基于深度学习的龙马溪组页岩孔缝识别与参数计算

Identification and Parameter Calculation of Shale in Longmaxi Formation Pore Fractures Basedon Deep Learning

孔缝的识别和定量表征是页岩储层精细评价的核心内容。龙马溪组页岩非均质性强,发育各类无机孔、有机孔和微裂缝。孔缝的尺度变化范围大,纳米、微米、毫米级均有发育。由于页岩中广泛存在的黄铁矿、石墨化和沥青质等多种成岩矿物和成岩后生作用的影响,难以使用传统的成像测井资料识别孔缝。采用数字岩心的方法,主要针对龙马溪组各小层的电镜薄片和微电阻率扫描成像测井图像进行灰度化、二值化,并采用阈值分割的方法,对原始图片进行采样,将图片转化为RGB像素点。将二值化后的RGB像素点作为输入参数,基于残差卷积神经网络对龙马溪组页岩的大量电镜薄片进行学习,再对新处理的电镜薄片进行判别,识别出其中的粒间孔、粒内孔、溶蚀孔、有机孔和微裂缝,并对孔缝的主要参数进行计算和统计分析。这一方法为快速研究龙马溪组页岩孔缝结构与定量表征孔缝结构参数提供了有益的参考。

The identification and quantitative characterization of pores and fractures in shale reservoirs is the core content of fine evaluation of shale reservoirs.The shale in Longmaxi formation has strong heterogeneity,and various inorganic pores,organic pores and micro-fractures are developed.The size of pores and slits varies widely,ranging from nm,μm,and mm.Due to the influence of various diagenetic minerals such as pyrite,graphitization and asphaltene widely existing in shale and sedimentary epigenesis,it is very difficult to use traditional imaging logging data to identify.In this paper,the method of digital core is used to grayscale and binarization the image of the electron microscope thin section and formation microscanner image logging data of each sublayer of the Longmaxi formation.The threshold segmentation method is used to sample the original image and convert the image are RGB pixels.Then,the binarized RGB pixels are used as input parameters,and a large number of scanning electron microscope photos of Longmaxi formation are learned by using Residual Convolutional Neural Network(ResNet-CNN),and the newly processed electron microscope slices are discriminated and identified.The intergranular pores,intragranular pores,dissolution pores,organic pores and micro-fractures are obtained,and the main parameters of pores and fractures are calculated and statistically analyzed.This method provides a useful reference for the rapid study of the pore-fracture structure of shale in Longmaxi formation and the quantitative characterization of pore-fracture structure parameters.