摘要
传统微电阻率成像(电成像)测井岩性识别主要依靠人工辨识,识别结果往往受经验和主观因素影响,同时还存在岩性表征困难等问题。为此,提出了一种基于电成像图像形状和颜色相结合的多维度岩性识别方法。利用Filtersim算法对电成像空白条带进行补全,并对补全后的数据进行K-means++聚类算法像素聚类,实现对裂缝、溶洞等弱噪声信息的标记,避免给颜色聚类引入噪声。借鉴异常检测思想,利用损失异常对强噪声样本进行筛选。根据电成像的纹理结构和电阻率响应特性,分别把电成像数据集解耦为形状集和颜色集,在此基础上提出一种形状和颜色相结合的电成像岩性识别模型;针对不同地质构造(块状、层状、纹层状)的电成像形状特征,引入标签精炼方法解决硬标签问题,建立Resnet-50网络模型实现形状特征的自动识别;针对不同电阻率响应(泥岩、灰质泥岩、砂质泥岩)的电成像颜色特征,利用K-means++聚类算法筛选出数据集总体分布的聚类中心,实现电成像颜色的快速分类。结合形状特征分类和颜色特征分类结果,识别出电成像岩性种类。应用新方法对济阳坳陷页岩油储层电成像图像进行岩性识别实验,结果表明:岩性识别准确率达83.5%,具有较高的识别精度。提出的基于电成像测井的多维度岩性识别方法,可为测井解释中的岩性识别提供良好的算法支撑。
The traditional microresistivity image logging lithology identification mainly depends on manual identification,and the identification results often affected by manual experience and subjective factors which leads to some issues such as difficulty in lithology characterization.In this paper,a multi-dimensional electrical imaging identification method based on the combination of shape and color is proposed for lithology identification.First,Filtersim algorithm is employed to fill the blank strip of electrical imaging,and K-means + + clustering in pixelwise is performed on the filled data to mark the weak noise such as cracks and karst caves,so as to avoid introducing noise into color clustering.Then,loss anomaly is used to screen strong noise samples.According to the texture structure and resistivity response characteristics of electro-imaging,the electrical imaging dataset is decoupled into shape set and color set,respectively.Then,a shape and color combined electrical imaging identification model is proposed.To solve the issue of hard labeling by introducing label refining method,Resnet-50 network is established to realize automatic recognition of shape features for different geological structures(massive,layered and laminated).For the electrical imaging color features of different resistivity responses(mudstone,calcareous mudstone and sandy mudstone),K-means + + algorithm is used to screen out the clustering centers of the overall distribution of the data set to achieve fast classification of the electro-imaging colors.Finally,combined with the results of shape classification and color classification,the types of electrical imaging lithology are identified.Lithology recognition experiment is carried out on the electrical imaging image of shale oil reservoir in Jiyang depression.The results show that the recognition accuracy is 83.5%,which has high recognition accuracy.The method can provide fine algorithm support for log interpretation of lithology recognition.
作者
刘娟
闵宣霖
漆仲黎
易军
赖富强
周伟
LIU Juan;MIN Xuanlin;QI Zhongli;YI Jun;LAI Fuqiang;ZHOU Wei(CCTEG Chongqing Research Institute,Chongqing 400037,China;School of Intelligent Technology and Engineering,Chongqing University of Science and Technology,Chongqing 401331,China;School of Petroleum Engineering,Chongqing University of Science and Technology,Chongqing 401331,China)
出处
《测井技术》
CAS
2023年第6期726-735,共10页
Well Logging Technology
基金
重庆市教育委员会科学技术研究项目“深层碳酸盐岩储层缝洞图像信息深度挖掘及精细评价”(KJZDK202301508)。
关键词
电成像
岩性识别
卷积网络
聚类分析
电成像形状特征
电阻率响应特性
济阳坳陷
electrical imaging
lithology identification
convolutional neural network
clustering analysis
electrical imaging shape feature
resistivity response characteristic
Jiyang depression
