Advanced Search
Volume 42 Issue 3
Jun.  2024
Turn off MathJax
Article Contents

YE Yu, CHENG Chao, JIANG YuQiang, YI JuanZi, DENG HongBing, LI Xi, GU YiFan, CHEN Yan. Lithology and Distribution Characteristics of the Oolitic Beach from the Feixianguan Formation, Eastern Sichuan Basin[J]. Acta Sedimentologica Sinica, 2024, 42(3): 1032-1046. doi: 10.14027/j.issn.1000-0550.2022.089
Citation: YE Yu, CHENG Chao, JIANG YuQiang, YI JuanZi, DENG HongBing, LI Xi, GU YiFan, CHEN Yan. Lithology and Distribution Characteristics of the Oolitic Beach from the Feixianguan Formation, Eastern Sichuan Basin[J]. Acta Sedimentologica Sinica, 2024, 42(3): 1032-1046. doi: 10.14027/j.issn.1000-0550.2022.089

Lithology and Distribution Characteristics of the Oolitic Beach from the Feixianguan Formation, Eastern Sichuan Basin

doi: 10.14027/j.issn.1000-0550.2022.089
Funds:

National Natural Science Foundation of China 41430316

National Science and Technology Major Project 2017ZX05008-004-008

  • Received Date: 2022-04-27
  • Accepted Date: 2022-09-16
  • Rev Recd Date: 2022-07-06
  • Available Online: 2022-09-16
  • Publish Date: 2024-06-10
  • Objective To solve the problems such as unclear lithological variations of the Feixianguan Formation on the western side of the marine trough's southern segment and within the plateau interior of the eastern part of the Sichuan Basin, Methods the lithological types and characteristics of the Feixianguan Formation were studied by comprehensively using multiple geological data such as cores, drilling and logging, and an intelligent identification method of lithological logging based on machine learning was proposed, which solved the technical problems of fine identification of lithology in the old area, and revealed the lithology, distribution and evolution of the Feixianguan Formation in the area. Results and Conclusions (1) The Feixianguan Formation is mainly composed of lithology such as mudstone, mud crystal limestone, argillary limestone, granule limestone, granulous dolomite, mud crystal dolomite, paste dolomite, and gypsum rock; (2) Comparison found that the improved gradient-boosted decision tree algorithm, namely Stochastic Gradient Boosting Decision Tree (SGBDT), is superior to other algorithms for constructing lithology models, and is more suitable for carbonate rock complex lithology identification; (3) The granule limestone developed intensively in the area south of the Kaijiang-Liangping Sea Trough between Feixianguan I period and Feixianguan Ⅲ period, and the granulous dolomite was concentrated in the Feixianguan Ⅱ period and distributed and scattered; (4) The distribution of the oolitic shoal in the area exhibits significant variation. The main development of the plateau ancient geomorphological high point and the edge of the platform in the Feixianguan I period. In the Feixianguan Ⅱ period, most of the the oolitic shoal at the edge of the platform were developed, and a small number of ancient landforms in the platform were developed.And the main development of the plateau ancient geomorphological high point in the Feixianguan Ⅲ period.
  • [1] 戴金星,倪云燕,刘全有,等. 四川超级气盆地[J]. 石油勘探与开发,2021,48(6):1081-1088.

    Dai Jinxing, Ni Yunyan, Liu Quanyou, et al. Sichuan super gas basin in southwest China[J]. Petroleum Exploration and Development, 2021, 48(6): 1081-1088.
    [2] 马新华,杨雨,文龙,等. 四川盆地海相碳酸盐岩大中型气田分布规律及勘探方向[J]. 石油勘探与开发,2019,46(1):1-13.

    Ma Xinhua, Yang Yu, Wen Long, et al. Distribution and exploration direction of medium-and large-sized marine carbonate gas fields in Sichuan Basin, SW China[J]. Petroleum Exploration and Development, 2019, 46(1): 1-13.
    [3] 刘景东,刘光祥,韦庆亮,等. 四川盆地元坝地区飞仙关组二段滩相储层孔隙演化特征[J]. 中国石油大学学报(自然科学版),2016,40(1):10-17.

    Liu Jingdong, Liu Guangxiang, Wei Qingliang, et al. Pore evolution characteristics of beach facies reservoir of Feixianguan Ⅱ member in Yuanba area, Sichuan Basin[J]. Journal of China University of Petroleum (Edition of Natural Science), 2016, 40(1): 10-17.
    [4] 陈辉,郭海洋,徐祥恺,等. 四川盆地剑阁—九龙山地区长兴期与飞仙关期古地貌演化特征及其对礁滩体的控制[J]. 石油与天然气地质,2016,37(6):854-861.

    Chen Hui, Guo Haiyang, Xu Xiangkai, et al. Features of paleogeomorphological evolution and its control on reef flat composite in Changxing-Feixianguan Formation in Jiange-Jiulongshan region, the Sichuan Basin[J]. Oil & Gas Ge-ology, 2016, 37(6): 854-861.
    [5] 蒋裕强,邓虹兵,易娟子,等. 开江—梁平海槽西侧飞仙关组不同类型鲕滩储层特征及其控制因素研究[J]. 特种油气藏,2020,27(1):17-24.

    Jiang Yuqiang, Deng Hongbing, Yi Juanzi, et al. Properties of different oolitic beach reservoirs in the Feixianguan Formation of Kaijiang-Liangping trough[J]. Special Oil & Gas Reservoirs, 2020, 27(1): 17-24.
    [6] 胡忠贵,董庆民,李世临,等. 川东—渝北地区长兴组—飞仙关组礁滩组合规律及控制因素[J]. 中国石油大学学报(自然科学版),2019,43(3):25-35.

    Hu Zhonggui, Dong Qingmin, Li Shilin, et al. Combination regularities of reef-beach and main controlling factors in Changxing-Feixianguan Formation of eastern Sichuan-northern Chongqing area[J]. Journal of China University of Petroleum, 2019, 43(3): 25-35.
    [7] 赫云兰,付孝悦,刘波,等. 川东北飞仙关组鲕滩沉积与成岩对储集层的控制[J]. 石油勘探与开发,2012,39(4):434-443.

    He Yunlan, Fu Xiaoyue, Liu Bo, et al. Control of oolitic beaches sedimentation and diagenesis on reservoirs in Feixianguan Formation, northeastern Sichuan Basin[J]. Petroleum Exploration and Development, 2012, 39(4): 434-443.
    [8] 赵文智,沈安江,胡素云,等. 中国碳酸盐岩储集层大型化发育的地质条件与分布特征[J]. 石油勘探与开发,2012,39(1):1-12.

    Zhao Wenzhi, Shen Anjiang, Hu Suyun, et al. Geological conditions and distributional features of large-scale carbonate reservoirs onshore China[J]. Petroleum Exploration and Development, 2012, 39(1): 1-12.
    [9] 蒋裕强,周亚东,陈智雍,等. 川东地区台内洼地二叠系生物礁、滩沉积格局及勘探意义[J]. 天然气地球科学,2019,30(11):1539-1550.

    Jiang Yuqiang, Zhou Yadong, Chen Zhiyong, et al. Sedimentary pattern and exploration significance of Permian reefs and shoals in intra-platform depressions, eastern Sichuan Basin[J]. Natural Gas Geoscience, 2019, 30(11): 1539-1550.
    [10] 刘建强,罗冰,谭秀成,等. 川东北地区飞仙关组台缘带鲕滩分布规律[J]. 地球科学:中国地质大学学报,2012,37(4):805-814.

    Liu Jianqiang, Luo Bing, Tan Xiucheng, et al. Distribution of marginal-platform oolitic shoal in Feixianguan Formation, northeast Sichuan, China[J]. Earth Science: Journal of China University of Geosciences, 2012, 37(4): 805-814.
    [11] 郭彤楼. 川东北元坝地区长兴组—飞仙关组台地边缘层序地层及其对储层的控制[J]. 石油学报,2011,32(3):387-394.

    Guo Tonglou. Sequence strata of the platform edge in the Changxing and Feixianguan Formations in the Yuanba area, northeastern Sichuan Basin and their control on reservoirs[J]. Acta Petrolei Sinica, 2011, 32(3): 387-394.
    [12] 谢增业,田世澄,单秀琴,等.川东北飞仙关组鲕滩天然气富集成藏特征及勘探前景[J].石油勘探与开发,2005,32(2):31-34.

    Xie Zengye, Tian Shicheng, Shan Xiuqin, et al. Features of gas accumulation and exploration foreground in oolitic reservoir of Feixianguan Formation in Sichuan Basin [J]. Petroleum Exploration and Development, 2005,32(2): 31-34.
    [13] 邹才能,徐春春,汪泽成,等. 四川盆地台缘带礁滩大气区地质特征与形成条件[J]. 石油勘探与开发,2011,38(6):641-651.

    Zou Caineng, Xu Chunchun, Wang Zecheng, et al. Geological characteristics and forming conditions of the large platform margin reef-shoal gas province in the Sichuan Basin[J]. Petroleum Exploration and Development, 2011, 38(6): 641-651.
    [14] 邹娟,杨迅,尹宏,等. 九龙山—剑阁地区长兴组、飞仙关组礁、滩储层特征及控制因素研究[J]. 天然气勘探与开发,2014,37(4):1-7.

    Zou Juan, Yang Xun, Yin Hong, et al. Characteristics of bioreef and shoal reservoirs in Changxing and Feixianguan Formations of Jiulongshan-Jiange area and their controlling factors[J]. Natural Gas Exploration and Development, 2014, 37(4): 1-7.
    [15] 冯林杰,蒋裕强,刘菲,等. 川东地区开江—梁平海槽南段飞仙关组鲕滩储层特征及主控因素[J]. 石油学报,2021,42(10):1287-1298.[

    Feng Linjie, Jiang Yuqiang, Liu Fei, et al. Reservoir characteristics and main controlling factors of oolitic shoal reservoir in Feixianguan Formation in the southern part of Kaijiang-Liangping trough, eastern Sichuan Basin[J]. Acta Petrolei Sinica, 2021, 42(10): 1287-1298.
    [16] 朱怡翔,石广仁. 火山岩岩性的支持向量机识别[J]. 石油学报,2013,34(2):312-322.

    Zhu Yixiang, Shi Guangren. Identification of lithologic characteristics of volcanic rocks by support vector machine[J]. Acta Petrolei Sinica, 2013, 34(2): 312-322.
    [17] Rosenblatt F. The perceptron: A probabilistic model for information storage and organization in the brain[J]. Psychological Review,1958, 65(6): 386-408.
    [18] 赵彦超,吴春萍,吴东平. 致密砂岩气层的测井评价:以鄂尔多斯盆地大牛地山西组一段气田为例[J]. 地质科技情报,2003,22(4):65-70.[

    Zhao Yanchao, Wu Chunping, Wu Dongping. Logging evaluation to tight gas sandstone: A case study from the First member of Shanxi Formation in Daniudi gas pool, Ordos Basin, China[J]. Geological Science and Technology Information, 2003, 22(4): 65-70.
    [19] Ghiasi M M, Zendehboudi S, Mohsenipour A A. Decision tree-based diagnosis of coronary artery disease: CART model[J]. Computer Methods and Programs in Biomedicine, 2020, 192: 105400.
    [20] 张俊玉,胡家豪,黄嵩. CART决策树方法在煤电厂节能降耗中的应用[J]. 控制与决策,2021,36(5):1232-1238.

    Zhang Junyu, Hu Jiahao, Huang Song. Application of CART decision tree model in reducing coal consumption in coal power plant[J]. Control and Decision, 2021, 36(5): 1232-1238.
    [21] 陈平,徐星. 基于CART算法的带钢抗拉强度影响因素研究[J]. 控制工程,2015,22(2):276-281.

    Chen Ping, Xu Xing. Research of tensile strength of strip steel based on CART[J]. Control Engineering of China, 2015, 22(2): 276-281.
    [22] 魏合理,陈秀红,戴聪明. 通用大气辐射传输软件(CART)及其应用[J]. 红外与激光工程,2012,41(12):3360-3366.

    Wei Heli, Chen Xiuhong, Dai Congming. Combined atmospheric radiative transfer (CART) model and its applications[J]. Infrared and Laser Engineering, 2012, 41(12): 3360-3366.
    [23] 刘玉茹,赵成萍,臧军,等. CART分析及其在故障趋势预测中的应用[J]. 计算机应用,2017,37(增刊2):57-59,73.

    Liu Yuru, Zhao Chengping, Zang Jun, et al. Analysis of CART regression tree and its application in fault trend forecasting[J]. Journal of Computer Applications, 2017, 37(Suppl.2): 57-59, 73.
    [24] 吴冠朋,黄伟,刘毅慧. CART算法在原发性肝癌放疗后HBV再激活的应用[J]. 生物信息学,2017,15(3):164-170.

    Wu Guanpeng, Huang Wei, Liu Yihui. Application of HBV reactivation in primary liver carcinoma after radiotherapy based on CART algorithm[J]. Chinese Journal of Bioinformatics, 2017, 15(3): 164-170.
    [25] 王晓畅,张军,李军,等. 基于交会图决策树的缝洞体类型常规测井识别方法:以塔河油田奥陶系为例[J]. 石油与天然气地质,2017,38(4):805-812.

    Wang Xiaochang, Zhang Jun, Li Jun, et al. Conventional logging identification of fracture-vug complex types data based on crossplots-decision tree: A case study from the Ordovician in Tahe oilfield, Tarim Basin[J]. Oil & Gas Geology, 2017, 38(4): 805-812.
    [26] 孙予舒,黄芸,梁婷,等. 基于XGBoost算法的复杂碳酸盐岩岩性测井识别[J]. 岩性油气藏,2020,32(4):98-106.

    Sun Yushu, Huang Yun, Liang Ting, et al. Identification of complex carbonate lithology by logging based on XGBoost algorithm[J]. Lithologic Reservoirs, 2020, 32(4): 98-106.
    [27] 李洪奇,谭锋奇,许长福,等. 基于决策树方法的砾岩油藏岩性识别[J]. 测井技术,2010,34(1):16-21.

    Li Hongqi, Tan Fengqi, Xu Changfu, et al. Lithology identification of conglomerate reservoir based on decision tree method[J]. Well Logging Technology, 2010, 34(1): 16-21.
    [28] Han J W, Kamber M. Data mining concepts and techniques[M]. 2nd ed. San Francisco: Morgan Kaufmann Publishers, 2006.
    [29] Witten I H, Frank E. 数据挖掘:实用机器学习技术[M]. 董琳,邱泉,于晓峰,等译. 2版. 北京:机械工业出版社,2006.

    Witten I H, Frank E. Data mining: practical machine learning tools and techniques[M]. Dong Lin, Qiu Quan, Yu Xiaofeng, et al, trans. 2nd ed. Beijing: China Machine Press, 2006.
    [30] 李雄炎,周金昱,李洪奇,等. 复杂岩性及多相流体智能识别方法[J]. 石油勘探与开发,2012,39(2):243-248.

    Li Xiongyan, Zhou Jinyu, Li Hongqi, et al. Computational intelligent methods for predicting complex lithologies and multiphase fluids[J]. Petroleum Exploration and Development, 2012, 39(2): 243-248.
    [31] Kohonen T. Self-organization and associative memory[M]. 3rd ed. Heidelberg: Springer, 1989.
    [32] Kohonen T. The self-organizing map[J]. Proceedings of the IEEE, 1990, 78(9): 1464-1480.
    [33] Shi G R, Zhou X X, Zhang G Y, et al. The use of artificial neural network analysis and multiple regression for trap quality evaluation: A case study of the northern Kuqa Depression of Tarim Basin in western China[J]. Marine and Petroleum Geology, 2004, 21(3): 411-420.
    [34] 王俊,曹俊兴,尤加春. 基于GRU神经网络的测井曲线重构[J]. 石油地球物理勘探,2020,55(3):510-520.

    Wang Jun, Cao Junxing, You Jiachun. Reconstruction of logging traces based on GRU neural network[J]. Oil Geophysical Prospecting, 2020, 55(3): 510-520.
    [35] 李霞,范宜仁,邓少贵,等. 自动划分层序单元的测井多尺度数据融合方法[J]. 石油勘探与开发,2009,36(2):221-227.

    Li Xia, Fan Yiren, Deng Shaogui, et al. Automatic demarcation of sequence stratigraphy using the method of well logging multiscale data fusion[J]. Petroleum Exploration and Development, 2009, 36(2): 221-227.
    [36] 肖小玲,靳秀菊,张翔,等. 基于常规测井与电成像测井多信息融合的裂缝识别[J]. 石油地球物理勘探,2015,50(3):542-547.

    Xiao Xiaoling, Jin Xiuju, Zhang Xiang, et al. Fracture identification based on information fusion of conventional logging and electrical imaging logging[J]. Oil Geophysical Prospecting, 2015, 50(3): 542-547.
    [37] 马陇飞,萧汉敏,陶敬伟,等. 基于梯度提升决策树算法的岩性智能分类方法[J]. 油气地质与采收率,2022,29(1):21-29.

    Ma Longfei, Xiao Hanmin, Tao Jingwei, et al. Intelligent lithology classification method based on GBDT algorithm[J]. Petroleum Geology and Recovery Efficiency, 2022, 29(1): 21-29.
    [38] Friedman J H. Greedy function approximation: A gradient boosting machine[J]. The Annals of Statistics, 2001, 29(5): 1189-1232.
    [39] 王瑞妮. 基于流形学习和GBDT的异常流量检测方法[D]. 哈尔滨:哈尔滨工程大学,2020.

    Wang Ruini. Anomaly traffic detection based on manifold learning and GBDT[D]. Harbin: Harbin Engineering University, 2020.
    [40] 王焱. 基于随机梯度提升决策树的行人检测算法设计与实现[D]. 杭州:浙江大学,2017.

    Wang Yan. Design and implementation of a pedestrian detection algorithm using s-GBDT[D]. Hangzhou: Zhejiang University, 2017.
    [41] Nassif A B. Short term power demand prediction using stochastic gradient boosting[C]//5th international conference on electronic devices, systems and applications (ICEDSA). Ras Al Khaimah: IEEE, 2016: 1-4.
    [42] 谢振平,孙桃. 自组织决策树的联想记忆在线学习模型[J]. 模式识别与人工智能,2017,30(1):21-31.

    Xie Zhenping, Sun Tao. Online associative memory model based on self-organizing decision tree[J]. Pattern Recognition and Artificial Intelligence, 2017, 30(1): 21-31.
    [43] Sakkari M, Hamdi M, Elmannai H, et al. Feature extraction-based deep self-organizing map[J]. Circuits, Systems, and Signal Processing, 2022, 41(5): 2802-2824.
  • 加载中
通讯作者: 陈斌, bchen63@163.com
  • 1. 

    沈阳化工大学材料科学与工程学院 沈阳 110142

  1. 本站搜索
  2. 百度学术搜索
  3. 万方数据库搜索
  4. CNKI搜索

Figures(10)  / Tables(6)

Article Metrics

Article views(76) PDF downloads(31) Cited by()

Proportional views
Related
Publishing history
  • Received:  2022-04-27
  • Revised:  2022-07-06
  • Accepted:  2022-09-16
  • Published:  2024-06-10

Lithology and Distribution Characteristics of the Oolitic Beach from the Feixianguan Formation, Eastern Sichuan Basin

doi: 10.14027/j.issn.1000-0550.2022.089
Funds:

National Natural Science Foundation of China 41430316

National Science and Technology Major Project 2017ZX05008-004-008

Abstract: Objective To solve the problems such as unclear lithological variations of the Feixianguan Formation on the western side of the marine trough's southern segment and within the plateau interior of the eastern part of the Sichuan Basin, Methods the lithological types and characteristics of the Feixianguan Formation were studied by comprehensively using multiple geological data such as cores, drilling and logging, and an intelligent identification method of lithological logging based on machine learning was proposed, which solved the technical problems of fine identification of lithology in the old area, and revealed the lithology, distribution and evolution of the Feixianguan Formation in the area. Results and Conclusions (1) The Feixianguan Formation is mainly composed of lithology such as mudstone, mud crystal limestone, argillary limestone, granule limestone, granulous dolomite, mud crystal dolomite, paste dolomite, and gypsum rock; (2) Comparison found that the improved gradient-boosted decision tree algorithm, namely Stochastic Gradient Boosting Decision Tree (SGBDT), is superior to other algorithms for constructing lithology models, and is more suitable for carbonate rock complex lithology identification; (3) The granule limestone developed intensively in the area south of the Kaijiang-Liangping Sea Trough between Feixianguan I period and Feixianguan Ⅲ period, and the granulous dolomite was concentrated in the Feixianguan Ⅱ period and distributed and scattered; (4) The distribution of the oolitic shoal in the area exhibits significant variation. The main development of the plateau ancient geomorphological high point and the edge of the platform in the Feixianguan I period. In the Feixianguan Ⅱ period, most of the the oolitic shoal at the edge of the platform were developed, and a small number of ancient landforms in the platform were developed.And the main development of the plateau ancient geomorphological high point in the Feixianguan Ⅲ period.

YE Yu, CHENG Chao, JIANG YuQiang, YI JuanZi, DENG HongBing, LI Xi, GU YiFan, CHEN Yan. Lithology and Distribution Characteristics of the Oolitic Beach from the Feixianguan Formation, Eastern Sichuan Basin[J]. Acta Sedimentologica Sinica, 2024, 42(3): 1032-1046. doi: 10.14027/j.issn.1000-0550.2022.089
Citation: YE Yu, CHENG Chao, JIANG YuQiang, YI JuanZi, DENG HongBing, LI Xi, GU YiFan, CHEN Yan. Lithology and Distribution Characteristics of the Oolitic Beach from the Feixianguan Formation, Eastern Sichuan Basin[J]. Acta Sedimentologica Sinica, 2024, 42(3): 1032-1046. doi: 10.14027/j.issn.1000-0550.2022.089
  • 自20世纪90年代以来,众多学者围绕四川盆地下三叠统飞仙关组沉积相类型、相带展布[115]及气藏主控因素[911,15]等开展了大量研究,相继在普光、龙门、高峰场、巫山坎、双家坝等海槽台缘地带取得重大突破。截至目前,天然气探明储量累计超过6 000×108 m3。然而,飞仙关组储层类型复杂、岩性多样,不同岩性测井曲线响应特征差异不明显,海槽南段西侧、台内等地区岩性变化规律及沉积格局演化规律不明确,严重制约了天然气勘探进度。

    根据各种测井曲线间的内在联系,传统的交会图和双矿物模型等在非均质性强烈的碳酸盐岩岩性识别中应用效果并不突出,数据重叠现象较为严重,岩性解释精度受限[413]。随着机器学习算法在岩性识别领域中的广泛应用,诸如人工神经网络(Artificial Neural Network,ANN)、支持向量机等“黑盒算法”[1618]均是近年的研究热点,这些黑盒算法对数据与属性的因果关系有所表征,但忽略了测井数据随深度变化的前后关联,因此其岩性识别的准确度有待提升。相较于黑盒算法,“白盒算法”决策树对复杂碳酸盐岩识别具有显著优势,对于各种影响因素有着深刻的表征以及指导作用[1936]。而梯度提升决策树(Grandient Boosting Decision Tree,GBDT)是解决不均衡数据的高预测精度算法[3738],通过残差分析对目标进行精确分类,但其运算量庞大,导致计算速度低、资源占用量大。为此,将随机因素引入GBDT,提出利用随机梯度提升决策树(Stochastic Grandient Boosting Decision Tree,SGBDT)算法[3941]建立老井岩性测井精细识别模型,并对全区飞仙关组岩性进行识别,在此基础上总结有利岩性分布及沉积格局演化规律,以期为后续地质研究提供有力依据。

  • 研究区位于四川盆地东部[1](以下简称“川东”)(图1a),受晚二叠世—早三叠世北部构造带和峨眉地裂的共同影响,形成以开江—梁平海槽为主的槽台沉积格局[5,8,15]。区内飞仙关组在长兴组古地貌基础下继承性沉积,海槽逐渐被“填平补齐”。水体逐渐变浅,至飞仙关组中—晚期,逐渐演化成碳酸盐岩开阔台地,末期则形成蒸发台地[49]。依据电性、岩性特征[915],将区内飞仙关期划分为四段,即:飞仙关组四段(飞四段)、飞仙关组三段(飞三段)、飞仙关组二段(飞二段)、飞仙关组一段(飞一段)(图1b)。飞四段岩性主要由紫红色泥岩、膏质云岩、石膏岩、泥晶云岩构成;飞三段以泥晶灰岩为主,局部夹薄层鲕粒灰岩;飞二段岩性主要为泥晶灰岩和鲕粒灰岩,受白云石化和组构溶蚀作用影响[15],部分地区形成鲕粒云岩;受陆源碎屑和古地貌差异等因素影响,区内自SW至NE向发育混积台地相和清水台地相[15],混积台地相飞一段底部为泥质灰岩,下部为泥晶灰岩夹薄层泥质灰岩,向上局部地区发育鲕粒灰岩,顶部则为紫红色泥岩,清水台地相主要发育泥晶灰岩,少部分地区发育鲕粒灰岩(图2)。

    Figure 1.  (a) Location of the study area (modified from reference [1]); (b) stratigraphic column of the Feixianguan Formation in the eastern Sichuan Basin

    Figure 2.  Feixianguan Formation litho well logging characteristics from eastern Sichuan Basin

  • 泥晶灰岩岩心主要为灰、深灰和暗褐灰色,见水平层理。从薄片可以看出,主要由泥—粉晶方解石构成,几乎不含泥纹,孔隙发育差,岩心孔隙度介于0.6%~1.54%,平均为1.03%。其沉积特征表明该类岩石形成于水体较为安静的环境中,在开阔台地内部静浅水、斜坡带以及较深水盆地(海槽)等环境中均有分布。常规测井响应特征为:自然伽马(GR)低值,补偿中子(CNL)低值,声波时差(AC)低值,补偿密度(DEN)中—高值,深侧向电阻率(RT)高值(图2a)。

  • 鲕粒云岩岩心为浅灰、褐灰色,见槽状、板状交错层理。从薄片可以看出,鲕粒主要为残余鲕,由他形—半自形粉晶白云石组成,孔隙以粒间溶孔、晶间溶孔为主,岩心孔隙度介于1.53%~12.34%,平均为5.07%,为最有利储层岩性,形成于水动力条件较强的沉积环境之中,如台地边缘(台缘滩)、台内局部地貌高地(台内点滩)等。常规测井响应特征为:补偿密度(DEN)中等,深侧向电阻率(RT)中值(图2b)。

  • 鲕粒灰岩岩心以灰、深灰和灰白色为主,分选、磨圆中等—好。从薄片可以看出,多为正常鲕粒,粒间孔发育,见少量粒内溶孔,部分鲕粒被白云石交代,岩心孔隙度介于1%~3%,平均为2.08%,为有利储层岩性,形成于台地边缘和台内点滩等水动力较强的沉积环境。常规测井响应特征为:补偿密度(DEN)中等,深侧向电阻率(RT)中—低值(图2c)。

  • 膏质云岩岩心为浅灰色,见石膏结核和少量针孔,发育层状层理,由泥晶白云石组成,岩心孔隙度介于0.35%~1.26%,孔隙发育差,多分布于潮坪和滩间海中。常规测井响应特征为:补偿密度(DEN)高等,深侧向电阻率(RT)中—高值(图2d)。

  • 泥晶云岩岩心为浅灰—灰色,发育层状层理,含少量灰质,岩心孔隙度介于0.86%~1.65%,与膏质云岩呈互层产出。常规测井响应特征为:补偿密度(DEN)高等,深侧向电阻率(RT)中—高值(图2e)。

  • 泥质灰岩岩心为灰—深灰色,发育层状层理,主要由泥—粉晶方解石构成,含较多的泥纹,含少量完整生物,如介形虫、腹足等,孔隙发育差,岩心孔隙度介于0.49%~1.33%,多见于斜坡—海槽环境。常规测井响应特征为:自然伽马(GR)低—中值,补偿密度(DEN)中—高等,深侧向电阻率(RT)中—低值(图2f)。

  • 膏岩岩心为白色、浅灰色,通常以结核状出现,纹层和变形构造发育,主要分布于飞四段时期的潮坪—蒸发潟湖环境之中,常与泥晶云岩伴生,岩心孔隙度小于1%,物性极差。常规测井响应特征为:自然伽马(GR)低值,补偿密度(DEN)高等,补偿中子(CNL)低值,声波时差(AC)低值,深侧向电阻率(RT)高值(图2g)。

  • 泥岩主要分布于飞四段时期的蒸发氧化环境中,颜色以红棕色为主,物性极差。常规测井响应特征为:自然伽马(GR)高值,补偿密度(DEN)中等,补偿中子(CNL)中—高值,声波时差(AC)低—中值,深侧向电阻率(RT)低值(图2h)。

    岩石类型、测井常规曲线交会图分析显示(图3),不同岩性阈值界限不清,数据重叠现象较为严重,可能导致后期岩性解释和鲕滩分布等出现极大偏差。为此,以大量薄片鉴定岩性为主,结合钻录井资料、岩性测井响应特征参数建立岩性分类标签,利用GBDT、SGBDT和ANN算法对数据进行高精度分类,即在已知标签监督下,对复杂的、多维的、模糊的不均衡数据精细归类,准确识别不同岩性,在此基础上对比不同算法优劣性,得到高适应岩性识别模型,从而完成老井岩性识别任务。

    Figure 3.  Cross⁃plot of RT⁃AC and GR⁃AC

  • 人工神经网络(ANN),由Frank Rosenblatt[17]在1958年提出的一种前馈式人工神经网络:连接多个特征值,经过线性和非线性的组合,产生输出影响其他神经元实现非线性映射。根据有限的数据信息,在代价函数的约束下,对多输入多输出非线性数据具有良好的预测能力[1719,4243]图4a)。

    Figure 4.  Artificial Neural Network (ANN) (a), Gradient Boosting Decision Tree (GBDT) (b), and Stochastic Gradient Boosting Decision Tree (SGBDT) (c) schematics (modified from references [34,36])

  • GBDT算法是以CART决策树为基学习模型、采用梯度提升(Grandient Boosting)对决策树多次迭代最终累加形成强学习模型的一种集成学习经典Boosting算法(图4b)[3438],具有高预测精度、对Robust损失函数的利用和对异常值极为敏感的属性,其开发目的主要是解决实际不均衡数据的分类和回归问题。

    基于Boosting思想,GBDT算法逐次拟合新模型,即在梯度方向上训练一个新的学习模型来降低前一个学习模型的残差,并且基于当前学习模型的基础迭代生成新的学习模型,使其最终与损失函数负梯度相关,并与整个集成系统相连接,其计算公式为:

    Ym+1x=Ymx+ρmhx         1mM (1)

    式中:Ym+1x)为第m+1个学习模型;Ymx)为第m个学习模型;ρm为第m次学习率;hx)为当前损失函数负梯度方向上拟合得到的基学习模型;M为迭代设置总次数。

    具体流程分为四步,即:

    (1) 初始化第一个学习模型,设迭代次数为M,其计算公式为:

    Y0x=arg min l˙=1nQUi,ρ (2)

    式中:Y0x)为初始化学习模型;QUi,ρ)为损失函数;Ui 为第i个预测目标;ρ为学习率。

    (2) 计算此次迭代中回归树的拟合目标。即当前损失函数的负梯度值δm,i,计算公式为:

    δm,i=QUi,YxiYxiYxi=Ym-1(xi) i=1,2,3,,n (3)

    (3) 经过m次迭代,得到模型最优的基分类模型Bm

    Bm=argmini=1n[δm,i-βh(xi;Bm)] (4)

    式中:β为计算乘子;hxi;B m )为最优基分类模型B m 的损失函数负梯度方向上拟合得到的基学习模型。

    通过线性寻优方式计算最优学习率ρm,更新下一个学习模型:

    Ymx=Ym-1x+ρmhxi;Bm (5)

    (4) 重复步骤1~3,直到m=M结束形成强学习模型G。

  • 由于GBDT算法每次迭代均选取全部训练集数据,导致计算速度慢,资源占用量高,可能发生过拟合现象。因此,基于Friedman[38]提出的随机梯度提升(Stochastic Grandient Boosting)方法将随机因素引入GBDT,改进后得到随机梯度提升决策树(SGBDT)(图4c),即设定子采样因子ff<1),在每次迭代中随机选择部分样本构建学习模型(样本值选择而不替换),提高模型泛化能力[3638]

    Ymx=f Ym-1x+ρmhxi;Bm (6)

    式中:f为采样因子。

  • 测井响应特征是地层中岩性、流体等物理变化的综合反应,不同测井参数对岩性的敏感性具有明显的区分度。在岩性识别前,采用Spearman秩相关系数矩阵表征各测井参数对岩性的敏感性,结果表明岩性与密度(DEN)、声波时差(AC)、深侧向电阻率(RT)、自然伽马(GR)、中子(CNL)整体上具有较高的相关性(表1)。因此,最终选用这5种作为输入参数,建立岩性精细识别模型。

    参数类型DENACRTGRCNLRXOCALLithology
    DEN1.00-0.10-0.010.370.050.10-0.390.45
    AC1.00-0.350.560.79-0.30-0.06-0.53
    RT1.00-0.40-0.330.83-0.040.35
    GR1.000.73-0.30-0.09-0.41
    CNL1.00-0.270-0.37
    RXO1.00-0.200.10
    CAL1.000.12
    Lithology1.00

    Table 1.  Spearman rank correlation coefficient matrix of lithology sensitive parameters in eastern Sichuan Basin

  • 考虑到岩性模型的适用性与准确性,实验样品取自川东地区不同区域13口取心井的不同深度(图1表2),共获得7 491个有效数据。选用其中12口取心井的3 743个样本作为训练集,用剩余一口井(天东100井)的样本作为模型评估的检验集(数据见表3),其中泥岩、泥晶灰岩、泥质灰岩、鲕粒灰岩、鲕粒云岩、泥晶云岩、膏质云岩、膏岩样本个数分别为515,308,1 901,2 422,1 017,628,315和385个。利用Python软件编程,将5条测井曲线与对应的岩性标签分别在ANN、GBDT算法和SGBDT算法程序中进行训练学习,并在测试集上进行分类预测,得到其对应岩性的混淆矩阵、克莱姆相关系数和交会图相关系数,作为模型检验效果的衡量标准。根据网格搜索算法调参确定,GBDT和SGBDT算法决策树深度为9,学习率为0.1,最小样本数为30,叶节点最小样本数为40,迭代次数根据期望损失计算公式[36]确定为50次,SGBDT子采样随机因子为60%(表4);ANN算法参数直接通过Keras程序包导入,其隐藏层神经元一般设置为输入层神经元个数的1~5倍,经过反复训练尝试,最终确定5-16-8-1模型效果最佳,其中隐藏层为2层,训练次数设置为10 000次,学习率取优为0.01(表5)。

    区块井名层位深度/m
    高峰场峰 4井飞三段3 773.05~3 824.5
    峰 15井3 774.68~3 923.65
    门—门西门7井飞二段2 964.40~2 972.77
    七里峡七里51井飞二段3 853.52~3 980.41
    七里52井飞二段—飞三段3 755.25~3 809.89,3 930.23~3 990.42
    七里58井飞二段3 928.00~3 986.00
    大池干池59井飞二段3 342.00~3 349.00
    池028-3井飞三段2 344.82~2 362.32
    天东天东100井飞仙关3 738.56~3 845.08
    天东110井飞二段3 448.31~3 457.50
    天东9井飞二段3 541.20~3 581.72
    黄草峡草10井飞一段1 772.85~1 803.94
    卧龙河卧79井飞一段、飞三段3 950.00~3 968.00,4 201.00~4 214.68

    Table 2.  Experimental samples and data sources

    泥岩泥晶灰岩泥质灰岩鲕粒灰岩鲕粒云岩泥晶云岩膏质云岩膏岩总数
    标签代码12345678
    训练集样本数3302694906939923642853203 743
    占比%8.817.1913.0918.5126.509.727.618.55100
    测试集样本数185391 4111 72925.026430653 748
    占比%4.941.0437.6546.130.677.040.801.73100

    Table 3.  Database lithology distribution

    算法类型迭代次数决策树深度最小样本数叶节点最小样本数学习步长子采样随机因子(v/v)标准偏差AUC (v/v)
    GBDT10650200.110.019 50.915 4
    20650200.110.019 40.916 1
    30650200.110.019 40.916 4
    40650200.110.019 20.919 3
    50650200.110.018 60.920 8
    60650200.110.019 30.916 6
    70650200.110.020 30.914 7
    50350200.110.020 50.913 5
    50550200.110.019 20.915 2
    50750200.110.018 70.916 4
    50950200.110.018 30.929 6
    501150200.110.019 20.918 3
    501350200.110.019 50.917 0
    501550200.110.020 00.913 8
    50910200.110.020 70.919 4
    50930200.110.002 00.933 4
    50950200.110.020 00.927 3
    50970200.110.020 20.925 6
    50990200.110.020 70.924 1
    50930100.110.021 50.915 8
    50930200.110.020 10.922 1
    50930300.110.017 70.923 5
    50930400.110.017 20.934 1
    50930500.110.017 00.922 8
    50930600.110.022 40.914 7
    50930400.00510.024 00.918 2
    50930400.0510.021 60.924 3
    50930400.110.018 60.941 5
    50930400.210.021 80.914 0
    50930400.310.02330.9117
    SGBDT50930400.10.50.020 20.933 4
    50930400.10.60.018 30.944 2
    50930400.10.70.018 60.940 3
    50930400.10.80.021 00.938 9
    50930400.10.90.021 90.936 6

    Table 4.  GBDT and SGBDT algorithm parameter statistics

    优化算法激活函数学习率动量控制Dropout权重约束方法权重初始化隐藏层数量隐藏层神经元数输入样本数迭代次数AUC/(v/v)
    ANNAdamsoftplus0.000 01011uniform116501 0000.759 8
    SGDsoftplus0.000 01011uniform116501 0000.729 5
    Adagradsoftplus0.000 01011uniform116501 0000.685 4
    Nadamsoftplus0.000 01011uniform116501 0000.795 3
    RMSpropsoftplus0.000 01011uniform116501 0000.810 6
    Adadeltasoftplus0.000 01011uniform116501 0000.796 1
    RMSpropsoftmax0.000 01011uniform116501 0000.689 4
    RMSpropsoftsign0.000 01011uniform116501 0000.694 0
    RMSproprelu0.00001011uniform116501 0000.788 5
    RMSproptanh0.000 01011uniform116501 0000.669 2
    RMSpropsigmoid0.000 01011uniform116501 0000.800 1
    RMSproplinear0.000 01011uniform116501 0000.710 6
    RMSpropsoftplus0.000 1011uniform116501 0000.761 5
    RMSpropsoftplus0.001011uniform116501 0000.769 5
    RMSpropsoftplus0.01011uniform116501 0000.816 2
    RMSpropsoftplus0.1011uniform116501 0000.723 8
    RMSpropsoftplus0.2011uniform116501 0000.763 5
    RMSpropsoftplus0.010.211uniform116501 0000.689 4
    RMSpropsoftplus0.010.411uniform116501 0000.826 1
    RMSpropsoftplus0.010.611uniform116501 0000.811 9
    RMSpropsoftplus0.010.811uniform116501 0000.762 5
    RMSpropsoftplus0.010.911uniform116501 0000.774 2
    RMSpropsoftplus0.010.401uniform116501 0000.809 5
    RMSpropsoftplus0.010.40.11uniform116501 0000.769 5
    RMSpropsoftplus0.010.40.21uniform116501 0000.849 1
    RMSpropsoftplus0.010.40.31uniform116501 0000.855 2
    RMSpropsoftplus0.010.40.41uniform116501 0000.721 9
    RMSpropsoftplus0.010.40.51uniform116501 0000.652 9
    RMSpropsoftplus0.010.40.61uniform116501 0000.706 2
    RMSpropsoftplus0.010.40.71uniform116501 0000.701 8
    RMSpropsoftplus0.010.40.81uniform116501 0000.686 2
    RMSpropsoftplus0.010.40.91uniform116501 0000.711 8
    RMSpropsoftplus0.010.40.32uniform116501 0000.762 8
    RMSpropsoftplus0.010.40.33uniform116501 0000.792 1
    RMSpropsoftplus0.010.40.34uniform116501 0000.706 8
    RMSpropsoftplus0.010.40.35uniform116501 0000.799 6
    RMSpropsoftplus0.010.40.31zero116501 0000.823 1
    RMSpropsoftplus0.010.40.31lecun_uniform116501 0000.810 6
    RMSpropsoftplus0.010.40.31normal116501 0000.796 8
    RMSpropsoftplus0.010.40.31glorot_normal116501 0000.755 9
    RMSpropsoftplus0.010.40.31uniform216-16501 0000.859 8
    RMSpropsoftplus0.010.40.31uniform316-16501 0000.819 2
    RMSpropsoftplus0.010.40.31uniform416-16501 0000.829 5
    RMSpropsoftplus0.010.40.31uniform216-5501 0000.762 1
    RMSpropsoftplus0.010.40.31uniform216-8501 0000.862 9
    RMSpropsoftplus0.010.40.31uniform216-11501 0000.823 6
    RMSpropsoftplus0.010.40.31uniform216-8101 0000.756 1
    RMSpropsoftplus0.010.40.31uniform216-81001 0000.812 6
    RMSpropsoftplus0.010.40.31uniform216-81501 0000.764 1
    RMSpropsoftplus0.010.40.31uniform216-82001 0000.689 2
    RMSpropsoftplus0.010.40.31uniform216-82501 0000.754 9
    RMSpropsoftplus0.010.40.31uniform216-83001 0000.816 4
    RMSpropsoftplus0.010.40.31uniform216-8502 0000.867 6
    RMSpropsoftplus0.010.40.31uniform216-8503 0000.870 9
    RMSpropsoftplus0.010.40.31uniform216-8504 0000.871 1
    RMSpropsoftplus0.010.40.31uniform216-8505 0000.873 5
    RMSpropsoftplus0.010.40.31uniform216-8506 0000.873 9
    RMSpropsoftplus0.010.40.31uniform216-8507 0000.875 2
    RMSpropsoftplus0.010.40.31uniform216-8508 0000.876 1
    RMSpropsoftplus0.010.40.31uniform216-8509 0000.879 2
    RMSpropsoftplus0.010.40.31uniform216-85010 0000.888 6
    RMSpropsoftplus0.010.40.31uniform216-85011 0000.856 2
    RMSpropsoftplus0.010.40.31uniform216-85012 0000.828 4
    续表

    Table 5.  ANN algorithm parameter statistics

  • 实验结果表明(图5~7),在天东100井上,SGBDT算法对泥岩、泥晶灰岩、泥质灰岩、鲕粒灰岩、鲕粒云岩、泥晶云岩、膏质云岩、膏岩的识别准确率分别为97.30%,95.59%,91.42%,96.92%,100%,95.45%,90.00%,87.08%,其中对储层有利岩性识别尤为准确,整体上其克莱姆相关系数、交会图相关系数分别达到了0.945、0.920,SGBDT算法判别岩性效果优,适合于碳酸盐岩复杂岩性评价。

    Figure 5.  Partial results of the logging interpretation for well TD100

    Figure 6.  SGBDT, GBDT, and ANN discriminant analysis lithology confusion matrices

    Figure 7.  Algorithm cross⁃plot histogram of the Gabriel Cramer and intersection coefficients

    相较于SGBDT,GBDT的整体识别准确率和相关系数略有下降(图5~7),表明加入随机因素的SGBDT泛化能力(Robust性)较GBDT有所提高,其改进后的算法足以提供可靠的预测结果(图6a)。而ANN算法的整体识别准确率和相关系数明显下降,对泥晶灰岩和泥质灰岩的识别效果较差,其识别准确率分别为67.94%、62.74%,对复杂岩性识别评价效果有待考量。

    利用SGBDT决策树建立的岩性识别模型对研究区全井进行岩性预测,为了验证识别模型的适用性能力,选取3口取心井段长短不一且沉积环境各异的取心井进行检验。其结果表明,SGBDT算法对区内不同沉积相区的各种岩性识别准确率和整体相关性均保持在88.5%以上,识别效果较好(表6)。

    井名泥岩/%泥质灰岩/%泥晶灰岩/%鲕粒灰岩/%鲕粒云岩/%泥晶云岩/%膏质云岩/%膏岩/%克莱姆相关系数(v/v)交会图相关系数(v/v)
    七里55井97.3092.6591.1596.8189.8395.4590.0089.470.9240.918
    天东9井96.5988.6994.5590.910.8970.933
    新13井92.2895.6591.670.9070.909

    Table 6.  Recall rate, Gabriel Cramer, and cross⁃plot correlation coefficients of single well lithology identification

  • 在岩性识别结果的基础上,总结川东飞仙关有利岩性分布特征。平面上,鲕粒岩多呈透镜体几何形态独立分布,其中鲕粒灰岩多分布在开江—梁平海槽以南,而鲕粒云岩分布比较分散,一些分布于开江—梁平海槽以北,剩余部分分布于开江—梁平海槽南部边缘(图8)。纵向上鲕粒灰岩在飞一段至飞三段均有发育,而鲕粒云岩集中发育于飞二段时期,从岩性连井剖面看出,有利岩性纵向多期叠置,夹薄层泥晶灰岩,非均质性强,连通性差(图9)。

    Figure 8.  Favorable lithological planar distribution of the Feixianguan Formation in eastern Sichuan Basin

    Figure 9.  Comparison of the lithologic columns from the Feixianguan Formation in eastern Sichuan Basin

  • 结合岩心分析化验、沉积相带展布等研究成果[515],总结了飞仙关组鲕粒滩演化与分布规律。飞一段沉积早期,研究区处于海侵体系域,伴随海平面的相对下降,古地貌高点的鲕滩滩体暴露于海平面之上,受淡水淋滤作用的影响,被方解石完全充填的鲕滩体发生组构性溶蚀[15],在此期间发育的鲕粒滩主要分布于开阔台地与台地边缘(图10a)。飞一段沉积晚期,陆源碎屑的注入导致开阔台地沉积环境逐渐演化为混合台地环境,其陡增的泥质含量抑制了台内鲕滩的发育,鲕粒滩在开阔台地古地貌高点偶有发育以及未受影响的台地边缘发育(图10b)。飞二段高位体系域时期,海平面持续下降,台地与海槽的转折处处于动荡的高能水体环境,鲕粒在台缘沉积下来,当鲕滩厚度足够大时,其障壁作用有利于形成富Mg2+的流体,使部分鲕滩滩体白云石化,此期间鲕粒滩储层主要发育于台地边缘,开阔台地古地貌高点零星发育(图10c)。飞三段时期,研究区从早期的海侵体系域逐渐过渡为高位体系域,海槽被逐渐填平补齐,台缘边缘逐渐向原海槽—斜坡方向迁移,台缘鲕滩也随之迁移,该阶段鲕粒滩以台内点滩为主(图10d)。飞四段时期,相对海平面处于极低点,受陆源碎屑和强烈蒸发效应的控制,全区演化为蒸发台地,鲕滩不发育。

    Figure 10.  Facies and the oolitic shoal distribution map of Feixianguan Formation in eastern Sichuan Basin

  • (1) 研究区岩性复杂,依据岩心、测录井资料,共划分出8种岩性,分别为泥岩、泥晶灰岩、泥质灰岩、鲕粒灰岩、鲕粒云岩、泥晶云岩、膏质云岩、膏岩。

    (2) SGBDT算法对不均衡的岩性识别数据精度足以提供可靠的预测结果,与ANN、GBDT算法相比,SGBDT算法的岩性识别克莱姆系数、交会图系数分别达到了0.945、0.920,表明该算法有良好的泛化能力且更适合碳酸盐岩复杂岩性识别。

    (3) 区内鲕粒灰岩于飞一段—飞三段时期在开江—梁平海槽以南地区集中发育,而鲕粒云岩于飞二段时期集中发育但分布分散,纵横向上非均质性强,连通性差。

    (4) 川东飞仙关时期鲕粒滩分布差异明显,飞一段时期鲕滩主要在台内古地貌高地和台地边缘发育;飞二段时期多发育台缘鲕滩,台内鲕滩零星出现;飞三段时期受海槽填平补齐影响,台缘鲕滩向原海槽—斜坡方向迁移,以台内点滩为主;飞四段时期鲕粒滩则不发育。

Reference (43)

Catalog

    /

    DownLoad:  Full-Size Img  PowerPoint
    Return
    Return