Distribution prediction of the sepiolite-containing succession in the Member 1 of Maokou Formation in Sichuan Basin based on multi-gradient boosting algorithm
1 State Key Laboratory of Oil and Gas Reservoir Geology and Exploitation,Chengdu University of Technology,Chengdu 610059,China
2 Xihua University,Chengdu 610039,China
3 PetroChina Southwest Oil & Gas Field Company,Chengdu 610051,China
SONG Jinmin,born in 1983,Ph.D.,is a professor at Chengdu University of Technology. He is specialises in research and teaching within the field of petroleum reservoir geology. E-mail: songjinmin2012@cdut.edu.cn.
The sepiolite-containing succession in the Member 1 of the Middle Permian Maokou Formation in the Sichuan Basin exhibit self-generation and self-accumulation characteristics,positioning them as a promising new frontier for the exploration of unconventional gas reservoirs. However,the identification and predictive distribution of these strata are still underdeveloped. This study utilizes core samples,thin sections,X-ray diffraction(XRD)analysis,well logging,and mud logging data,performing sensitivity analysis to select six key well log curves: CNL,DEN,GR,RT,RXO,and AC. The SMOTE algorithm is employed to address feature imbalance. The workflow for predicting sepiolite-containing succession based on multi-gradient boosting algorithms is as follows: (1)The CatBoost algorithm is used for binary classification to determine the presence of sepiolite;(2)CatBoost performs multi-class classification to categorize the morphology of sepiolite-containing succession;(3)XGBoost is applied for regression analysis to predict the talc content in the sepiolite-containing succession;(4)The effective thickness of sepiolite-containing succession is identified based on talc content. The sepiolite-containing succession in the Member 1 of Maokou Formation primarily exhibit three morphologies: spotty,lenticular,and layered. Prediction results indicate that spotty talc is mainly developed in the northern and central southern regions of Sichuan Basin,with thickness increasing toward the northeast;lenticular talc is predominantly distributed in the western Sichuan region;and layered talc is mainly found in north western,central and southern Sichuan Basin. Overall,the sepiolite-containing succession in the Member 1 of Maokou Formation show a distribution pattern of‘thicker in the northeast,thinner in the southwest.’The Tongjiang-Changshou intracratonic sag area and the Hechuan-Weiyuan-Luzhou region serve as the sedimentary centers for the effective thickness of sepiolite-containing succession,providing a basis for future exploration and deployment in these areas.
[Author(id=1271896929056534705, tenantId=1045748351789510663, journalId=1155139928303341606, articleId=1242840881607729651, orderNo=0, firstName=null, middleName=null, lastName=null, nameCn=null, orcid=null, stid=null, country=null, authorPic=null, dead=0, email=songjinmin2012@cdut.edu.cn, emailSecond=null, emailThird=null, correspondingAuthor=0, authorType=1, ext={EN=AuthorExt(id=1271896929123643573, tenantId=1045748351789510663, journalId=1155139928303341606, articleId=1242840881607729651, authorId=1271896929056534705, language=EN, stringName=Jinmin SONG, firstName=Jinmin, middleName=null, lastName=SONG, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=1, address=1 State Key Laboratory of Oil and Gas Reservoir Geology and Exploitation,Chengdu University of Technology,Chengdu 610059,China, bio={"content":"
SONG Jinmin,born in 1983,Ph.D.,is a professor at Chengdu University of Technology. He is specialises in research and teaching within the field of petroleum reservoir geology. E-mail: songjinmin2012@cdut.edu.cn.
"}, bioImg=null, bioContent=
SONG Jinmin,born in 1983,Ph.D.,is a professor at Chengdu University of Technology. He is specialises in research and teaching within the field of petroleum reservoir geology. E-mail: songjinmin2012@cdut.edu.cn.
随着人工智能、机器学习及大数据运算技术的快速发展,油气勘探领域正在经历由经验驱动向数据驱动的转型。这些前沿技术能够在大规模、多维度的地质与测井资料中高效提取特征与规律为复杂地质体系的精细刻画和有利区预测提供了新的途径。Wang等(2024)以测井数据为基础,采用反向传播神经网络(BPNN)预测区块内裂缝发育度,建立了地应力控制下的裂缝发育分布模型。Koray等(2024)结合无监督聚类与多种监督回归算法改进机器学习流程,有效提升了孔隙度与渗透率建模精度。前人已将机器学习算法应用到茅一段的研究中,包括含滑石多矿物优化模型(殷树军等,2023)、BP神经网络算法(王心乾等,2024)、多元回归预测方法(蒋文博等,2018)、基于极限学习机算法的神经网络模型(白烨等,2021),主要是利用多维测井数据进行黏土矿物含量的预测。目前针对茅一段海泡石—滑石类黏土矿物产状的预测则相对欠缺,且海泡石(滑石)含量相对较低(刘树根等,2022),直接运用“端到端”预测模型可能出现偏差。因此需要先利用机器学习模型判定海泡石是否发育,然后增加海泡石层系类型的分类任务,目前这方面的研究尚显薄弱。作者将滑石视为海泡石演化的最终产物,利用梯度提升器算法(Friedman,1999)中的CatBoost(Prokhorenkova et al., 2018)及XGBoost(Chen and Guestrin,2016)模型重新审视“端到端”预测流程,以海泡石(滑石)作为预测分析载体对含海泡石层系存在性(二分类问题)、海泡石层系的滑石产状(多分类问题)和滑石含量(回归问题)进行综合分析,建立适用于含海泡石层系的综合预测的多梯度提升器算法工作流程,对四川盆地茅一段含海泡石层系不同滑石产状的分布进行预测,为后续的勘探提供参考。
梯度提升器算法(Grandient Boosting Machine,GBM)是Boosting(提升)算法的一种,是基于加法模型(additive model)和梯度下降的集成学习方法(Friedman,1999)。核心思想是串行地生成多个弱学习器,每个弱学习器的目标是拟合先前累加模型的损失函数的负梯度,使累积模型损失往负梯度的方向减少。通过不同的权重将机器学习器进行线性组合,使表现更优秀的学习器得到重用,有效提高预测精度,广泛应用于分类和回归问题。其代表性的算法包括CatBoost(Prokhorenkova et al., 2018)、XGBoost(Chen and Guestrin,2016)等,其在处理类别特征和提高准确性方面表现突出。
梯度提升器算法中,CatBoost使用“有序提升”(Ordered Boosting)和随机排列技术,直接处理类别特征,减少了目标泄漏(target leakage)问题,能够更好地捕捉数据中的复杂关系,适用于分类任务(Prokhorenkova et al., 2018)。XGBoost是一种高效的梯度提升算法,其分布式计算能力和优化大规模数据集能力能够高效处理高维数据并且能有效减少过拟合(Chen and Guestrin,2016)。因此,针对四川盆地含海泡石层系存在性二分类问题以及滑石产状多分类问题选取CatBoost模型,滑石含量回归预测任务选取XGBoost模型。
3.1 数据处理
收集四川盆地127口含海泡石层系数据,采用Z-score方法对每一道曲线数据进行标准化处理,标注出—含海泡石层系存在性—滑石产状分类标签。对于测井数据Xk及其对应标签Y,构成集合D, D={Xk,Yk}k=1, 2, …, n, 其中,Xk表示第k道测井曲线数据,对于实际测井曲线构成的特征数据集,Xk表示为:
数据预处理包含四川盆地待预测测井数据类型统计和完整性检验。在保证数据集和预测集数据特征一致的前提下,将测井曲线最低频率和完整性下限制设置为0.9,统计结果见表 1。引入随机森林参数打分法进一步检验与滑石发育的相关性(Hastie et al., 2009)(表 2),在确保测井曲线完整度前提下,综合结果选择自然伽马(GR)、声波时差(AC)、补偿中子(CNL)、密度(DEN)、地层真电阻率(RT)、冲洗带地层电阻率(RXO)曲线作为数据特征,采用MinMaxScaler方法(Pedregosa et al., 2011)进行数据归一化。
然而,含海泡石—滑石层系二分类问题和多分类问题中,存在滑石空间局部分布不均、二分类和多分类任务过程中的焦点差异导致样本标签分布不均衡,造成模型过采样/欠采样(Cao et al., 2019)引入SMOTE(Synthetic Minority Oversampling Technique)算法对少数样本进行补充(Chawla et al., 2002)。SMOTE通过在少数类样本的特征空间中构造合成样本来扩充数据集(公式4),共获得测井数据—滑石/非滑石、测井—滑石产状标签50800条(图 5)。
[FanJ P, SongJ M, JiangQ C, LiuS G, YeY H, HuangS P, WangJ R, SuW, LiL J, JinX, FengY X. 2022. Reservoir characteristics and development model of the Middle Permian Mao-1 Member in eastern Sichuan Basin. Oil & Gas Geology, 43(6): 1413-1430]
[FanJ P, SongJ M, LiuS G, JiangQ C, LiZ W, YangD, JinX, SuW, YeY H, HuangS P, WangJ R, JiangH, LuoP. 2023. Paleotemperature evolution and its driving mechanism during the formation of limestone-marl alternations in the first Member of Middle Permian Maokou Formation in Sichuan Basin. Petroleum Geology & Experiment, 45(4): 726-738]
[HuD F, WangL J, ZhangH R, DuanJ B, XiaW Q, LiuZ J, WeiQ C, WangK, PanL. 2020. Discovery of carbonate source rock gas reservoir and its petroleum geological implications: a case study of the gas reservoir in the first Member of Middle Permian Maokou Formation in the Fuling area,Sichuan Basin. Natural Gas Industry, 40(7): 23-33]
[JiangQ C, WangZ C, SuW, HuangS P, ZengF Y, FengZ, BiH, LuW H. 2021. Accumulation conditions and favorable exploration orientation of unconventional natural gas in the marl source rock of the first Member of the Middle Permian Maokou Formation,Sichuan Basin. China Petroleum Exploration, 26(6): 82-97]
[JiangW B, DingX J, GuoH N, XuS S, HuangB, LiX. 2018. Establishment of prediction equations for clay minerals and their application in the Dongying Oilfield. Petroleum Geology and Engineering, 32(3): 95-97,125 ]
[LiangX, XuZ Y, LiW M, MaL Q, JiY B, LuoY F, DingB C. 2023. Characteristics and main controlling factors of the limy source rock gas reservoir in the first Member of the Middle Permian Maokou Formation in the southern Sichuan and western Chongqing area: a case study of well DB 1. Petroleum Geology & Experiment, 45(4): 714-725]
[LiangX, ZhangL, JiangL W, ShanC A, PanF, LuoY F, MengY, WangG C, ShuH L, ZhuD X, LiB S, WangZ D, YaoQ C, XuZ Y. 2024. Exploration breakthrough and resource potential of limy source rock gas in Da’an area in the western Chongqing,Sichuan Basin. China Petroleum Exploration, 29(2): 30-45]
[LiuS, FanC H, ZhangB J, ZhangY, WangW, LuoB, BaiX L. 2023. Distribution characteristics of Gufeng Member of the Middle Permian Maokou Formation,eastern Sichuan Basin and its petrogeological significance. Oil & Gas Geology, 44(4): 993-1008]
[SongJ M, LiuS G, JinX, WangH, FanJ P, JiangQ C, TianX B, YeY H, LiY, LiZ W, YangD, WangJ R, LuoP. 2022. A new model of source-reservoir association and new fields of oil and gas exploration in the Middle Permian sepiolite-bearing strata,Sichuan Basin,China. Journal of Chengdu University of Technology(Science & Technology Edition), 49(2): 129-149]
[SongJ M, LiuS G, LiZ W, ZhangB J, DengB, JiangQ C, JinX, SunW, WangH, GuoH Y, ChenX, WangJ R, TianX B, ZhangX H, YeY H, YangD, FanJ P, LuoP. 2023. Accumulation model and favorable exploration area distribution of the Middle Permian oil and gas in the Sichuan Basin. Natural Gas Industry, 43(11): 54-71]
[SongJ M, WangJ K, LiuS G, WenL, YeY H, LuoB, LiZ W, ZhangB J, JinX, YangD, ZhangX H, WangJ R, ZhouG, GuoJ X, ZhangZ Y, LuoP. 2025. Occurrence types and enrichment model of organic matter in the sepiolite-containing successions: a case study of the Mao-1 Member of Mid-Permian Maokou Formation,Eastern Sichuan Basin,SW China. Petroleum Exploration and Development, 52(5):1054-1067]
[WangB, SongJ M, LiuS G, JinX, LiZ W, YeY H, YangD, WangJ R, WangJ K, DengH S, GuoJ X, LuoP. 2024. Types and genetic modes of dolomite in the Member 2 of Middle Permian Maokou Formation in western Sichuan Basin. Acta Petrolei Sinica, 45(12): 1743-1760]
[WangX Q, YuW D, MaX D, ZhouT, TaiH, CuiQ Y, DengK, LuY C, LiuZ H. 2024. Identification and application of shale lithofacies based on conventional logging curves: a case study of the second Member of Funing Formation in Qintong sag,Subei Basin. Petroleum Reservoir Evaluation and Development, 14(5): 699-706,713 ]
[XiaW Q, LiuJ, YuL J, ZhangH R, WangJ H, WeiF B, ZhangS Y. 2023. Characteristics and main controlling factors of argillaceous limestone reservoir in the first Member of Permian Maokou Formation in Fuling area,southeast Sichuan Basin. Mineralogy and Petrology, 43(4): 100-108]
[YangY M, YangY, WenL, ZhangX H, ChenC, ChenK, ZhangY, DiG D, WangH, XieC. 2020. New exploration progress and prospect of Middle Permian natural gas in the Sichuan Basin. Natural Gas Industry, 40(7): 10-22]
[YinS J, WenZ, JinX Y, YanW L, WangC Y, ShiJ B, XiaW H, LiJ F, NieN, ZhengS D. 2023. Characteristics and logging characterization method of unconventional marl reservoir: a case study of the Member 1 of Maokou Formation in Hechuan-Tongnan area,Sichuan Basin. Acta Petrolei Sinica, 44(7): 1105-1117]
[ZhaiC B, LinL B, YouD H, LiuF B, LiuS Y. 2024. Sedimentary microfacies characteristics and organic matter enrichment pattern of the first Member of the Middle Permian Maokou Formation,southwestern Sichuan Basin. Oil & Gas Geology, 45(2): 440-456]
[ZhangP X, HeX P, GaoQ F, GaoY Q, SunB, CaiX, HeG S, ZhangZ P, LiuN N. 2021. Geological characteristics and enrichment pattern of Permian Mao-1 Member shale gas reservoirs at the southeastern margin of Sichuan Basin. Oil & Gas Geology, 42(1): 146-157]
[ZhangY, CaoQ G, LuoK P, LiL L, LiuJ L. 2022. Reservoir exploration of the Permian Maokou Formation in the Sichuan Basin and enlightenment obtained. Oil & Gas Geology, 43(3): 610-620]
[49]
CaoK, WeiC, GaidonA, ArechigaN,Ma, TY. 2019. Learning imbalanced datasets with label-distribution-aware margin loss. Advances in Neural Information Processing Systems, 32(NeurIPS 2019).
[50]
ChawlaN V, BowterK W, HallL O, KegelmeyerW P. 2002. SMOTE: synthetic minority over-sampling technique. Journal of Artificial Intelligence Research, 16: 321-357.
[51]
ChenT, GuestrinC. 2016. Xgboost: a scalable tree boosting system. Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining: 785-794.
[52]
DorogushA V, ErshovV, GulinA. 2018. The CatBoost algorithm. Microsoft Research Technical Report.
[53]
FriedmanJ H. 1999. Greedy function approximation: a gradient boosting machine. Annals of Statistics, 29(5): 1189-1232.
[54]
HastieT, TibshiraniR, FriedmanJ. 2009. The Elements of Statistical Learning(2nd ed). New York: Springer.
[55]
HyndmanR J, KoehlerA B. 2006. Another look at measures of forecast accuracy. International Journal of Forecasting, 22(4): 679-688.
[56]
JinH, SongQ, HuX. 2019. Auto-Keras: an efficient neural architecture search system. 25th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining: 1946-1956.
[57]
KorayA-M, BuiD, KubiE, AmpomahW, AmosuA. 2024. Machine learning based reservoir characterization and numerical modeling from integrated well log and core data. Geoenergy Science & Engineering,243: 213296.
ProkhorenkovaL, GusevG, VorobevA, DorogushA V, GulinA. 2018. CatBoost: unbiased boosting with categorical features. Advances in Neural Information Processing Systems,31.
[60]
WangZ H, CaiY D, LiuD M, LuJ, QiuF, SunF R, HuJ H, LiZ T. 2024. Characterization of natural fracture development in coal reservoirs using logging machine learning inversion,well test data and simulated geostress analyses. Engineering Geology: 107696.