Lithologic Reservoirs ›› 2026, Vol. 38 ›› Issue (5): 159-169.doi: 10.12108/yxyqc.20260515

• PETROLEUM EXPLORATION • Previous Articles     Next Articles

Intelligent seismic prediction method for porosity of coal rock gas reservoirs based on hybrid random forest: A case study of Carboniferous Benxi Formation 8# coal in northern Mizhi area, Ordos Basin

ZHAO Sen(), CHEN Sheng(), LI Xinyu, DU Wenhui, WANG Xiujiao, YANG Hao, LI Yandong, DAI Chunmeng   

  1. Research Institute of Petroleum Exploration & Development, Beijing 100083, China
  • Received:2026-03-12 Revised:2026-04-17 Online:2026-09-01 Published:2026-09-04
  • Contact: CHEN Sheng E-mail:1581808972@qq.com;cs69@petrochina.com.cn

RichHTML

12

PDF (PC)

9

Abstract:

To address the problem that conventional methods are insufficient for accurately characterizing high‑porosity reservoirs in Carboniferous Benxi Formation 8# coal in northern Mizhi area of Ordos Basin, a “classification‑regression” two‑stage collaborative hybrid random forest porosity prediction method was proposed by selecting sensitive seismic attributes based on Spearman rank correlation analysis, and constructing a low‑redundancy input feature set. It was then compared with six conventional methods. The results show that:（1） The hybrid random forest porosity prediction method uses a class‑balanced classifier to identify invalid va-lues and output their probabilities, introduces porosity‑segmented sample weights into the regressor training to incorporate weights of low‑, medium‑, and high‑porosity intervals into optimization, and adaptively adjusts the predictions through a probability‑weighted mechanism, which suppresses false high‑porosity anomalies in non‑reservoir sections while maintaining regression accuracy in reservoir sections. （2） On the blind well test set, determination coefficient（R2）of the hybrid random forest porosity predictions is 0.864 4, which significantly outperforms single‑attribute regression (0.720 4), multi‑attribute linear regression (0.766 5), XGBoost (0.717 5), conventional random forest (0.791 0), and SMOTE + random forest (0.771 3). In the low‑porosity background interval, the bias of the proposed method is only 0.416 4%, markedly lower than that of conventional random forest (0.829 8%) and XGBoost (0.881 6%), effectively suppressing false high‑porosity anomalies. In the sweet‑spot interval (porosity ≥ 6%), it also maintains a high prediction accuracy with an MAE of 0.703 8%. （3） Well‑tie profiles and planar prediction results demonstrate that the proposed method outperforms the compared methods in lateral reservoir continuity, sweet‑spot identification, and background control, and can accurately delineate the planar distribution characteristics of porosity of Benxi Formation 8# coal.

Key words: hybrid random forest algorithm, coal rock gas reservoir, porosity prediction, intelligent geophysical prospecting, 3D seismic, Benxi Formation 8# coal, northern Mizhi area, Ordos Basin

CLC Number: 

  • TE311

Fig. 1

Tectonic map of northern Mizhi area (a) and comprehensive stratigraphic column of Upper Carboniferous Benxi Formation to Lower Permian Shanxi Formation (b)， Ordos Basin"

Fig. 2

Schematic diagram of random forest model"

Fig. 3

Flowchart of hybrid random forest porosity prediction model"

Table 1

Seismic attribute names used in the hybrid random forest porosity prediction model"

属性
种类		 属性名称
基础
属性		 振幅包络(Amplitude Envelope)、瞬时相位(Instantaneous
Phase)、瞬时频率(Instantaneous Frequency)
导数
属性		 一阶导数(Derivative)、二阶导数(Second Derivative)
积分
属性		 导数瞬时振幅(Derivative Instantaneous Amplitude)、
积分(Integrate)
频率
属性		 积分绝对值(Integrate Absolute Amplitude)、主频(Dominant Frequency)、平均频率(Average Frequency)、振幅
加权频率(Amplitude Weighted Frequency)
相位
属性		 正交迹(Quadrature Trace)、余弦瞬时相位(Cosine Instantaneous Phase)、振幅加权相位(Amplitude Weighted Phase)、振幅加权余弦相位(Amplitude Weighted Cosine Phase)
极性
属性		 视极性(Apparent Polarity)
滤波
属性		 基础滤波(Filter)、5~10 Hz滤波(Filter 5-10 Hz)、15~
20 Hz滤波(Filter 15-20 Hz)、25~30 Hz滤波(Filter 25-30 Hz)、35~40 Hz滤波(Filter 35-40 Hz)、45~50 Hz滤波(Filter 45-50 Hz)、55~60 Hz滤波(Filter 55-60 Hz)
复合
属性		 二阶导数瞬时振幅(Second Derivative Instantaneous
Amplitude)

Fig. 4

Histogram of logging-derived porosity distribution of Carboniferous Benxi Formation 8# coal and sur-rounding rocks in northern Mizhi area, Ordos Basin"

Fig. 5

Statistics of average porosity of Carboniferous Benxi Formation 8# coal from 40 wells in northern Mizhi area, Ordos Basin"

Fig. 6

Heatmap of correlation between porosity and optimized feature parameters of Carboniferous Benxi Formation 8# coal in northern Mizhi area, Ordos Basin"

Fig. 7

Crossplot of porosity-optimized feature parameters of Carboniferous Benxi Formation 8# coal in northern Mizhi area, Ordos Basin"

Fig. 8

Well-tie profiles of optimized feature parameters of Carboniferous Benxi Formation 8# coal in northern Mizhi area, Ordos Basin"

Table 2

Hyperparameter settings in the hybrid random forest porosity prediction model of Carboniferous Benxi Formation 8# coal in northern Mizhi area, Ordos Basin"

超参数 参数搜索
范围		 分类器
最佳
参数		 回归器
最佳
参数
决策树数量 10~500 30 180
单棵决策树的最大深度 1~50、None 16 16
节点分裂所需最小样本数 1~10 2 2
叶节点所需最小样本数 1~10 1 1
单棵树在节点分裂时最大特征参数数量 None, sqrt, log2,1 sqrt 1

Table 3

Settings of porosity weights and segmentation thresholds in the hybrid random forest porosity prediction model of Carboniferous Benxi Formation 8# coal in northern Mizhi area, Ordos Basin"

超参数 搜索范围 最佳参数
低孔段权重 1~10 8
中孔段权重 1~10 2
高孔段权重 1~10 8
低孔、中孔分界/% 2~4 2
中孔、高孔分界/% 6~10 10

Fig. 9

Distribution of grid search scores in the hybrid random forest porosity prediction model of Carboniferous Benxi Formation 8# coal in northern Mizhi area, Ordos Basin"

Fig. 10

Comparison of feature parameter importance scores in the hybrid random forest porosity prediction model of Carboniferous Benxi Formation 8# coal in northern Mizhi area, Ordos Basin"

Fig. 11

Comparison of porosity prediction effects on blind wells using multiple methods of Carboniferous Benxi Formation 8# coal in northern Mizhi area, Ordos Basin"

Table 4

Comparison of porosity prediction errors using multiple methods of Carboniferous Benxi Formation 8# coal in northern Mizhi area, Ordos Basin"

方法 整体平均
绝对
误差/%		 非“甜点”
段平均
绝对误差/%		 非“甜点”
段平均
偏差/%		 “甜点”段平均绝对误差/% 整体
R2
单属性回归 1.243 6 1.343 4 0.703 3 0.732 8 0.720 4
多属性回归 1.252 3 1.364 7 0.670 1 0.677 1 0.766 5
XGBoost 1.177 5 1.251 2 0.881 6 0.800 4 0.717 5
随机森林 0.995 9 1.049 1 0.829 8 0.723 6 0.791 0
SMOTE_RF 1.021 3 1.073 7 0.859 9 0.753 4 0.771 3
混合随机森林 0.791 4 0.808 5 0.416 4 0.703 8 0.864 4

Fig. 12

Comparison of well-tie profile effects of porosity prediction using multiple methods of Carboniferous Benxi Formation 8# coal in northern Mizhi area, Ordos Basin"

Fig. 13

Planar distribution of porosity predicted by random forest (a) and hybrid random forest (b) of Carboniferous Benxi Formation 8# coal in northern Mizhi area, Ordos Basin"

[1] 李国欣, 张水昌, 何海清, 等. 煤岩气:概念、内涵与分类标准[J]. 石油勘探与开发, 2024, 51(4):783-795.
LI Guoxin, ZHANG Shuichang, HE Haiqing, et al. Coal-rock gas:Concept,connotation and classification criteria[J]. Petroleum Exploration and Development, 2024, 51(4):783-795.
[2] 邹才能, 赵群, 刘翰林, 等. 中国煤岩气突破及意义[J]. 天然气工业, 2025, 45(4):1-18.
ZOU Caineng, ZHAO Qun, LIU Hanlin, et al. China’s breakthrough in coal-rock gas and its significance[J]. Natural Gas Industry, 2025, 45(4):1-18.
[3] 桂志先, 朱广生, 熊新斌, 等. 孔隙度与纵横波速度间的关系及其在孔隙度预测中的应用[J]. 石油物探, 1998, 37(2):36-40.
GUI Zhixian, ZHU Guangsheng, XIONG Xinbin, et al. Relationship between porosity and P- and S-wave velocities and its application in porosity prediction[J]. Geophysical Prospecting for Petroleum, 1998, 37(2):36-40.
[4] HAMPSON D, SCHUELKE J S, QUIREIN J. Use of multiattribute transforms to predict log properties from seismic data[J]. Geophysics, 2001, 66(1):220-236.
[5] BERRYMAN J G, PRIDE S R, WANG H F. A differential scheme for elastic properties of rocks with dry or saturated cracks[J]. Geophysical Journal International, 2002, 151(2):597-611.
[6] HE Xile, HE Zhenhua, WANG Xuben, et al. Rock skeleton mo-dels and seismic porosity inversion[J]. Applied Geophysics, 2012, 9:349-358.
[7] 杨午阳, 魏新建, 李海山. 智能物探技术的过去、现在与未来[J]. 岩性油气藏, 2024, 36(2):170-188.
YANG Wuyang, WEI Xinjian, LI Haishan. The past,present and future of intelligent geophysical technology[J]. Lithologic Reservoirs, 2024, 36(2):170-188.
[8] CHEN Wei, YANG Liuping, ZHA Bei, et al. Deep learning re-servoir porosity prediction based on multilayer long short-term memory network[J]. Geophysics, 2020, 85(4):WA213-WA225.
[9] 路萍, 郭京哲, 高春云, 等. 基于机器学习的CO2封存致密砂岩储层孔隙度预测[J]. 地球物理学进展, 2024, 39(3):1129-1140.
LU Ping, GUO Jingzhe, GAO Chunyun, et al. Machine learning-based prediction of porosity in tight sandstone reservoirs for CO2 geological sequestration[J]. Progress in Geophysics, 2024, 39(3):1129-1140.
[10] 桂金咏, 李胜军, 高建虎, 等. 基于特征变量扩展的含气饱和度随机森林预测方法[J]. 岩性油气藏, 2024, 36(2):65-75.
GUI Jinyong, LI Shengjun, GAO Jianhu, et al. A random fo-rests prediction method for gas saturation based on feature variable extension[J]. Lithologic Reservoirs, 2024, 36(2):65-75.
[11] 甄艳, 康锦涛, 赵晓明, 等. 基于改进SMOTE和随机森林算法的致密砂岩成岩相测井解释方法[J]. 西南石油大学学报(自然科学版), 2025, 47(4):62-74.
ZHEN Yan, KANG Jintao, ZHAO Xiaoming, et al. A logging interpretation method for tight sandstone diagenetic facies based on improved SMOTE and random forest algorithm[J]. Journal of Southwest Petroleum University (Science & Technology Edition), 2025, 47(4):62-74.
[12] 翟咏荷, 何登发, 开百泽. 鄂尔多斯盆地及邻区中—晚二叠世构造-沉积环境与原型盆地演化[J]. 岩性油气藏, 2024, 36(1):32-44.
ZHAI Yonghe, HE Dengfa, KAI Baize. Tectonic-depositional environment and prototype basin evolution of Middle-Late Permian in Ordos Basin and adjacent areas[J]. Lithologic Re-servoirs, 2024, 36(1):32-44.
[13] 何登发, 包洪平, 开百泽, 等. 鄂尔多斯盆地及其邻区关键构造变革期次及其特征[J]. 石油学报, 2021, 42(10):1255-1269.
HE Dengfa, BAO Hongping, KAI Baize, et al. Critical tectonic modification periods and its geologic features of Ordos Basin and adjacent area[J]. Acta Petrolei Sinica, 2021, 42(10):1255-1269.
[14] 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 cha-racteristics 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.
[15] 张盟勃, 彭剑康, 崔晓杰, 等. 鄂尔多斯盆地米脂北地区石炭系本溪组煤岩气储层地震预测技术[J]. 岩性油气藏, 2026, 38(1):162-171.
ZHANG Mengbo, PENG Jiankang, CUI Xiaojie, et al. Seismic prediction technology for coal rock gas reservoir of Carbonife-rous Benxi Formation in northern Mizhi area,Ordos Basin[J]. Lithologic Reservoirs, 2026, 38(1):162-171.
[16] MALLEY J D, MALLEY K G, PAJEVIC S. Random forests-trees everywhere[M]// Statistical learning for biomedical data. Cambridge: Cambridge University Press,2011:137-154.
[17] BREIMAN L. Bagging predictors[J]. Machine Learning, 1996, 24:123-140.
[18] BREIMAN L, FRIEDMAN J H, OLSHEN R A, et al. Classification and regression trees[M]. Belmont:Wadsworth, 1984.
[19] 方匡南, 吴见彬, 朱建平, 等. 随机森林方法研究综述[J]. 统计与信息论坛, 2011, 26(3):32-38.
FANG Kuangnan, WU Jianbin, ZHU Jianping, et al. A review of technologies on random forests[J]. Journal of Statistics and Information, 2011, 26(3):32-38.
[20] 宋述芳, 何入洋. 基于随机森林的重要性测度指标体系[J]. 国防科技大学学报, 2021, 43(2):25-32.
SONG Shufang, HE Ruyang. Importance measure index system based on random forest[J]. Journal of National University of Defense Technology, 2021, 43(2):25-32.
[21] 王光宇, 宋建国, 徐飞, 等. 不平衡样本集随机森林岩性预测方法[J]. 石油地球物理勘探, 2021, 56(4):679-687.
WANG Guangyu, SONG Jianguo, XU Fei, et al. Random forest lithology prediction method for imbalanced data sets[J]. Oil Geophysical Prospecting, 2021, 56(4):679-687.
[22] SPEARMAN C. The proof and measurement of association between two things[J]. The American Journal of Psychology, 1904, 15(1):72-101.
[23] 马如辉, 张平. 利用波阻抗反演预测地层孔隙率[J]. 石油地球物理勘探, 2002, 37(5):537-540.
MA Ruhui, ZHANG Ping. Study of method for prediction of formation porosity by wave impedance inversion[J]. Oil Geophysical Prospecting, 2002, 37(5):537-540.
[24] 邹冠贵, 彭苏萍, 张辉, 等. 地震波阻抗反演预测采区孔隙度方法[J]. 煤炭学报, 2009, 34(11):1507-1511.
ZOU Guangui, PENG Suping, ZHANG Hui, et al. Predicting aquifer porosity of coal mine by impedance inversion[J]. Journal of China Coal Society, 2009, 34(11):1507-1511.
[25] 罗登贵, 刘江平, 金聪, 等. 活断层的地震响应特征与瞬时地震属性[J]. 地球科学, 2017, 42(3):462-470.
LUO Denggui, LIU Jiangping, JIN Cong, et al. Instantaneous seismic attributes and response characteristics of active faults[J]. Earth Science, 2017, 42(3):462-470.
[26] 王开燕, 徐清彦, 张桂芳, 等. 地震属性分析技术综述[J]. 地球物理学进展, 2013, 28(2):815-823.
WANG Kaiyan, XU Qingyan, ZHANG Guifang, et al. Summary of seismic attribute analysis[J]. Progress in Geophysics, 2013, 28(2):815-823.
[27] 张军华, 侯静, 辛星, 等. 道积分属性理论诠释及其在薄河道砂体预测中的应用[J]. 地球物理学进展, 2018, 33(1):326-333.
ZHANG Junhua, HOU Jing, XIN Xing, et al. Theory annotation and application of trace integration attribute in the prediction of thin channel sand body[J]. Progress in Geophysics, 2018, 33(1):326-333.
[28] 田文忠, 乔林, 袁剑, 等. 浅水三角洲前缘薄砂岩储层地震预测方法:以川西坳陷侏罗系蓬莱镇组二段为例[J]. 岩性油气藏, 2026, 38(1):78-88.
TIAN Wenzhong, QIAO Lin, YUAN Jian, et al. Seismic prediction method for thin sandstone reservoirs in shallow water delta front:A case study of the second member of Jurassic Penglaizhen Formation in Western Sichuan Depression[J]. Lithologic Reservoirs, 2026, 38(1):78-88.
[29] 邹冠贵, 彭苏萍, 郝小霞, 等. 基于楔形模型分析煤厚与地震振幅属性关系[J]. 煤炭科学技术, 2014, 42(4):88-91.
ZOU Guangui, PENG Suping, HAO Xiaoxia, et al. Analysis on relationship between seam thickness and seismic amplitude attribute based on wedge model[J]. Coal Science and Techno-logy, 2014, 42(4):88-91.
[30] WU Xiaoyang, CHAPMAN M, LI Xiangyang, et al. Quantitative gas saturation estimation by frequency-dependent amplitude-versus-offset analysis[J]. Geophysical Prospecting, 2014, 62(6):1224-1237.
[31] 李毓, 张春霞. 基于out-of-bag样本的随机森林算法的超参数估计[J]. 系统工程学报, 2011, 26(4):566-572.
LI Yu, ZHANG Chunxia. Estimation of the hyper-parameter in random forest based on out-of-bag sample[J]. Journal of Systems Engineering, 2011, 26(4):566-572.
[32] 殷疆, 焦雪君, 李小龙, 等. 基于随机森林优化算法的低电阻率储层含油饱和度评价方法[J]. 岩性油气藏, 2026, 38(1):55-66.
YIN Jiang, JIAO Xuejun, LI Xiaolong, et al. Evaluation method for oil saturation in low resistivity reservoirs based on random forest optimization algorithm[J]. Lithologic Reservoirs, 2026, 38(1):55-66.
[1] HUANG Li, ZHAO Longmei, CHEN Dong, ZHAO Haoyang, ZHANG Yixin, SHI Shi, ZHANG Wen, JI Liang. Characteristics and exploration potential of coal rock gas reservoirs in Carboniferous-Permian in Yichuan area, Ordos Basin [J]. Lithologic Reservoirs, 2026, 38(5): 40-50.
[2] ZHANG Weigang, LIU Guanglin, WANG Yong, NING Fan, MA Shaohui, YAN Baiquan. Reservoir characteristics and main controlling factors of interlayered shale oil in Triassic Chang 7 member of Tiebiancheng area， Ordos Basin [J]. Lithologic Reservoirs, 2026, 38(5): 94-103.
[3] FAN Yujie, WAN Yongping, PU Renhai, SHE Yinnan, CUI Hongjun, GAO Xiaoping. Sedimentary characteristics and exploration potential of the large-scale channel belt in Shan23 submember of Yan’an-Ganquan area, Ordos Basin [J]. Lithologic Reservoirs, 2026, 38(5): 125-135.
[4] LI Xinze, WU Hao, CHAO Jiahao, LI Na, WANG Suran, BAI Yuhu, FENG Ruyong, CHENG Jialu. Prediction method for coalbed methane deliverability of Permian Taiyuan Formation in Shenfu block, Ordos Basin [J]. Lithologic Reservoirs, 2026, 38(5): 191-200.
[5] BAI Yiyuan, GONG Hujun, WANG Dongxu, LI Yuan, LUO Fenhong, WANG Junjun. Distribution characteristics and controlling factors of helium in Yanchang exploration area of Ordos Basin [J]. Lithologic Reservoirs, 2026, 38(3): 149-161.
[6] REN Jiawei, LI Li, WANG Deyu, BAI Xiaohu, KANG Bo, BAI Yuen, BAI Jianwen, CHEN Junbin. Four-dimensional in-situ stress evolution and optimization method for refracturing timing in shale oil reservoir horizontal wells： A case study of Triassic Chang 7 member in Qingcheng Oilfield, Ordos Basin [J]. Lithologic Reservoirs, 2026, 38(3): 190-200.
[7] XUE Bowen, ZHANG Zhaohui, ZHANG Jiaosheng, ZOU Jiandong, ZHANG Wenting. Intelligent identification of fluid logging in tight sandstone reservoirs based on GWO-XGBoost model: A case study of Triassic Chang 8 member in Hongde area, Ordos Basin [J]. Lithologic Reservoirs, 2026, 38(2): 111-121.
[8] YANG Zhanlong, HAO Bin, TAN Kaijun, ZHANG Jing, ZHANG Liping, LIAO Jianbo, LI Zaiguang, SHI Jianglong. Geological features and hydrocarbon exploration directions of medium to small-scale basin group, onshore China [J]. Lithologic Reservoirs, 2026, 38(2): 12-31.
[9] ZHOU Wenjuan, PU Renhai, LU Zixing, WANG Kangle, ZHANG Peng, WEN Xingyu, WANG Tong, GUAN Yunwen. Distribution characteristics and controlling factors of evaporites in Member 5 of Ordovician Majiagou Formation, central and eastern Ordos Basin [J]. Lithologic Reservoirs, 2026, 38(2): 122-133.
[10] LI Tao, MA Guofu, ZHAO Leyi, YUAN Li, MA Qilin, XIE Jingyu, ZHANG Bo, LI Henan. Characteristics and oil-source correlation of Triassic Chang 7 member source rocks in Huanxian area, Ordos Basin [J]. Lithologic Reservoirs, 2026, 38(2): 134-144.
[11] LONG Liwen, XIAO Wenhua, YAN Baonian, WANG Jianguo, LI Shaoyong, LI Conglin, GUO Yaoxuan, REN Xueyao. Main controlling factors for hydrocarbon accumulation of Triassic Chang 81 reservoir in Huanxi area， Ordos Basin [J]. Lithologic Reservoirs, 2026, 38(2): 65-75.
[12] GUO Yuxin, BAI Yubin, ZHAO Jingzhou, ZHANG Jun, CAO Dandan. Heterogeneity and oil control effects of Triassic Chang 7 shale in Zhidan area, Ordos Basin [J]. Lithologic Reservoirs, 2026, 38(2): 97-110.
[13] ZHANG Mengbo, PENG Jiankang, CUI Xiaojie, ZHANG Dong, NI Na, LONG Shengfang, WEI Penghui. Seismic prediction technology for coal rock gas reservoir of Carboniferous Benxi Formation in northern Mizhi area, Ordos Basin [J]. Lithologic Reservoirs, 2026, 38(1): 162-171.
[14] XUN Xiaoquan, LI Hongtao, LI Changping, YANG Fan, LIU Xiong. A new method for splitting the production of fractured horizontal wells in strong heterogeneous gas reservoirs:Taking Permian He1 member gas reservoir in Dongsheng Gasfield of Ordos Basin as an example [J]. Lithologic Reservoirs, 2026, 38(1): 191-200.
[15] YIN Jiang, JIAO Xuejun, LI Xiaolong, LI Taifu, SHEN Zhanyong, LI Mengxi, SUN Rui, ZHU Yushuang. Evaluation method for oil saturation in low resistivity reservoirs based on random forest optimization algorithm [J]. Lithologic Reservoirs, 2026, 38(1): 55-66.
Viewed
Full text


Abstract

Cited
  Shared   
  Discussed   
No Suggested Reading articles found!
TRENDMD:

Copyright © 2018 Lithologic Reservoirs

Support by Beijing Magtech support@magtech.com.cn