FLU-NET: A DEEP FULLY CONVOLUTIONAL NEURAL NETWORK FOR SHALE RESERVOIR MICRO-PORE CHARACTERIZATION
-
摘要:
页岩孔隙是页岩气储集的主要空间，孔隙的形状、大小、连通性与发育程度很大程度上决定了页岩储集层的储集性能，因此，页岩气开采首先需要对其孔隙有充分的认识。基于阈值分割法获取页岩孔隙结构参数是目前页岩微观结构表征的一种重要手段，但是受扫描电镜图像灰度分布差异的影响，该方法需要逐一修改图像的最佳分割阈值以达到最好的孔隙分割效果，且阈值分割方法无法直接划分孔隙类别，这给后续的页岩微观结构定量表征带来了麻烦。为了实现页岩孔隙的智能识别和分类，笔者设计基于像素级语义分割的深度全卷积神经网络FLU-net，对页岩孔隙识别并分类为有机孔、无机孔（粒内孔、粒间孔）及裂缝，并结合孔隙尺度分类统计方法，分析不同类型孔隙发育数量、孔径大小、孔隙度等参数，实现页岩储层微观孔隙结构的自动化定量表征。以重庆渝西区块足201井区和四川盆地威远地区威204井区的页岩扫描电镜图像为研究对象，在对1 600幅页岩扫描电镜图像原始数据进行人工标注并划分数据集后，使用FLU-net进行孔隙识别，结果表明，本方法具有较高的准确率，同时自动化程度和泛化能力均高于传统预测方法。因此，扫描电镜与基于深度学习的语义分割模型结合是定量研究页岩微观结构表征的有效手段。
Abstract:Shale pores are the main space of shale gas reservoir. The shape, size, connectivity and development degree of pores may determine the performance of reservoir to a great extent. A full understanding of shale pores is critical before exploitation. At present, obtaining structure parameters of shale pores based on threshold segmentation method plays an important role in shale microstructure characterization. However, due to the difference in gray distribution of SEM images, this method requires modification of the best segmentation threshold individually to achieve the best pore segmentation effect, and the threshold segmentation method cannot classify the pore directly. It may cause trouble to the subsequent quantitative characterization of shale microstructure. In order to realize the intelligent recognition and classification of shale pores, a Deep Convolution Neural Network FLU-net based on pixel-level semantic segmentation is designed in this paper. The pores of shale are identified and classified into organic pores, inorganic pores (intragranular pores and intergranular pores) and fractures. Combined with the statistical method of pore scale classification, the number of pores, pore size, porosity and other parameters for different types of pores are analyzed, and thus automatic quantitative characterization of micro pore structure of shale reservoir is realized. This paper takes the Scanning Electron Microscope (SEM) images of shale from the Zu-201 well area, Yuxi block in Chongqing and the Wei-204 well area, Weiyuan area in Sichuan Basin as the research objects. After manually labeling and dividing the original dataset of 1600 SEM images of shale, FLU-net is used for pore recognition. The results show that this method not only keeps high accuracy, but also has higher automation and generalization ability than traditional prediction methods. Therefore, the combination of SEM and semantic segmentation model based on Deep Learning is an effective mean for quantitative study of shale microstructure characterization.
-
Key words:
- shale pores /
- shale reservoir /
- semantic segmentation /
- pore identification /
- deep learning
-
-
表 1 威204井和足201井图像样本中各类别孔隙的孔隙度
Table 1. Porosity of all kinds of pores in image samples from Well Wei-204 and Well Zu-201
所属井 样本编号 有机质孔 晶（粒）间孔 晶（粒）间孔 微裂缝 威204 1号 0.05% 1.86% 1.56% 8.87% 威204 2号 0.71% 0.51% 1.65% 1.77% 威204 3号 5.90% 1.88% 0.29% 0.00% 
威204 平均值 3.45% 1.92% 0.88% 0.63% 足201 1号 0.42% 0.11% 0.07% 0.00% 足201 2号 0.45% 0.00% 0.06% 4.92% 足201 3号 0.72% 0.13% 0.07% 0.22% 
足201 平均值 1.02% 0.51% 0.33% 0.44% 
下载: 导出CSV
表 2 5种网络的验证集评价指标
Table 2. The ACC and mIoU indexes of the five networks on the validation set
模型 ACC mIoU U-net 0.9855 0.4381 deeperU-net 0.9854 0.4284 U-ent+Dropout 0.9850 0.4264 U-net+Focal Loss 0.9864 0.4588 FLU-net 0.9853 0.4726
下载: 导出CSV
表 3 U-net和FLU-net的4种微观孔隙度预测结果与真实孔隙度的均方根误差（RMSE）
Table 3. The RMSE of the predicted porosity by U-net and FLU-net and the real porosity of the four kinds of pores
微裂缝 晶（粒）间孔 晶（粒）内孔 有机质孔 总孔隙 U-net 0.005099 0.008179 0.00500 0.006296 0.008250 FLU-net 0.004260 0.004707 0.003918 0.005116 0.006283
下载: 导出CSV
表 4 通过不同语义分割模型定量表征的孔隙特征
Table 4. The quantitatively characterized porosity characters by different semantic segmentation models
人工标签 U-net FLU-net 最大孔径长度 
平均孔径长度 
孔隙总面积 
最大孔隙面积 
平均孔隙面积 
平均形状系数 
周长面积法分形维数 
下载: 导出CSV
-
[1] ADNAN A,REZA R,LIONEL E,et al. Comparisons of pore size distribution:a case from the Western Australian gas shale formations[J]. Journal of Unconventional Oil and Gas Resources,2014,8(12):1-13. doi: 10.1016/j.juogr.2014.06.002
[2] 赵日新,卢双舫,薛海涛,等. 扫描电镜分析参数对定量评价页岩微观孔隙的影响[J]. 石油与天然气地质,2019, 40(5):1141-1154. doi: 10.11743/ogg20190520
[3] 张林晔,李钜源,李政,等. 北美页岩油气研究进展及对中国陆相页岩油气勘探的思考[J]. 地球科学进展,2014, 29(6):700-711. doi: 10.11867/j.issn.1001-8166.2014.06.0700
[4] 李俊乾,卢双舫,张婕,等. 页岩油吸附与游离定量评价模型及微观赋存机制[J]. 石油与天然气地质,2019,40(3):583-592. doi: 10.11743/ogg20190313
[5] 孙寅森,郭少斌. 基于图像分析技术的页岩微观孔隙特征定性及定量表征[J]. 地球科学进展,2016,31(7):751-763. doi: 10.11867/j.issn.1001-8166.2016.07.0751.
[6] 王羽,金婵,汪丽华,等. 应用氩离子抛光-扫描电镜方法研究四川九老洞组页岩微观孔隙特征[J]. 岩矿测试,2015(3):278-285.
[7] 白名岗,夏响华,张聪,等. 场发射扫描电镜及PerGeos系统在安页1井龙马溪组页岩有机质孔隙研究中的联合应用[J]. 岩矿测试,2018,37(3):225-234.
[8] 武瑾,梁峰,吝文,等. 渝东北地区龙马溪组页岩储层微观孔隙结构特征[J]. 成都理工大学学报(自然科学版),2016,43(3):308-319.
[9] 王欣,齐梅,李武广,等. 基于分形理论的页岩储层微观孔隙结构评价[J]. 天然气地球科学,2015,26(4):754-759.
[10] 焦淑静,张慧,薛东川,等. 泥页岩有机显微组分的扫描电镜形貌特征及识别方法[J]. 电子显微学报,2018,37(2):137-144. doi: 10.3969/j.issn.1000-6281.2018.02.006
[11] 姜振学,唐相路,李卓,等. 川东南地区龙马溪组页岩孔隙结构全孔径表征及其对含气性的控制[J]. 地学前缘,2016,23(2):126-134.
[12] YU L,ZHU Y M. Comparison of pore characteristics in the coal and shale reservoirs of Taiyuan Formation,Qinshui Basin,China[J]. International Journal of Coal Science & Technology,2016,3(3):1-9.
[13] 傅志强,贺翠翠,刘锡贝,等. 扫描电镜测量纳米尺度的影响因素[J]. 电子显微学报,2012,31(3):226-229. doi: 10.3969/j.issn.1000-6281.2012.03.006
[14] 吴一全,孟天亮,吴诗婳. 图像阈值分割方法研究进展20年(1994—2014)[J]. 数据采集与处理,2015,30(1):1-23.
[15] 俞雨溪,王宗秀,程明,等. 页岩微观结构灰度图像的标准化方法[J]. 煤炭学报,2019,44(7):2178-2187.
[16] 于炳松. 页岩气储层孔隙分类与表征[J]. 地学前缘,2013,20(4):211-220.
[17] 杨峰,宁正福,胡昌蓬,等. 页岩储层微观孔隙结构特征[J]. 石油学报,2013,34(2):301-311. doi: 10.7623/syxb201302012
[18] DENG L,YU D. Deep learning:methods and applications[J]. Foundations and Trends in Signal Processing,2013,7(3/4):197-387.
[19] 刘建伟,刘媛,罗雄麟. 深度学习研究进展[J]. 计算机应用研究,2014,31(7):1921-1930. doi: 10.3969/j.issn.1001-3695.2014.07.001
[20] XING H,ZHANG G,SHANG M. Deep learning[J]. International Journal of Semantic Computing,2016,10(3):417-439. doi: 10.1142/S1793351X16500045
[21] 林年添,张栋,张凯,等. 地震油气储层的小样本卷积神经网络学习与预测[J]. 地球物理学报,2018,61(10):4110-4125. doi: 10.6038/cjg2018J0775
[22] 张东晓,陈云天,孟晋. 基于循环神经网络的测井曲线生成方法[J]. 石油勘探与开发,2018,45(4):598-607.
[23] 占文枢,伦增珉,陈洪刚,等. 基于像素及梯度域双层深度卷积神经网络的页岩图像超分辨率重建[J]. 科学技术与工程,2018,18(3):85-91. doi: 10.3969/j.issn.1671-1815.2018.03.014
[24] RONNEBERGER O, FISCHER P, BROX T. U-Net: convolutional networks for biomedical image segmentation[J]. Springer, Cham, 2015, 234-241.
[25] 赵子元. 威204井区页岩气井早期产能评价[D]. 重庆: 重庆科技学院, 2018.
[26] 梁峰. 中上扬子地区五峰组—龙马溪组页岩气富集模式及有利区优选评价[D]. 徐州: 中国矿业大学, 2018.
[27] 马新华,谢军. 川南地区页岩气勘探开发进展及发展前景[J]. 石油勘探与开发,2018,45(1):161-169.
[28] 姜枫,顾庆,郝慧珍,等. 基于内容的图像分割方法综述[J]. 软件学报,2017,28(1):160-183.
[29] 魏云超,赵耀. 基于DCNN的图像语义分割综述[J]. 北京交通大学学报,2016,40(4):82-91. doi: 10.11860/j.issn.1673-0291.2016.04.013
[30] KRIZHEVSKY A, SUTSKEVER I, HINTON G E. Imagenet classification with deep convolutional neural networks[C]//International Conference on Neural Information Processing Systems. 2012, 25: 1097-1105.
[31] SHELHAMER E,LONG J,DARRELL T. Fully convolutional networks for semantic segmentation[J]. IEEE Transactions on Pattern Analysis & Machine Intelligence,2014,39(4):640-651.
[32] HE K, ZHANG X, REN S, et al. Deep residual learning for image recognition[C]//Proceedings of the IEEE conference on computer vision and pattern recognition. Las Vegas, 2016: 770-778.
[33] SRIVASTAVA N,HINTON G,Krizhevsky A,et al. Dropout:a simple way to prevent neural networks from overfitting[J]. Journal of Machine Learning Research,2014,15(1):1929-1958.
[34] LIN T Y, GOYAL P, GIRSHICK R, et al. Focal Loss for Dense Object Detection[C]//Proceedings of the IEEE international conference on computer vision, Honolulu. 2017: 2980-2988.
[35] CHOU W C, BEEREl P A, Ginosar R, et al. Average-case optimized technology mapping of one-hot domino circuits[C]// Proceedings Fourth International Symposium on Advanced Research in Asynchronous Circuits and Systems, IEEE. San Diego, 1998: 80-91, doi: 10.1109/ASYNC.1998.666496.
[36] GARCIA-GARCIA A, ORTS-ESCOLANO S, OPREA S, et al. A review on deep learning techniques applied to semantic segmentation[J]. arXiv preprint: 1704.06857, 2017: 1-22.
[37] CHAI T,DRAXLER R. Root mean square error (RMSE) or mean absolute error (MAE)?:arguments against avoiding RMSE in the literature[J]. Geoscientific Model Development,2014,7(3):1247-1250. doi: 10.5194/gmd-7-1247-2014
[38] 焦堃,姚素平,吴浩,等. 页岩气储层孔隙系统表征方法研究进展[J]. 高校地质学报,2014,20(1):151-161. doi: 10.3969/j.issn.1006-7493.2014.01.015
-
图(7)
表(4)
计量
- 文章访问数: 2298
- PDF下载数: 549
- 施引文献: 0