基于LSTM神经网络的随钻方位电磁波测井数据反演

康正明; 秦浩杰; 张意; 李新; 倪卫宁; 李丰波

doi:10.11911/syztjs.2023047

基于LSTM神经网络的随钻方位电磁波测井数据反演

康正明^1,,
秦浩杰¹,
张意^2,*, ,,
李新^{3, 4},
倪卫宁^{3, 4},
李丰波^{3, 4}

1.
西安石油大学电子工程学院, 陕西西安 710065
2.
中煤科工西安研究院（集团）有限公司, 陕西西安 710077
3.
页岩油气富集机理与有效开发国家重点实验室, 北京 102206
4.
中石化石油工程技术研究院有限公司, 北京 102206

基金项目: 国家自然科学基金企业创新发展联合基金项目“海相深层油气富集机理与关键工程技术基础研究”（编号：U19B6003）、中国博士后科学基金项目“煤岩层界面及低阻异常体随钻方位电磁波探测方法研究”（编号：2022M711442）、陕西省重点研发计划项目“煤矿井下方位电磁波探测技术与仪器研究”（编号：2023-YBGY-111）、陕西省教育厅重点科学研究计划项目“基于随钻电成像测井的页岩气储层裂缝参数计算模型研究”（编号：22JY053）和西安石油大学研究生创新与实践能力培养计划（编号：YCS22214245）联合资助

详细信息

作者简介:
康正明（1989—），男，陕西靖边人，2014年毕业于西安石油大学勘查技术与工程专业，2019年获中国石油大学（北京）地质资源与地质工程专业博士学位，讲师，主要从事电法测井理论方法研究。E-mail:190720@xsyu.edu.cn。

通讯作者:
张意，yizhang86@163.com

中图分类号: P631.8⁺13
计量
- 文章访问数: 445
- HTML全文浏览量: 148
- PDF下载量: 109
出版历程
- 收稿日期: 2022-07-24
- 修回日期: 2023-03-15
- 网络出版日期: 2023-04-02
- 刊出日期: 2023-03-24

Data Inversion of Azimuthal Electromagnetic Wave Logging While Drilling Based on LSTM Neural Network

KANG Zhengming^1,,
QIN Haojie¹,
ZHANG Yi^2,*, ,,
LI Xin^{3, 4},
NI Weining^{3, 4},
LI Fengbo^{3, 4}

1.
School of Electronic Engineering, Xi’an Shiyou University, Xi’an, Shaanxi, 710065, China
2.
Xi’an Research Institute Co. Ltd., China Coal Technology and Engineering Group Corp., Xi’an, Shaanxi, 710077, China
3.
State Key Laboratory of Shale Oil and Gas Enrichment Mechanisms and Effective Development, Beijing, 102206, China
4.
Sinopec Research Institute of Petroleum Engineering Co., Ltd, Beijing, 102206, China

摘要

摘要:
随钻方位电磁波测井仪器在地质导向和储层评价等方面具有重要作用，但其测量响应不具有直观性，需要用反演方法获得地层信息，高斯–牛顿法、随机反演算法等传统反演方法计算速度较慢，难以满足实时反演的要求。为此，提出了一种基于长短期记忆人工神经网络（LSTM）的新反演方法，用于求取地层电阻率。首先，基于广义反射系数法建立正演算法，完成样本集的制作；然后，搭建LSTM神经网络模型，基于样本集进行训练和测试，通过遍历的方法优选出合适的网络参数；最后，在测试集上完成电阻率的反演，将反演电阻率与正演电阻率进行对比，对比反演所需时间和相对误差，并在测试集中加入白噪声验证了模型的抗噪能力。研究结果表明，模型能够准确快速地反演地层电阻率信息，能够满足对含有噪声数据的反演需要，具有较好的鲁棒性。此反演方法为测井资料处理提供了新的思路和方向。
- LSTM神经网络 /
- 电阻率反演 /
- 随钻方位电磁波测井 /
- 正演计算 /
- 地质导向
Abstract:
Azimuthal electromagnetic wave logging while drilling (LWD) tool plays an important role in geosteering and reservoir evaluation, but its measurement response is not intuitive. So inversion method is needed to obtain formation information. Traditional inversion methods (i.e., Gauss-Newton method, random inversion method, etc.) are difficult to meet the requirements of real-time inversion due to the slow calculation speed. In this paper, a new inversion method based on a long and short-term memory (LSTM) artificial neural network was proposed to obtain formation resistivity. Firstly, the forward algorithm was established based on the method of generalized reflection coefficient to produce the sample set. Then, the LSTM neural network model was built, and it was trained and tested on the sample set. The appropriate network parameters were optimized by the traversal method. Finally, the resistivity inversion was completed on the test set. The inverted resistivity was compared with the forward resistivity, and the inversion time and relative error were compared as well. Meanwhile, the anti-noise property of the model is verified by adding white noise to the test set. The results show that the model can accurately and rapidly invert formation resistivity and can invert data containing noise, indicating that the model has good robustness. This inversion method can provide a new idea and direction for logging data processing.
- LSTM neural network /
- resistivity inversion /
- azimuthal electromagnetic wave LWD /
- forward calculation /
- geosteering

HTML全文

图 1 方位电磁波测井常用线圈结构

Figure 1. Commonly used coil structure in azimuthal electromagnetic wave logging

下载: 全尺寸图片幻灯片

图 2 电阻率反演流程

Figure 2. Inversion flow of resistivity

下载: 全尺寸图片幻灯片

图 3 LSTM网络架构

Figure 3. Architecture of LSTM network

下载: 全尺寸图片幻灯片

图 4 水平层状多层地层模型

Figure 4. Horizontal stratified formation model with multiple layers

下载: 全尺寸图片幻灯片

图 5 批尺寸为32时不同学习率下的损失函数曲线对比

Figure 5. Comparison of loss function curves under different learning rates when batch size is 32

下载: 全尺寸图片幻灯片

图 6 批尺寸为64时不同学习率下的损失函数曲线对比

Figure 6. Comparison of loss function curves under different learning rates when batch size is 64

下载: 全尺寸图片幻灯片

图 7 批尺寸为128时不同学习率下的损失函数曲线对比

Figure 7. Comparison of loss function curves under different learning rates when batch size is 128

下载: 全尺寸图片幻灯片

图 8 批尺寸为256时不同学习率下的损失函数曲线对比

Figure 8. Comparison of loss function curves under different learning rates when batch size is 256

下载: 全尺寸图片幻灯片

图 9 三层地层模型反演结果

Figure 9. Inversion results of three-layer formation model

下载: 全尺寸图片幻灯片

图 10 四层地层模型反演结果

Figure 10. Inversion results of four-layer formation model

下载: 全尺寸图片幻灯片

图 11 五层地层模型反演结果

Figure 11. Inversion results of five-layer formation model

下载: 全尺寸图片幻灯片

图 12 不同噪声强度下电阻率反演误差分布直方图

Figure 12. Histogram of resistivity inversion error distribution under different noise intensities

下载: 全尺寸图片幻灯片

表 1 不同批尺寸和学习率的损失误差

Table 1 Loss errors for different batch sizes and learning rates

η	训练集误差				测试集误差
η	n=32	n=64	n=128	n=256	n=32	n=64	n=128	n=256
0.000 5	0.011 0	0.012 4	0.012 5	0.014 1	0.009 3	0.007 5	0.011 1	0.011 7
0.001 0	0.009 8	0.010 9	0.014 3	0.012 1	0.008 7	0.007 5	0.011 9	0.010 6
0.002 0	0.011 0	0.010 1	0.010 8	0.011 6	0.008 8	0.007 2	0.010 4	0.010 7
0.004 0	0.011 3	0.010 7	0.011 1	0.013 1	0.009 1	0.007 7	0.010 9	0.010 8
0.006 0	0.010 8	0.011 2	0.011 6	0.013 3	0.008 9	0.008 3	0.011 2	0.011 6
0.008 0	0.012 5	0.012 3	0.012 5	0.011 9	0.009 0	0.008 1	0.010 4	0.014 0

下载: 导出CSV

表 2 电阻率反演相对误差

Table 2 Relative error of resistivity inversion

电阻率	相对误差，%	采样点数	百分比，%
R_h	<5	1 397 313	91.0
	≥5～<10	112 652	7.3
	≥10～<20	18 852	1.2
	≥20	7 183	0.5
R_v	<5	1 357 925	88.4
	≥5～<10	124 935	8.1
	≥10～<20	28 557	1.9
	≥20	24 583	1.6

下载: 导出CSV

表 3 不同方法反演时间比较

Table 3 Comparison of inversion time between different methods

地层模型层数	反演单个样本所需时间/s
地层模型层数	LSTM网络	监督下降法	Occam法
3	0.04～0.06	0.5～4.0	>120
5	0.04～0.06	0.5～4.0	>240

下载: 导出CSV

参考文献(21)

[1]	NIE Xiaochun, YUAN Ning, LIU C R. Simulation of LWD tool response using a fast integral equation method[J]. IEEE Transactions on Geoscience and Remote Sensing, 2010, 48(1): 72–81. doi: 10.1109/TGRS.2009.2027112
[2]	LEE H O, TEIXEIRA F L, SAN MARTIN L E, et al. Numerical modeling of eccentered LWD borehole sensors in dipping and fully anisotropic earth formations[J]. IEEE Transactions on Geoscience and Remote Sensing, 2012, 50(3): 727–735. doi: 10.1109/TGRS.2011.2162736
[3]	吴柏志,杨震,郭同政,等. 多尺度随钻方位电磁波测井系统响应特征研究[J]. 石油钻探技术,2022,50(6):7–13. doi: 10.11911/syztjs.2022107 WU Baizhi,YANG Zhen,GUO Tongzheng,et al. Response characteristics of logging while drilling system with multi-scale azimuthal electromagnetic waves[J]. Petroleum Drilling Techniques, 2022, 50(6): 7–13. doi: 10.11911/syztjs.2022107
[4]	刘天淋,岳喜洲,李国玉,等. 超深探测随钻电磁波测井地质信号特性研究[J]. 石油钻探技术,2022,50(6):41–48. doi: 10.11911/syztjs.2022110 LIU Tianlin, YUE Xizhou, LI Guoyu, et al. Study over the geo-signal properties of ultra-deep electromagnetic wave logging while drilling[J]. Petroleum Drilling Techniques, 2022, 50(6): 41–48. doi: 10.11911/syztjs.2022110
[5]	WANG Gongli, BARBER T, WU P, et al. Fast inversion of triaxial induction data in dipping crossbedded formations[J]. Geophysics, 2017, 82(2): D31–D45. doi: 10.1190/geo2015-0610.1
[6]	杨震,于其蛟,马清明. 基于拟牛顿法的随钻方位电磁波电阻率仪器响应实时反演与现场试验[J]. 石油钻探技术,2020,48(3):120–126. doi: 10.11911/syztjs.2020025 YANG Zhen, YU Qijiao, MA Qingming. Real time inversion and field test of LWD azimuthal electromagnetic waves based on quasi-newton method[J]. Petroleum Drilling Techniques, 2020, 48(3): 120–126. doi: 10.11911/syztjs.2020025
[7]	JIN Yuchen, SHEN Qiuyang, WU Xuqing, et al. A physics-driven deep-learning network for solving nonlinear inverse problems[J]. Petrophysics, 2020, 61(1): 86–98.
[8]	ZHU Gaoyang, GAO Muzhi, KONG Fanmin, et al. A fast inversion of induction logging data in anisotropic formation based on deep learning[J]. IEEE Geoscience and Remote Sensing Letters, 2020, 17(12): 2050–2054. doi: 10.1109/LGRS.2019.2961374
[9]	LECUN Y, BENGIO Y, HINTON G. Deep learning[J]. Nature, 2015, 521(7553): 436–444. doi: 10.1038/nature14539
[10]	ZHU Xiaoyu, LI Jincai, ZHU Min, et al. An evaporation duct height prediction method based on deep learning[J]. IEEE Geoscience and Remote Sensing Letters, 2018, 15(9): 1307–1311. doi: 10.1109/LGRS.2018.2842235
[11]	ARAYA-POLO M, JENNINGS J, ADLER A, et al. Deep-learning tomography[J]. The Leading Edge, 2018, 37(1): 58–66. doi: 10.1190/tle37010058.1
[12]	OH S, NOH K, YOON D, et al. Salt delineation from electromagnetic data using convolutional neural networks[J]. IEEE Geoscience and Remote Sensing Letters, 2019, 16(4): 519–523. doi: 10.1109/LGRS.2018.2877155
[13]	COLOMBO D, TURKOGLU E, LI Weichang, et al. Physics-driven deep-learning inversion with application to transient electromagnetics[J]. Geophysics, 2021, 86(3): E209–E224. doi: 10.1190/geo2020-0760.1
[14]	JIN Yuchen, WU Xuqing, CHEN Jiefu, et al. Using a physics-driven deep neural network to solve inverse problems for LWD azimuthal resistivity measurements[R]. SPWLA-2019-IIII, 2019.
[15]	HU Yanyan, GUO Rui, JIN Yuchen, et al. A supervised descent learning technique for solving directional electromagnetic logging-while-drilling inverse problems[J]. IEEE Transactions on Geo-science and Remote Sensing, 2020, 58(11): 8013–8025. doi: 10.1109/TGRS.2020.2986000
[16]	SHAHRIARI M, PARDO D, PICON A, et al. A deep learning approach to the inversion of borehole resistivity measurements[J]. Computational Geosciences, 2020, 24(3): 971–994. doi: 10.1007/s10596-019-09859-y
[17]	NOH K, TORRES-VERDÍN C, PARDO D. Real-time 2.5D inversion of LWD resistivity measurements using deep learning for geosteering applications across faulted formations[R]. SPWLA-2021 − 0104, 2021.
[18]	倪卫宁,张晓彬,万勇,等. 随钻方位电磁波电阻率测井仪分段组合线圈系设计[J]. 石油钻探技术,2017,45(2):115–120. NI Weining, ZHANG Xiaobin, WAN Yong, et al. The design of the coil system in LWD tools based on azimuthal electromagnetic-wave resistivity combined with sections[J]. Petroleum Drilling Techni-ques, 2017, 45(2): 115–120.
[19]	杨震,肖红兵,李翠. 随钻方位电磁波仪器测量精度对电阻率及界面预测影响分析[J]. 石油钻探技术,2017,45(4):115–120. doi: 10.11911/syztjs.201704020 YANG Zhen, XIAO Hongbing, LI Cui. Impacts of accuracy of azimuthal electromagnetic logging-while-drilling on resistivity and interface prediction[J]. Petroleum Drilling Techniques, 2017, 45(4): 115–120. doi: 10.11911/syztjs.201704020
[20]	杨震,文艺,肖红兵. 随钻方位电磁波仪器探测电阻率各向异性新方法[J]. 石油钻探技术,2016,44(3):115–120. doi: 10.11911/syztjs.201603021 YANG Zhen, WEN Yi, XIAO Hongbing. A new method of detecting while drilling resistivity anisotropy with azimuthal electromagnetic wave tools[J]. Petroleum Drilling Techniques, 2016, 44(3): 115–120. doi: 10.11911/syztjs.201603021
[21]	HOCHREITER S, SCHMIDHUBER J. Long short-term memory[J]. Neural Computation, 1997, 9(8): 1735–1780. doi: 10.1162/neco.1997.9.8.1735

施引文献(8)

期刊类型引用(6)

1.	董添奇，杨旭辉. 近钻头钻压和扭矩的高精度温度补偿方法. 石油机械. 2024(04): 71-79 . 百度学术
2.	孙养清，易先中，万继方，易军，吴霁薇，刁斌斌，陈志湘. 机械式复合冲击器的工作特性分析. 石油机械. 2024(06): 29-37+108 . 百度学术
3.	白彬珍，曾义金，芦鑫，张进双，陶亮. 钻头破岩能量与岩石自适应匹配提速技术. 石油钻探技术. 2023(03): 30-36 . 本站查看
4.	罗杰，柳军，李强. 深井钻柱纵-扭耦合下的黏滑振动特性分析. 石油机械. 2023(08): 139-147 . 百度学术
5.	汪伟，陈杰，张鹏翔，柳贡慧，李军，查春青. 旋转配流式扭转冲击钻具设计与分析. 断块油气田. 2023(06): 1047-1052 . 百度学术
6.	兰永飞，韩玉香，陈明勇，王虎. 扭力冲击器-螺杆复合钻进工艺应用实践. 钻探工程. 2023(S1): 399-404 . 百度学术