基于成像测井的孔缝智能分割与识别

Intelligent segmentation and recognition of pores and fractures based on imaging logging

樊永东 庞惠文 金衍 王汉青

FAN Yongdong PANG Huiwen JIN Yan WANG Hanqing

中国石油大学(北京) 人工智能学院 中国石油大学(北京) 油气资源与探测国家重点实验室 中国石油大学(北京) 理学院 中国石化石油勘探开发研究院

School of Artificial Intelligence, China University of Petroleum (Beijing), Beijing 102249, China State Key Laboratory of Oil & Gas Resources and Exploration, China University of Petroleum (Beijing), Beijing 102249, China College of Science, China University of Petroleum (Beijing), Beijing 102249, China Research Institute of Petroleum Exploration and Development, SINOPEC, Beijing 102206, China

在碳酸盐岩储层的钻探开发过程中，确定地层中的溶孔和裂缝所处区域及种类，对判断储层漏失通道和储存空间具有重要意义。借助图像识别技术，识别成像测井图像中裂缝与溶孔是当前研究的难点，该方法要求样本数据量较大，在小样本情况下识别效果较差，因此提出了通过图像分割提高样本质量，实现小样本情况下高准确度的孔缝识别。研究主要包括图像分割与图像识别两部分，图像分割以阈值分割为主，应用K均值聚类与遗传算法对阈值分割进行了逐步优化；图像识别主要是应用深度神经网络，基于图像分割后的高质量图像识别成像测井图像中的孔缝结构。研究结果表明，图像分割前后整体识别准确度由63.3%提高到90.0%。在模型的实际应用中，该模型成功识别了样本中包含的高导缝与溶蚀孔，通过图像分割提高图像质量，可以实现小样本高准确度的孔缝结构识别。

When drilling and developing carbonate rock reservoirs, determining the location and type of the dissolved pores and fractures in the formation is of great significance for judging the leakage channels and storage space in the reservoir. With the help of image recognition technology, recognizing the fractures and dissolved pores in imaging logging images is a difficult point in current research. Recognizing the images requires a large amount of sample data, and the recognition effect is poor in the case of small samples. Therefore, it is proposed to improve sample quality through image segmentation, so as to realize high-accuracy pore and fracture recognition in the case of small samples. The research mainly includes image segmentation and image recognition. Image segmentation is mainly based on threshold segmentation, that is, applying -means clustering and genetic algorithm to gradually optimize threshold segmentation. Image recognition mainly refers to deep neural network, which recognizes the pore and fracture structures in imaging log images based on the high-quality image after image segmentation. The research results show that the overall recognition accuracy increases from 63.3% to 90.0% before and after image segmentation. In the practical application of the model, the model successfully recognized the high conductivity fractures and the dissolved pores contained in the sample, and improved the image quality through image segmentation, which can realize the high-accuracy recognition of pore structure in small samples.

10.13639/j.odpt.2022.04.015