, 107614.
Dawson, H. L., Dubrule, O., & John, C. M. (2023). Impact of dataset size and convolutional neural network architecture on transfer learning for carbonate rock classification. Computers & Geosciences, 171, 105284.
De Medeiros, R. S. P., Basso, M., Chinelatto, G. F., Theodoro Soares, M. V., Matheus, G. F., Villacreses Morales, J. F., De Carvalho Mendes, L., & Vidal, A. C. (2024). Unravelling the origin of reworked deposits in Aptian lacustrine carbonate reservoirs of the Santos Basin, SE Brazil. Marine and Petroleum Geology, 161, 106700.
Ding, L., Chen, B., Zhu, Y., Dong, H., & Zhang, P. (2024). Mineral prediction based on prototype learning. Computers & Geosciences, 184, 105540.
Gomes, J. P., Bunevich, R. B., Tedeschi, L. R., Tucker, M. E., & Whitaker, F. F. (2020). Facies classification and patterns of lacustrine carbonate deposition of the Barra Velha Formation, Santos Basin, Brazilian Pre-salt. Marine and Petroleum Geology, 113, 104176.
Mohammadi, M., Niri, M. E., Bahroudi, A., Soleymanzadeh, A., & Kord, S. (2025). Enhancing formation resistivity factor estimation in carbonate reservoirs using electrical zone indicator and multi-resolution graph-based clustering methods. Scientific Reports, 15(1), 30823.
Mohammadian, E., Kheirollahi, M., Liu, B., Ostadhassan, M., & Sabet, M. (2022). A case study of petrophysical rock typing and permeability prediction using machine learning in a heterogenous carbonate reservoir in Iran. Scientific Reports, 12(1), 4505.
Montano, D., Gasparrini, M., Gerdes, A., Della Porta, G., & Albert, R. (2021). In-situ U-Pb dating of Ries Crater lacustrine carbonates (Miocene, South-West Germany): Implications for continental carbonate chronostratigraphy. Earth and Planetary Science Letters, 568, 117011.
Nawal, M., Kumar, S., & Shekar, B. (2022). LithoBot: An AutoML approach to identify lithofacies. Second International Meeting for Applied Geoscience & Energy, 1885-1889.
Qi, L., & Carr, T. R. (2006). Neural network prediction of carbonate lithofacies from well logs, Big Bow and Sand Arroyo Creek fields, Southwest Kansas. Computers & Geosciences, 32(7), 947-964.
Silva Dos Santos, V., Gloaguen, E., & Tirdad, S. (2025). Lithological mapping using Spatially Constrained Bayesian Network (SCB-Net): A deep learning model for generating field-data-constrained predictions with uncertainty evaluation using remote sensing data. Computers & Geosciences, 204, 105964.
Xin, C., et al. (2023). Integrated carbonate reservoir types modeling based on the PRT deep learning and multi-parameters seismic inversion and its application. Third International Meeting for Applied Geoscience & Energy Expanded Abstracts, 1239-1243.
Yang, Y.-Q., Qiu, L.-W., Gregg, J., Shi, Z., & Yu, K.-H. (2016). Formation of fine crystalline dolomites in lacustrine carbonates of the Eocene Sikou Depression, Bohai Bay Basin, East China. Petroleum Science, 13(4), 642-656.
Zheng, L., Wang, C., Jiang, Z., Wu, Y., Kong, X., Zhu, X., & Zhang, Y. (2024). Classification and reservoir characteristics of lacustrine carbonate conglomerate in the Shulu Sag of the Bohai Bay Basin. Marine and Petroleum Geology, 164, 106806.
SHOW FULL COLUMNS FROM `hl_config_system` [ RunTime:0.001742s ]
SELECT `value` FROM `hl_config_system` WHERE `type` = 'baseweb' AND `stype` = 'sys_view' AND `versions` = 0 LIMIT 1 [ RunTime:0.000507s ]
SELECT * FROM `hl_config_system` WHERE `type` = 'website' [ RunTime:0.000727s ]
SHOW FULL COLUMNS FROM `hl_foot_column` [ RunTime:0.002047s ]
SELECT `id`,`pid`,`title`,`url`,`type`,`link_to`,`link_way` FROM `hl_foot_column` WHERE `versions` = 0 AND `is_show` = 1 ORDER BY `sorted` DESC,`id` ASC [ RunTime:0.000647s ]
SELECT `value`,`stype` FROM `hl_config_system` WHERE `type` = 'baseweb' AND `versions` = 0 [ RunTime:0.000355s ]
SHOW FULL COLUMNS FROM `hl_journal_subject` [ RunTime:0.001320s ]
SELECT `id`,`subject_name` FROM `hl_journal_subject` WHERE `versions` = 0 ORDER BY `sort` DESC,`id` DESC [ RunTime:0.001025s ]
SHOW FULL COLUMNS FROM `hl_journal` [ RunTime:0.002874s ]
SELECT `jid`,`name` FROM `hl_journal` WHERE `versions` = 0 AND `isdelete` = 0 ORDER BY `is_recommend` DESC,`sort` DESC,`jid` DESC [ RunTime:0.001971s ]
SHOW FULL COLUMNS FROM `hl_paper` [ RunTime:0.003202s ]
SELECT * FROM `hl_paper` WHERE `id` = 3555 LIMIT 1 [ RunTime:0.002107s ]
UPDATE `hl_paper` SET `cnum` = `cnum` + 1 WHERE `id` = 3555 [ RunTime:0.005440s ]
SELECT `name` FROM `hl_journal` WHERE `jid` = 65 LIMIT 1 [ RunTime:0.000655s ]
SHOW FULL COLUMNS FROM `hl_journal_period` [ RunTime:0.002129s ]
SELECT `pid`,`name`,`cyear`,`imgurl` FROM `hl_journal_period` WHERE `isdelete` = 0 AND `id` = 477 LIMIT 1 [ RunTime:0.000400s ]
SELECT `pid`,`name`,`cyear`,`imgurl` FROM `hl_journal_period` WHERE `id` = 397 LIMIT 1 [ RunTime:0.000458s ]
SHOW FULL COLUMNS FROM `hl_paper_tag` [ RunTime:0.002111s ]
SELECT * FROM `hl_paper_tag` WHERE `id` IN (247,337) ORDER BY `sort` DESC,`id` ASC [ RunTime:0.001115s ]
SELECT * FROM `hl_journal` WHERE `jid` = 65 LIMIT 1 [ RunTime:0.000725s ]
SELECT `pid`,`name`,`type`,`cyear`,`id` FROM `hl_journal_period` WHERE `jid` = 65 AND `isdelete` = 0 ORDER BY `cyear` DESC,`type` ASC,`name` ASC,`ctime` ASC [ RunTime:0.000762s ]
SHOW FULL COLUMNS FROM `hl_journal_menu` [ RunTime:0.003247s ]
SELECT `id`,`column_title`,`column_content` FROM `hl_journal_menu` WHERE `jid` = 65 [ RunTime:0.002289s ]
SHOW FULL COLUMNS FROM `hl_paper_menu` [ RunTime:0.001980s ]
SELECT `column_title`,`column_content` FROM `hl_paper_menu` WHERE `pid` = 3555 [ RunTime:0.005049s ]
UPDATE `hl_journal` SET `sections_number` = `sections_number` + 1 WHERE `jid` = 65 [ RunTime:0.002139s ]
