Intelligent identification of deep coal macrolithotypes based on weighted ensemble in machine learning: A case study of the Shenmu area, Ordos Basin

Authors

• YAO Yuhao, Key Laboratory of Coalbed Methane Resources and Accumulation Process, Ministry of Education, China University of Mining and Technology, Xuzhou 221008, ChinaFollow
• YANG Zhaobiao, Key Laboratory of Coalbed Methane Resources and Accumulation Process, Ministry of Education, China University of Mining and Technology, Xuzhou 221008, ChinaFollow
• HUANG Daojun, PetroChina Changqing Oilfield Company Exploration and Development Research Institute, Xi’an 710018, China
• WANG Jianan, Key Laboratory of Coalbed Methane Resources and Accumulation Process, Ministry of Education, China University of Mining and Technology, Xuzhou 221008, China
• DONG Guodong, PetroChina Changqing Oilfield Company Exploration and Development Research Institute, Xi’an 710018, China
• CHEN Yilin, Key Laboratory of Coalbed Methane Resources and Accumulation Process, Ministry of Education, China University of Mining and Technology, Xuzhou 221008, China
• HE Jinxian, Key Laboratory of Coalbed Methane Resources and Accumulation Process, Ministry of Education, China University of Mining and Technology, Xuzhou 221008, China
• ZHANG Shang, Key Laboratory of Coalbed Methane Resources and Accumulation Process, Ministry of Education, China University of Mining and Technology, Xuzhou 221008, China

Abstract

Objective Coal macrolithotype represents a key geological property in the exploration and recovery of deep coalbed methane (CBM) resources, and its accurate identification is of great significance for selecting the optimal geological and engineering sweet-spot intervals. However, conventional coal macrolithotype identification relies heavily on core observation, suffering from prohibitively high costs since core data are extremely valuable and thus feature a small sample size. Meanwhile, conventional log-based threshold methods face challenges in effectively characterizing the complex boundaries between coal macrolithotypes.Methods This study investigated the No. 8 coal seam of the Benxi Formation, Shenmu area, eastern Ordos Basin. Based on the systematic core sample observation and log data analysis of 25 cored wells, this study develops a weighted soft voting ensemble strategy that combines multiple prediction models. Specifically, following the depth-thickness joint corrections of log data, along with the selection of the optimal synthetic minority over-sampling technique and Tomek Links (SMOTE-Tomek) class balance technique for hybrid resampling, multiple machine learning models are systematically compared, including random forest (RF), XGBoost, and Light Gradient Boosting Machine (LightGBM). Models with high performance are then selected through 5-fold cross-validation. Subsequently, a soft voting ensemble model is constructed using four weighted soft-voting strategies: equal, cross-validated accuracy, Macro-F1, and G-mean weightings. The ensemble model integrates the discriminative information from heterogeneous models, thus improving the stability in identifying transitional boundaries between coal macrolithotypes. Results and Conclusions Validation on an independent test set indicates that the weighted ensemble model achieved an overall accuracy of 84.91%, representing an increase of 1.89% compared to the RF model—the optimal single model. Leave-one-well-out cross-validation (LOOCV)-based blind well verification reveals that the ensemble model yielded an average accuracy of 83.56%, highlighting its potential for cross-well applications. This model was used to predict the coal macrolithotypes in the No. 8 coal seam of the Benxi Formation within the Shenmu area. The prediction results show that bright to semi-bright coals are primarily distributed in the south-central part, and semi-dull to dull coals are concentrated in the northwestern part, forming a distinct spatial differentiation framework. These findings provide a basis for selecting the optimal deep CBM sweet spots. The novel log-based method for identifying coal macrolithotypes developed in this study can provide a scientific basis and technical support for fine-scale CBM reservoir evaluation and well placement optimization in the Shenmu area and blocks with similar geological settings.

Keywords

deep coalbed methane (CBM), coal macrolithotype, log curve, machine learning, Ordos Basin

DOI

10.12363/issn.1001-1986.26.06.0353

Recommended Citation

YAO Yuhao, YANG Zhaobiao, HUANG Daojun, et al. (2026) "Intelligent identification of deep coal macrolithotypes based on weighted ensemble in machine learning: A case study of the Shenmu area, Ordos Basin," Coal Geology & Exploration: Vol. 54: Iss. 9, Article 15.
DOI: 10.12363/issn.1001-1986.26.06.0353
Available at: https://cge.researchcommons.org/journal/vol54/iss9/15

Reference

[1] 鞠玮,陶树,杨兆彪,等. 中国深部煤层气研究与勘探开发现状及其发展趋势[J]. 石油实验地质,2025,47(1):9−16. JU Wei,TAO Shu,YANG Zhaobiao,et al. Current status and development trends of deep coalbed methane research in China[J]. Petroleum Geology & Experiment,2025,47(1):9−16.

[2] ZHOU Keyu,SUN Fengrui,YANG Chao,et al. Evaluation of deep coalbed methane potential and prediction of favorable areas within the Yulin area,Ordos Basin,based on a multi–level fuzzy comprehensive evaluation method[J]. Processes,2024,12(4):820.

[3] 程慧杰. 沁水盆地郑庄某井区3#煤储层非均质性精细表征及剩余气分布[D]. 徐州: 中国矿业大学, 2022. CHENG Huijie. Fine characterization of heterogeneity and residual gas distribution of 3# coal reservoir in a well area of Zhengzhuang,Qinshui Basin[D]. Xuzhou: China University of Mining and Technology, 2022.

[4] 王金伟,许浩,刘一楠,等. 鄂尔多斯盆地临兴–神府地区深部煤储层储渗空间发育特征及产水能力评价[J]. 石油实验地质,2025,47(1):54−63. WANG Jinwei,XU Hao,LIU Yinan,et al. Storage and permeation space development characteristics and water production capacity evaluation of deep coal reservoirs in Linxing–Shenfu area of Ordos Basin[J]. Petroleum Geology & Experiment,2025,47(1):54−63.

[5] 张群,潘治贵. 烟煤的宏观煤岩分类系统研究[J]. 煤田地质与勘探,1999,27(2):2−5. ZHANG Qun,PAN Zhigui. Establishment of macroscopic classification system for bituminous coal in China[J]. Coal Geology & Exploration,1999,27(2):2−5.

[6] 邵龙义,董大啸,李明培,等. 华北石炭–二叠纪层序–古地理及聚煤规律[J]. 煤炭学报,2014,39(8):1725−1734. SHAO Longyi,DONG Daxiao,LI Mingpei,et al. Sequence–paleogeography and coal accumulation of the Carboniferous–Permian in the North China Basin[J]. Journal of China Coal Society,2014,39(8):1725−1734.

[7] 代世峰,任德贻,李生盛,等. 内蒙古准格尔黑岱沟主采煤层的煤相演替特征[J]. 中国科学(D辑：地球科学),2007,37(增刊1):119−126.

[8] 赵志刚,朱学申,王存武,等. 基于资源性与可压性的深部煤层气"甜点"预测[J]. 煤田地质与勘探,2024,52(8):22−31. ZHAO Zhigang,ZHU Xueshen,WANG Cunwu,et al. Predicting the "sweet spot" of deep coalbed methane based on resource conditions and fracability[J]. Coal Geology & Exploration,2024,52(8):22−31.

[9] 崔超. 宏观煤岩类型的测井识别与量化评价:以郑庄区块为例[D]. 北京: 中国地质大学(北京), 2021. CUI Chao. Logging identification and quantitative evaluation of coal macrolithotypes: A case of Zhengzhuang field[D]. Beijing: China University of Geosciences (Beijing), 2021.

[10] ZHAO Junlong,TANG Dazhen,QIN Yong,et al. Experimental study on structural models of coal macrolithotypes and its well logging responses in the Hancheng area,Ordos Basin,China[J]. Journal of Petroleum Science and Engineering,2018,166:658−672.

[11] XU Hao,TANG Dazhen,MATHEWS J P,et al. Evaluation of coal macrolithotypes distribution by geophysical logging data in the Hancheng Block,eastern margin,Ordos Basin,China[J]. International Journal of Coal Geology,2016,165:265−277.

[12] GORDON J B,SANEI H,PEDERSEN P K. Predicting hydrogen and oxygen indices (HI,OI) from conventional well logs using a Random Forest machine learning algorithm[J]. International Journal of Coal Geology,2022,249:103903.

[13] 邵先杰,董新秀,汤达祯,等. 韩城矿区煤岩类型测井解释技术及产能预测方法[J]. 测井技术,2013,37(6):671−675. SHAO Xianjie,DONG Xinxiu,TANG Dazhen,et al. Coal type definition and productivity production of Hancheng mining area based on logging data[J]. Well Logging Technology,2013,37(6):671−675.

[14] 蔡天,孙建孟,孙红华,等. 基于机器学习的元素录井宏观煤岩类型识别方法研究[J]. 测井技术,2026,50(1):97−107. CAI Tian,SUN Jianmeng,SUN Honghua,et al. Research on the macroscopic coal rock type identification method of elemental logging based on machine learning[J]. Well Logging Technology,2026,50(1):97−107.

[15] 石玉江,何羽飞,万金彬,等. 深层煤岩气地质品质及含气量测井评价方法研究[J]. 中国石油勘探,2024,29(4):126−141. SHI Yujiang,HE Yufei,WAN Jinbin,et al. Research on logging evaluation methods for geological quality and gas content of deep coal measure gas[J]. China Petroleum Exploration,2024,29(4):126−141.

[16] 高飞,曲志鹏,魏震,等. 基于机器学习方法的测井岩相分类研究[J]. 地球物理学进展,2024,39(3):1173−1192. GAO Fei,QU Zhipeng,WEI Zhen,et al. Study on well–log lithofacies classification based on machine learning methods[J]. Progress in Geophysics,2024,39(3):1173−1192.

[17] 王民,杨金路,王鑫,等. 基于随机森林算法的泥页岩岩相测井识别[J]. 地球科学,2023,48(1):130−142. WANG Min,YANG Jinlu,WANG Xin,et al. Identification of shale lithofacies by well logs based on random forest algorithm[J]. Earth Science,2023,48(1):130−142.

[18] 赵石虎,于腾腾,何云超,等. 基于测井的柿庄地区宏观煤岩类型划分[J]. 中国矿业,2017,26(增刊1):382−384. ZHAO Shihu,YU Tengteng,HE Yunchao,et al. Classification of macroscopic type based on logging in Shizhuang area[J]. China Mining Magazine,2017,26(Sup.1):382−384.

[19] 张云骥,李倩,蔡益栋,等. 基于机器学习测井反演的宏观煤岩类型评价:以鄂尔多斯盆地佳县区块本溪组8号煤为例[J]. 现代地质,2025,39(5):1427−1436. ZHANG Yunji,LI Qian,CAI Yidong,et al. Evaluation of coal macrolithotypes based on machine learning logging inversion: A case study of the No. 8 coal seam in the Benxi Formation,Jiaxian Block,Ordos Basin[J]. Geoscience,2025,39(5):1427−1436.

[20] 范立勇,周国晓,杨兆彪,等. 鄂尔多斯盆地深部煤层气差异富集的地质控制[J]. 煤炭科学技术,2025,53(1):203−215. FAN Liyong,ZHOU Guoxiao,YANG Zhaobiao,et al. Geological control of differential enrichment of deep coalbed methane in the Ordos Basin[J]. Coal Science and Technology,2025,53(1):203−215.

[21] 杨兆彪,卢本举,周国晓,等. 深部煤层气地质特殊性与释放产出规律[J]. 石油学报,2025,46(8):1464−1476. YANG Zhaobiao,LU Benju,ZHOU Guoxiao,et al. Geological particularities and desorption–production patterns of deep coalbed methane[J]. Acta Petrolei Sinica,2025,46(8):1464−1476.

[22] 郭晓娇,王雷,姚仙洲,等. 深部煤岩地质特征及煤层气富集主控地质因素:以鄂尔多斯盆地东部M区为例[J]. 石油实验地质,2025,47(1):17−26. GUO Xiaojiao,WANG Lei,YAO Xianzhou,et al. Geological characteristics of deep coal rock and main geological factors controlling coalbed methane enrichment: A case study of the M area in the eastern Ordos Basin[J]. Petroleum Geology & Experiment,2025,47(1):17−26.

[23] 牛小兵,喻健,徐旺林,等. 鄂尔多斯盆地上古生界煤岩气成藏地质条件及勘探方向[J]. 天然气工业,2024,44(10):33−50. NIU Xiaobing,YU Jian,XU Wanglin,et al. Reservoir–forming geological conditions and exploration directions of Upper Paleozoic coal–rock gas in the Ordos Basin[J]. Natural Gas Industry,2024,44(10):33−50.

[24] 赵喆,徐旺林,赵振宇,等. 鄂尔多斯盆地石炭系本溪组煤岩气地质特征与勘探突破[J]. 石油勘探与开发,2024,51(2):234−247. ZHAO Zhe,XU Wanglin,ZHAO Zhenyu,et al. Geological characteristics and exploration breakthroughs of coal rock gas in Carboniferous Benxi Formation,Ordos Basin,NW China[J]. Petroleum Exploration and Development,2024,51(2):234−247.

[25] 魏虎,林进,李涛,等. 鄂尔多斯盆地东南缘深部煤储层有利区测井评价[J]. 现代地质,2026,40(1):154−168. WEI Hu,LIN Jin,LI Tao,et al. Logging evaluation of favorable areas for deep coalbed methane reservoirs in the southeastern margin of the Ordos Basin[J]. Geoscience,2026,40(1):154−168.

[26] 周立宏,李曙光,王渊,等. 深层煤岩气效益开发关键技术与实践:以鄂尔多斯盆地大宁–吉县区块为例[J]. 石油学报,2025,46(8):1536−1549. ZHOU Lihong,LI Shuguang,WANG Yuan,et al. Key technologies and practices for cost–effective development of deep coal–rock gas: A case study of the Daning–Jixian Block in Ordos Basin[J]. Acta Petrolei Sinica,2025,46(8):1536−1549.

[27] 余成林,马立民,于洋洋,等. 煤岩气地质工程一体化高效开发技术实践:以鄂尔多斯盆地佳县区块为例[J]. 油气藏评价与开发,2026,16(5):963−973. YU Chenglin,MA Limin,YU Yangyang,et al. Practice of integrated geology–engineering efficient development technology of coal–rock gas: A case study of the Jiaxian Block in Ordos Basin[J]. Petroleum Reservoir Evaluation and Development,2026,16(5):963−973.

[28] MAATEN L V D,HINTON G. Visualizing data using t–SNE[J]. Journal of Machine Learning Research,2008,9(86):2579−2605.

[29] 秦瑞宝,叶建平,李利,等. 基于机器学习的煤层含气量测井评价方法:以沁水盆地柿庄南区块为例[J]. 石油物探,2023,62(1):68−79. QIN Ruibao,YE Jianping,LI Li,et al. Artificial–intelligence and machine–learning models of coalbed methane content based on geophysical logging data: A case study in Shizhuang South Block of Qinshui Basin,China[J]. Geophysical Prospecting for Petroleum,2023,62(1):68−79.

[30] LI Cunlei,YANG Zhaobiao,TIAN Wenguang,et al. Construction and application of prediction methods for coal texture of CBM reservoirs at the block scale[J]. Journal of Petroleum Science and Engineering,2022,219:111075.

[31] LIU Weikai,ZHAO Yanbin,YANG Mei,et al. XGBoost formation thickness identification based on logging data[J]. Frontiers in Earth Science,2022,10:918384.

[32] DEV V A,EDEN M R. Formation lithology classification using scalable gradient boosted decision trees[J]. Computers & Chemical Engineering,2019,128:392−404.

[33] AHMADI A,SHARIF S S,BANAD Y M. A comparative study of sampling methods with cross–validation in the FedHome framework[J]. IEEE Transactions on Parallel and Distributed Systems,2025,36(3):570−579.

[34] CHAWLA N V,BOWYER K W,HALL L O,et al. SMOTE: Synthetic minority over–sampling technique[J]. Journal of Artificial Intelligence Research,2002,16(1):321−357.