基于改进长短记忆神经网络的深层致密储层裂缝测井识别

doi:10.13745/j.esf.sf.2025.7.18

摘要/Abstract

摘要：

辽河坳陷中央凸起深层致密基岩潜山发育裂缝性油气储层,资源潜力巨大,但埋深大、岩性多样,裂缝与测井参数间映射关系复杂,裂缝测井识别多解性强,准确率低。针对以上问题,本文对长短记忆神经网络算法(LSTM)进行改进用于深层潜山地层裂缝测井识别,在双层LSTM之间增加Dropout层,通过正则化防止过拟合,引入采用高斯核函数的最小二乘支持向量机(LSSVM)将LSTM中的Dense层和用于分类的Softmax函数进行替换,直接对LSTM层所提取的特征成分进行分类预测,在保留了长短记忆神经网络算法对测井曲线的序列性学习优势基础上,有效提升了分类预测效率和准确性,避免了裂缝特征信息的丢失以及对小样本训练数据的过度拟合,增强了算法的快速收敛能力。结果显示,测试集准确率达91.56%,识别准确率高于支持向量机和标准长短记忆神经网络模型,为深层复杂岩性基岩潜山储层裂缝识别提供了高效解决方案。

关键词: 深层, 基岩潜山, 改进长短记忆神经网络, 裂缝识别

Abstract:

The deep tight bedrock buried hills in the Central Uplift of the Liaohe Depression host fractured hydrocarbon reservoirs with significant resource potential. However, challenges arise from great burial depths, diverse lithologies, complex nonlinear relationships between fractures and logging parameters, and strong multi-solution ambiguity in fracture identification via conventional logging—resulting in low prediction accuracy. To address these challenges, the modified Long Short-Term Memory (LSTM) neural network algorithm is developed for fracture identification in deep buried hill formations. In the new algorithm, the Dropout layer is inserted between dual LSTM layers and regularization is used to mitigate overfitting; the Dense layer and the Softmax function used for classification in LSTM are replaced by Least Squares Support Vector Machines (LSSVM) with Gaussian kernel functions. Classification prediction based on features extracted by the LSTM layers can be directly approached. The proposed algorithm preserves LSTM’s sequential learning advantage for logging curves while significantly improving classification efficiency and accuracy. It effectively prevents loss of fracture-feature information and overfitting on limited training samples, while accelerating algorithm convergence. Validation results demonstrate a test-set accuracy of 91.56%, outperforming both Support Vector Machine (SVM) and standard LSTM models. This provides an efficient methodology for fracture identification in deep, complex-lithology bedrock buried hill reservoirs.

Key words: deep layer, base rock buried hill, Improved Long Short-Term Memory (LSTM) neural network, fracture identification

中图分类号:

张涛, 李艳萍, 李泽凯, 刘东成, 王静. 基于改进长短记忆神经网络的深层致密储层裂缝测井识别[J]. 地学前缘, 2025, 32(5): 456-465.

ZHANG Tao, LI Yanping, LI Zekai, LIU Dongcheng, WANG Jing. Fractures identification of deep tight reservoir with well logging based on Improved Long Short-Term Memory neural network[J]. Earth Science Frontiers, 2025, 32(5): 456-465.

图/表 15

图1 研究区构造位置图 (左:辽河坳陷构造概况;右:研究区基底太古宇顶面断裂系统图)

Fig.1 Structural location map of study area (left: structural outline map of Liaohe Depression; right: fault system map of the Archean basement top in the study area)

图2 研究区基岩潜山岩心裂缝类型统计

Fig.2 Bedrock buried hill core fracture type statistics

图3 不同岩性岩心裂缝发育类型统计

Fig.3 Fracture types of different lithologies

图4 LSTM“神经模块”结构图[38]

Fig.4 LSTM “neural module” structure[38]

图5 LSSVM-LSTM模型结构示意图

Fig.5 LSSVM-LSTM model structure diagram

图6 不同类型裂缝测井响应特征对比图

Fig.6 Comparison of logging response characteristics of different types of fractures

表1 裂缝测井响应特征表

Table 1 Logging response characteristics of fractures

图7 不同测井参数特征重要性得分统计图

Fig.7 Feature importance scores map of different logging parameters

图8 特征个数与分类准确率关系图

Fig.8 Relationship between the number of input features and classification

图9 LSSVM- LSTM方法驱动的多参数测井裂缝识别技术流程

Fig.9 Technical workflow of multi-parameter fracture logging identification driven by LSSVM- LSTM algorithm

表2 LSSVM- LSTM超参数数值表

Table 2 Hyperparameter values of LSSVM- LSTM model

参数类型	keep_prob 值	LSTM_1 单元数	LSTM_2 单元数	学习率	核参数	惩罚因子
参数值	0.36	132	64	0.008	112.316	32.653

图10 LSSVM- LSTM模型训练过程图

Fig.10 Training process of LSSVM- LSTM model

图11 LSSVM- LSTM模型验证集及测试集测试结果(a.验证集结果;b.测试集结果)

Fig.11 Validation and test sets performances of the LSSVM- LSTM model (a. validation set results; b. test set results)

图12 ZG1井裂缝识别成果图 1—代表有效裂缝;0—代表无裂缝或无效裂缝。

Fig.12 Fracture prediction results of ZG1

表3 LSSVM、LSTM、LSSVM- LSTM模型F1得分与准确率统计表

Table 3 F1-score and accuracy statistics of LSSVM, LSTM and LSSVM-LSTM models

评价项		LSSVM准确率/%			LSTM准确率/%			LSSVM-LSTM准确率/%
评价项		训练	验证	测试	训练	验证	测试	训练	验证	测试
F1值	有效裂缝	92.21	91.33	90.28	94.27	92.86	92.33	95.35	94.81	94.28
F1值	无效裂缝	75.47	71.81	68.61	80.28	76.21	74.91	83.83	81.11	78.93
样本集总体准确率		90.59	90.01	89.95	93.45	91.16	90.52	94.67	93.33	91.56

参考文献 40

[1]	胡英杰, 蔡国钢, 刘兴周, 等. 辽河坳陷东部滩海深层: 超深层天然气勘探重大发现及意义[J]. 中国石油勘探, 2023, 28(2): 24-34.
[2]	朱毅秀, 王欢, 单俊峰, 等. 辽河坳陷茨榆坨潜山太古界基岩储层岩性和储集空间特征[J]. 石油与天然气地质, 2018, 39(6): 1225-1236.
[3]	刘宝鸿, 郭彦民, 田志, 等. 利用地震相控反演预测辽河坳陷兴隆台潜山中生界致密砂砾岩优质储层[J]. 石油地球物理勘探, 2022, 57(增刊2): 100-109.
[4]	李毓. 储层裂缝的测井识别及其地质建模研究[J]. 测井技术, 2009, 33(6): 575-578.
[5]	吴琼, 林冬萍, 于春燕, 等. 新立油田低渗透油层裂缝测井识别方法[J]. 大庆石油地质与开发, 2007, 26(2): 112-115.
[6]	张凤生, 司马立强, 赵冉, 等. 塔河油田储层裂缝测井识别和有效性评价[J]. 测井技术, 2012, 36(3): 261-266.
[7]	廖海博, 刘红岐, 崔云江, 等. 基于常规测井的变质岩裂缝识别方法研究[J]. 海洋地质与第四纪地质, 2023, 43(4): 189-198.
[8]	曾联波, 柯式镇, 刘洋. 低渗透储层裂缝研究方法[M]. 北京: 石油工业出版社, 2010: 38-57.
[9]	孙建孟, 刘蓉, 梅基席, 等. 青海柴西地区常规测井裂缝识别方法[J]. 测井技术, 1999, 23(4): 268-275.
[10]	潘保芝, 刘文斌, 张丽华, 等. 一种提高储层裂缝识别准确度的方法[J]. 吉林大学学报(地球科学版), 2018, 48(1): 298-306.
[11]	杜书恒, 师永民, 关平. 松辽盆地扶余低渗非均质储层压裂缝定量预测[J]. 地学前缘, 2017, 24(6): 381-389.
[12]	钟高润, 张小莉, 杨振, 等. 鄂尔多斯盆地定边—志丹地区延长组致密砂岩储层裂缝测井识别[J]. 地球物理学进展, 2021, 36(4): 1669-1675.
[13]	DONG S, ZENG L, LYU W, et al. Fracture identification and evaluation using conventional logs in tight sandstones: a case study in the Ordos Basin, China[J]. Energy Geoscience, 2020, 1: 115-123.
[14]	汪明锐, 张冲, 聂昕, 等. 基于测井曲线重构的致密砂岩储层裂缝识别[J]. 西安石油大学学报(自然科学版), 2022, 37(5): 10-20.
[15]	赵星, 王贵文, 李亚锋, 等. 柴达木盆地英雄岭页岩油储层天然裂缝发育特征与测井识别[J]. 天然气地球科学, 2025, 36(4): 713-733.
[16]	LV H, GUO J, GU B, et al. A fracture identification method using fractal dimension and singular spectrum entropy in array acoustic logging: Insights from the Asmari Formation in the M oilfield, Iraq[J]. Journal of Applied Geophysics, 2025, 233: 105631.
[17]	王杰, 胡晨光, 潘勇利, 等. 吉华1地区潜山变质岩储层裂缝发育特征及综合评价[J]. 地质科学, 2022, 57(2): 463-477.
[18]	赖锦, 王贵文, 孙思勉, 等. 致密砂岩储层裂缝测井识别评价方法研究进展[J]. 地球物理学进展, 2015, 30(4): 1712-1724.
[19]	刘浩杰, 盖姗姗, 于文政, 等. 准噶尔盆地永进油田超深层致密砂岩裂缝测井识别[J]. 测井技术, 2022, 46(5): 592-598.
[20]	梁利喜, 许强, 刘向君. 基于人工神经网络实现裂缝性储集层的测井识别[J]. 新疆石油天然气, 2005, 1(3): 51-53.
[21]	赖富强, 夏炜旭, 龚大建, 等. 基于小波高频属性的泥页岩裂缝测井识别方法研究[J]. 地球物理学进展, 2020, 35(1): 124-131.
[22]	李伟, 王民, 肖佃师, 等. 天然裂缝发育特征及智能化识别方法: 以四川盆地川西坳陷上三叠统须家河组为例[J]. 油气藏评价与开发, 2025, 15(3): 443-454.
[23]	陆国青, 董少群, 黄立良, 等. 准噶尔盆地玛湖凹陷风城组陆相页岩油储层测井裂缝智能识别[J]. 地球科学, 2023, 48(7): 2690-2702.
[24]	董少群, 曾联波, 冀春秋, 等. 超深层致密砂岩裂缝测井识别深度核方法[J]. 地学前缘, 2024, 31(5): 166-176.
[25]	WANG Z, CAI Y, LIU D, et al. Characterization of natural fracture development in coal reservoirs using logging machine learning inversion, well test data and simulated geostress analyses[J]. Engineering Geology, 2024, 341: 107696.
[26]	郑军, 刘鸿博, 周文, 等. 阿曼五区块Daleel油田储层裂缝识别方法研究[J]. 测井技术, 2010, 34(3): 251-256.
[27]	TABASI S, TEHRANI PS, RAJABI M. Optimized machine learning models for natural fractures prediction using conventional well logs[J]. Fuel, 2022, 326: 124952.
[28]	HOCHREITER S, SCHMIDHUBER J. Long Short-Term Memory[J]. Neural Computation, 1997, 9(8): 1735-1780.
[29]	杨艺, 赵惊涛, 付国强. 基于PSO-LSTM模型的地热储层温度预测研究[J]. 矿业科学学报, 2024, 9(4): 538-548.
[30]	SHERSTINSKY A. Fundamentals of Recurrent Neural Network (RNN) and Long Short-Term Memory (LSTM) network[J]. Physica D: Nonlinear Phenomena, 2020, 404: 132306.
[31]	贺书洲, 程超, 谌丽, 等. 物模-数模双驱下的细粒沉积岩储层参数预测方法[J]. 地球物理学报, 2025, 68(7): 2712-2735.
[32]	程超, 李培彦, 陈雁, 等. 基于机器学习的储层测井评价研究进展[J]. 地球物理学进展, 2022, 37(1): 164-177.
[33]	安鹏, 曹丹平, 赵宝银, 等. 基于LSTM循环神经网络的储层物性参数预测方法研究[J]. 地球物理学进展, 2019, 34(5): 1849-1858.
[34]	张凤博, 马雪玲, 董珍珍, 等. 基于CNN和LSTM的机器学习模型在测井岩性识别的应用[J]. 西安石油大学学报(自然科学版), 2024, 39(5): 96-103.
[35]	QIAO L, XIN H C, XU Z M. Gas production prediction using AM-BiLSTM model optimized by Whale Optimization Algorithm[J]. Applied Geophysics, 2023, 20(4): 499-506.
[36]	陈波, 何贞铭, 李晓光, 等. 辽河盆地中央凸起带油气聚集规律[J]. 勘探家, 1998, 3(4): 32-36.
[37]	邱锡鹏. 神经网络与深度学习[M]. 北京: 机械工业出版社, 2020.
[38]	HAO W, SUN X, WANG C, et al. A hybrid EMD-LSTM model for non-stationary wave prediction in offshore China[J]. Ocean Engineering, 2022, 246: 110566.
[39]	ZHANG A, LIPTON Z C, LI M. Dive into Deep Learning[M]. New York: Cambridge University Press, 2020: 397-404.
[40]	司学强, 彭博, 庞志超, 等. 储集层多尺度裂缝特征及控制因素: 以准噶尔盆地南缘侏罗系—白垩系为例[J]. 中国矿业大学学报, 2022, 51(4): 731-741.