Characterization of reservoir properties and pore structure based on micro-resistivity imaging logging: Porosity spectrum, permeability spectrum, and equivalent capillary pressure curve

According to the capillary theory,an equivalent capillary model of micro-resistivity imaging logging was built.On this basis,the theoretical models of porosity spectrum(Ф_(i)),permeability spectrum(K_(i))and equivalent capillary pressure curve(pe)were established to reflect the reservoir heterogeneity.To promote the application of the theoretical models,the Archie's equation was introduced to establish a general model for quantitatively characterizing bi,K,and pei.Compared with the existing models,it is shown that:(1)the existing porosity spectrum model is the same as the general equation of gi;(2)the Ki model can display the permeability spectrum as compared with Purcell's permeability model;(3)the per model is constructed on a theoretical basis and avoids the limitations of existing models that are built only based on the component of porosity spectrum,as compared with the empirical model of capillary pressure curve.The application in the Permian Maokou Formation of Well TsX in the Central Sichuan paleo-uplift shows that the Ф_(i),K_(i),and p_(ci) models can be effectively applied to the identification of reservoir types,calculation of reservoir properties and pore structure parameters,and evaluation of reservoir heterogeneity.

A new method to construct reservoir capillary pressure curves using NMR log data and its application

By analyzing hundreds of capillary pressure curves, the controlling factors of shape and type of capillary pressure curves are found and a novel method is presented to construct capillary pressure curves by using reservoir permeability and a synthesized index. The accuracy of this new method is verified by mercury-injection experiments. Considering the limited quantity of capillary pressure data, a new method is developed to extract the Swanson parameter from the NMR T2 distribution and estimate reservoir permeability. Integrating with NMR total porosity, reservoir capillary pressure curves can be constructed to evaluate reservoir pore structure in the intervals with NMR log data. An in-situ example of evaluating reservoir pore structure using the capillary pressure curves by this new method is presented. The result shows that it accurately detects the change in reservoir pore structure as a function of depth.

Discrimination of Natural Fractures Using Well Logging Curve Unit

It is very difficult to discriminate natural fractures using conventional well log data, especially for most of the matured oilfields in China, because the raw data were acquired with relatively obsolete tools. The raw data include only GR and SP curves, indicative of lithology, AC curves, used to calculate the porosity of the formation, and a set of logging curves from various electrode length resistivity by laterolog. On the other hand, these oilfields usually have a large amount of core data which directly display the characteristics of the formation, and enough information of injection and production. This paper describes an approach through which logging curves are calibrated in terms of the raw data, and then a prototype model of natural fractures is established based on the investigation of core data from 43 wells, totaling 4 000 m in length. A computer program has been developed according to this method. Through analysis and comparison of the features of logging curves, this paper proposes a new concept, the well logging curve unit. By strictly depicting its shape through mathematical methods, the natural facture can be discriminated. This work also suggests an equation to estimate the probability of fracture occurrence, and finally other fracture parameters are calculated using some experimental expressions. With this methodology, logging curves from 100 wells were interpreted, the results of which agree with core data and field information.

Studies on phase and group velocities from acoustic logging

It is still argued whether we measure phase or group velocities using acoustic logging tools. In this paper, three kinds of models are used to investigate this problem by theoretical analyses and numerical simulations. First, we use the plane-wave superposition model containing two plane waves with different velocities and able to change the values of phase velocity and group velocity. The numerical results show that whether phase velocity is higher or lower than group velocity, using the slowness-time coherence （STC） method we can only get phase velocities. Second, according to the results of the dispersion analysis and branch-cut integration, in a rigid boundary borehole model the results of dispersion curves and the waveforms of the first-order mode show that the velocities obtained by the STC method are phase velocities while group velocities obtained by arrival time picking. Finally, dipole logging in a slow formation model is investigated using dispersion analysis and real-axis integration. The results of dispersion curves and full wave trains show similar conclusions as the borehole model with rigid boundary conditions.

The quadratic B-spline curve fitting for the shape of log cross sections

This paper describes a new method for simulation of the cross section shape of log. The self-developed MQK3102 log shape recognizing machine was used to acquire the finite discrete sampling points on the cross section of log and those points were fitted with the quadratic B-spline parametric curve. This method can clearly stimulate the real shape of the log cross section and is characterized by limited sampling points and high speed computing. The computed result of the previous curve does not affect the next one, which may avoid the graphic distortion caused by the accumulative error. The method can be used to simulate the whole body shape of log approximately by sampling the cross sections along the length direction of log, thus providing a reference model for optimum saw cutting of log.

Fractal Reconstruction of Log Curve

Sedimentary cyclothems at different scales show formations’ fractal structure which can be reflected on logs. The slope of the power spectrum of log is related to the fractal dimension of formations. The fractal dimensions from two logs with similar vertical resolutions are the same. Using fractal interpolating algorithm density log can be reconstructed. The reconstructed log can be compared with core density in washout intervals.

Linearly optimum seeking K-value in the fitting of exponential and geometric curves

Based on logarithmic transformation of exponential and geometric curves, the paper proposes a method of choosing k-value to improve the goodness of fit. Functions between the sum of squares of regression(SSR) and k-value are built up; and the maximum of SSR value can be solved by derivation in which an iterative algorithm is used. A computer program has been developed for application. The result shows that the method is simple, and the goodness of fit is not inferior to other optimum methods of curve fitting.

基于小波变换和CNN-Transformer模型的测井储层流体识别

针对具有复杂储集空间和极强的非均质性的低孔低渗储层,常规测井响应特征不够明显,使用传统解释手段难以有效识别储层流体的问题,提出了一种基于小波变换和CNN-Transformer混合模型的储层流体识别方法。首先,使用小波变换将测井信号从时域扩展到时频域,并生成时频谱图以增强信号特征,然后使用滑动时窗沿着测井曲线深度方向滑动采样,获取代表解释深度处地层信息的频谱特征图,最后,通过训练CNN-transformer模型深度挖掘特征图信息,实现储层流体识别。混合模型在利用储层对应深度处测井数据的同时,又兼顾测井曲线随深度的变化趋势和地层前后信息的关联性,挖掘时频谱图的局部细节和全局特征表示,自动识别流体类型。将模型应用于大港油田22口实测测井资料中,并与CNN和BiLSTM等多个模型的流体识别效果进行对比分析,基于小波变换和CNN-Transformer模型识别效果明显优于其他方法,在测试集上识别准确率达到了92.7%。研究结果表明该方法可以作为低孔渗油藏常规测井资料识别储层流体的有效手段,为流体评价提供了新思路。

基于Stacking集成学习的声波时差测井曲线复原研究

声波时差测井曲线在石油勘探中发挥着不可或缺的作用,但是受地质或仪器的影响,经常会出现部分甚至完整的声波测井曲线缺失的情况。针对这一问题,提出了一种基于Stacking集成学习的声波时差测井曲线复原方法,该模型使用随机森林(RF)、梯度提升决策树(GBDT)、轻量梯度提升机(LightGBM)和极限梯度提升(XGBoost)作为基学习器,支持向量回归(SVR)作为元学习器,同时采用5折交叉验证的方法。实验选取了大庆油田某区块的实际测井数据,分别进行了同井和异井间的缺失声波时差测井曲线复原实验,结果表明,所提方法比单一模型预测更加准确,验证了此方法的可行性。

基于深度学习的栅格测井曲线数字化研究

栅格测井曲线是记录测井数据的模拟曲线,地质专家可根据所得曲线判断出具体岩性、层位等.目前无监督计算机视觉技术解释数字曲线算法中仍然存在对人工干预有依赖,处理时间较慢且准确率不高的问题,提出了一种基于深度学习的栅格测井曲线数字化研究方法.该方法采用改进的U⁃net启发架构作为基础模型,结合栅格测井图像分辨率高的特性改变上下采样的残差块数量,并利用注意力增强的处理策略平衡在处理不同的输入和输出大小时保留其关键信号,再对栅格图像中的曲线进行语义分割减少背景干扰,增加数字化准确率.实验结果表明,与多条曲线的二值分割的真实数据相比,所提出的架构有效地对曲线进行了分类和数字化,在栅格测井图像数字化方面表现出较高的准确度和性能.

基于主成分自组织神经网络法的测井曲线分层技术

在砂岩型铀矿找矿工作中,提高测井岩性分层效率和精度至关重要。为提高砂岩型铀矿岩性分层效果,本文采用主成分分析法对多个测井曲线进行降维处理,将主成分分析法的第一主成分、第二主成分、第三主成分作为自组织神经网络的样本数据,进行自组织神经网络训练,将训练好的网络模型用于砂岩型铀矿岩性的自动化分层。实验结果显示:主成分自组织神经网络法岩性分层精度可达到85%以上,高于传统自组织神经网络算法78%的分层精度,具有更好的测井岩性分层效果。因此,主成分自组织神经网算法的岩性分层方法有效减少了输入样本的种类,简化了自组织神经网络结构,其自动化分层效果要优于传统的自组织神经网络算法。本文的研究结果表明,主成分自组织神经网算法在砂岩型铀矿领域岩性识别工作中具有较好的应用效果。

牛东地区砂砾岩储层测井层内非均质性评价

牛东地区砂砾岩储层具有低孔、低渗等特征,储层结构复杂,非均质性较强,常规测井对储层类型划分困难,影响了储层评价精度。岩心孔渗、薄片分析、X衍射等测试资料表明,牛东地区砂砾岩储层具有较强的非均质性,且根据毛管压力形态可将储层分为3个类别。本次研究利用电成像测井资料对储层层内非均质性进行评价:首先对电成像测井资料进行处理得到储层孔隙度谱,再对随深度变化的孔隙度谱进行研究,分别计算不同深度上孔隙度谱的均值、方差、洛伦兹系数和集中程度函数;根据综合概率模型,利用层次分析法确定评价指标的权重,得到储层非均质性综合指数,据此对储层进行划分,并建立储层非均质性的评价标准。应用该方法得到的研究结果与压汞实验结果一致,说明利用该方法能够有效地评价储层非均质性,可丰富现有的储层非均质性评价方法,为储层的精细评价提供一定的理论支撑。

基于模态分解与LSTM注意表征的测井曲线重构研究

测井曲线记录着地层物理性质随深度变化的幅值范围,是测井与地震资料之间的纽带,对储层岩性分析与识别和后续的油气勘探工程十分重要.然而,在实际的测井过程中仪器故障等原因会造成测井曲线缺失的问题,重新测井不仅价格昂贵而且难以实现.针对地质勘探时测井数据时常缺失的问题,本文提出了一种LSTM(Long Short-Term Memory)注意力表征的测井曲线重构方法.同时,对原始测井信号进行两种模态分解,计算分解后得到模态分量与原始信号之间的相关性,去除冗余分量,实现对缺失的测井曲线高效、高精度的人工补全.将该方法用于声波(ACoustic,AC)与密度(DENsity,DEN)曲线重构实验,并将实验结果与LSTM网络和BP(Back Propagation)神经网络预测的结果进行对比分析.结果表明,LSTM-Attention模型有着更为优异的预测效果,重构后的AC和DEN与原始曲线之间的相关性分别达到了86.8%和74.8%,高于传统LSTM和BP神经网络预测方法.在去除冗余的信号分量后,相关系数分别提高了1.4%和4.0%.同时,本文所提方法预测出的测井曲线具有最低的预测误差.因此,基于LSTM注意表征的网络结构对测井曲线重构具有较好的预测精度.

基于多信息融合的层次聚类测井曲线自动分层方法

随着人工智能的快速发展,机器学习的应用范围越来越广泛,将机器学习的方法用于测井曲线分层可以提高分层效率和精度。在利用测井资料进行岩性识别、沉积相分析等研究时,先要对测井曲线进行分层。文中提出一种基于多信息融合的层次聚类分层方法,实现了对测井曲线的自动分层。首先,采用滤波的方式滤除曲线上的噪点,对数据进行归一化处理,消除量纲的影响;其次,通过特征优选,选择包含较多地层信息的特征曲线,构造一个滤波器,将其中相似性较高的曲线融合,曲线融合的权值通过遗传算法求得;最后,使用层次聚类方法对多信息融合后的测井数据进行划分,将分层结果与人工分层结果进行对比验证。该方法能够提高分层效率,为地质勘探工作提供可靠的分层依据。

轮廓曲线的LoG变换及图像共变区域的检测

本文将LoG(Laplacian of Gaussian)变换应用到轮廓曲线上,其范数的平方具有稳健的曲率特性,由此构造了轮廓角点的响应函数;在此基础上,结合轮廓LoG所检测的角点和角点邻域内轮廓方向的尺度共变性设计了一种新的关于旋转和尺度共变的特征区域检测算法.最后,利用相关度准则对多组受干扰的图像进行特征共变区域匹配,实验结果验证了所提出的共变特征区域检测算法具有计算简单、容易实现和较强的鲁棒性等特点.

基于常规测井曲线的页岩岩相识别与应用——以苏北盆地溱潼凹陷阜宁组二段为例

页岩岩相的识别、划分在页岩油气勘探开发工作中具有重要的理论及实际意义。以苏北盆地溱潼凹陷古近系阜宁组二段页岩为研究对象,通过对典型钻井溱页1井岩心样品进行全岩/黏土X射线衍射分析,应用前人的页岩矿物组分“三端元”图解,得出该地区的页岩岩相类型。同时,利用一种基于原子搜索优化算法优化的BP(反向传播算法)神经网络方法对测井信息进行数据挖掘,以此建立黏土矿物、硅质矿物、碳酸盐岩矿物相对含量的预测模型,实现了通过自然能谱到页岩矿物含量的定量表征。最后应用该模型对溱页1井和沙垛1井阜宁组二段进行岩性预测及岩相识别,其识别结果与样品实测数据的岩相划分结果高度一致。研究为实现页岩岩相的间接识别提供了一个经济、快速、高效的方法,可通过测井曲线有效地预测页岩岩相组合体的主要矿物成分,为缺乏取心、实测样品井段的岩相识别提供依据。

基于LoG算子改进的自适应阈值小波去噪算法

图像在传输过程中会受到各种噪声干扰,为了实现消除噪声的目的,提出一种基于LoG算子改进的自适应阈值去噪算法。首先,利用LoG算子提取图像的边缘特征信息。接着,根据图像的边缘特征和非边缘特征分别求取改进的阈值函数:对于图像非边缘部分的阈值函数,在软阈值函数的基础上添加一个阈值修正系数,构建新的阈值函数;对于图像边缘部分的阈值函数,将边缘部分小波系数附近的能量和阈值相结合,构建新的阈值函数。最后利用改进的阈值函数对图像R、G、B 3个通道分别处理,保留图像所有的细节信息。实验结果表明,消噪后图像与含噪图像的PSNR值高于传统自适应算法12.09%;MAE值低于传统自适应算法22%。该算法有效保存了图像的边缘信息,综合去噪效果明显提高。

基于Log算子边缘检测的车牌定位方法

车牌定位是车牌自动识别系统中的关键技术之一,提出一种基于Log算子边缘检测的车牌精定位方法。首先对彩色车辆图片在多颜色空间内进行色彩分割去除大量的背景干扰信息,然后利用分块的思想实现车牌的粗定位,大大缩小车牌的搜索区域,最后对粗定位图用Log算子检测边缘突出车牌的纹理特征再结合投影的方法准确定位出车牌。通过对静态车牌图像定位仿真实验和分析表明,该方法对于车牌定位准确率较高。

基于多关键井平均曲线标准化方案的致密砂岩渗透率精确计算

X地区致密砂岩储层地质条件复杂,纵横向非均质性强;测井资料采集时间跨度大、标准化程度低;储层参数影响因素复杂,导致孔隙度、渗透率的计算精度低。为此提出一种多关键井平均、双标准层控制、趋势面法与直方图法相互验证的测井曲线标准化方案,系统解决单一标准层存在的相变、缺失、非均质性、特征值异常等一系列问题,标准化后孔隙度计算精度显著提高。在此基础上,基于流体流动单元储层分类理论,确立储层分类标准,构建基于流体流动单元的分类渗透率精确计算模型。利用多元线性回归方法建立测井曲线表征的流体流动单元计算模型,实现储层分类和渗透率计算的连续、定量评价。提出的标准化方案可提高测井曲线标准化的稳定性、适应性和可靠性,为储层分类及参数计算提供坚实的数据基础。该方法为该地区致密储层精细评价提供有力的技术支持,具有重要的应用和推广价值。

基于MAE神经网络的测井曲线地层自动识别方法

针对目前计算机自动分层识别准确率低,人工分层效率不高的问题,提出了一种基于掩码自编码器(MAE)神经网络算法的地层分层识别新方法。通过测井曲线优选,以自然电位、自然伽马、声波时差、电阻率、井位坐标和补心海拔作为输入特征变量进行模型训练及预测,再利用损失函数、准确率、精确率、召回率、F1值进行模型性能评估。研究结果表明:(1)基于MAE神经网络训练得到地层分层识别模型,预测准确率能够达到95.54%;(2) MAE神经网络模型与卷积神经网络和关注分层边界的卷积神经网络模型进行地层分层实验对比,MAE神经网络模型的性能和预测精度均较高,准确率分别提高了8.30%和6.32%,且无地层紊乱情况出现,具有明显分层优势;(3) MAE神经网络模型应用于未知井的地层分层中,自动分层预测准确率为98.07%。表明该方法具有较高的地层识别效果,为油田地层识别提供了一种基于自监督神经网络算法的理论支持和有益探索。