利用泥浆侵入效应评估储层渗透率的模型和方法（英文）

渗透率评价是储层评价中的一个复杂问题,传统测井方法难以对储层渗透率参数进行直接和有效的评估。储层渗透率对钻井过程中的泥浆滤液侵入有着较大的影响,因此本文提出一种利用泥浆侵入效应来评价储层渗透率的数学模型和方法。首先构造含泥饼的泥浆侵入数值模型,然后以达西渗流理论为基础导出泥浆侵入深度和储层渗透率的近似数学关系,以此可以利用侵入深度测量值来估算储层渗透率。对孔隙度、渗透率和含水饱和度三个主要储层参数的敏感性分析,发现该方法适用于受到泥浆侵入影响的低孔隙度、低渗透率的油层或油水同层。采用现场测井和取芯数据进行数值模拟计算,结果表明估算出来的渗透率曲线和预设的渗透率曲线吻合较好,证明该方法具有一定的可行性。     

基于LSTM循环神经网络的储层物性参数预测方法研究

储层物性参数是反映储层油气储集能力的重要参数,表征了不同地质时期的沉积特征.地球物理测井参数由深及浅反映了不同地质时期的声、放、电等沉积特征,因而测井参数和泥质含量(孔隙度)之间有很强非线性映射关系,并具有时间序列特征.充分利用多种测井参数预测储层泥质含量和孔隙度对于储层精细描述具有十分重要的意义.深度学习技术具有极强的数据结构挖掘能力,目前,全连接的深度神经网络已经在泥质含量预测进行了初步尝试并取得了较好的效果.而长短时记忆(LSTM)循环神经网络更适合解决序列化的数据问题,因此本文提出基于LSTM循环神经网络利用多种测井参数进行泥质含量和孔隙度预测的方法,预测结果的均方根误差比常规全连接深度神经网络分别下降了42.2%和48.6%,实际应用表明,对于具有序列化特性的泥质含量和孔隙度,LSTM循环神经网络预测的准确性和稳定性要明显优于常规全连接深度神经网络.     

碎屑岩泥质分布形式的测井评价研究进展

泥质含量是分析碎屑岩地层沉积环境的重要指标,也是储层测井评价的关键参数,更是岩性和物性精细解释的基础.目前在泥质的测井响应特征认识和泥质含量参数精准计算方法上取得了大量的研究成果,但对于泥质分布形式的研究还不够完善.为了探寻计算泥质分布形式的有效方法,本文通过广泛调研,系统讨论了泥质分布形式与储层参数间的关系.其中不同分布形式泥质对包括孔隙度、渗透率、含水饱和度及束缚水饱和度在内的储层参数造成的影响主要表现在以下几个方面:层状泥质使有效孔隙度降低;结构泥质对有效孔隙度、渗透率影响不大;分散泥质使效孔隙度减小,渗透率变差,束缚水饱和度增加.随后深入分析了不同泥质分布形式对常规测井曲线的影响,其中分散泥质与结构泥质在伽马测井及自然电位测井曲线上表现出明显的特征.并初步探索了在包括电成像测井及核磁共振测井的特殊测井上响应特征;归纳总结了现有的泥质分布形式的计算方法,并对现有计算方法的适用性及优缺点进行了对比分析,最后利用Thomas Stieber交会图优化分析法在南海某油田低阻储层进行了应用并得到了良好的效果.     

砂泥岩储层孔隙度和含水饱和度同步反演

基于岩石物理和地震反演理论，提出了一种同步反演储层孔隙度和含水饱和度的方法.以岩石物理为基础，建立了砂泥岩储层物性和弹性参数之间定量的关系-Simon模型，以贝叶斯理论为手段，结合不同类型的砂泥岩储层，建立了多信息联合约束的物性参数反演目标函数，并通过蒙特卡罗和遗传算法相结合的思路求解该目标函数，最终得到孔隙度和含水饱和度的同步反演结果.将该方法应用于河道砂和砂砾岩两种不同的砂泥岩储层中，孔隙度和含水饱和度数据的联合应用，进一步减少了储层预测的多解性，为石油地质综合研究提供了更加丰富准确的基础数据.     

基于数据分布域变换与贝叶斯神经网络的渗透率预测及不确定性估计

渗透率是储层评价和油气藏开发的关键参数.传统测井方法与常规机器学习方法估算的渗透率都是固定值.但由于测井数据本身存在噪声，渗透率的预测结果可能受到噪声的影响出现测量性的随机误差(即任意不确定性);同时，当测试数据与训练数据存在差异时，机器学习模型在预测渗透率时可能出现模型参数的不确定性(即认知不确定性).为实现渗透率的准确预测并量化两种不确定性对结果的影响，本文提出基于数据分布域变换和贝叶斯神经网络同时实现渗透率预测及其不确定性的估计.提出方法主要包括两个部分：一部分是不同域数据分布的相互转换，另一部分是基于贝叶斯理论的神经网络渗透率建模预测和不确定性估计.由于贝叶斯神经网络存在数据分布的假设，当标签的概率分布与网络的分布保持一致时，贝叶斯神经网络可以更好的学习到数据之间的关系.因此通过寻找一个函数将一个原始域的渗透率标签转换为目标域的与渗透率有关的变量(我们称为目标域渗透率),使得该变量符合贝叶斯神经网络的分布假设.我们使用贝叶斯神经网络预测目标域渗透率以及任意不确定性和认知不确定性.随后，通过分布域的逆变换，我们将目标域渗透率还原回原始域渗透率.应用本文方法到某油田的18口井的测井...     

深度Transformer迁移学习的页岩气储层核心参数预测案例

地层纵横向非均质性强，工区间数据分布存在差异.这导致基于已有工区数据构建的机器学习储层参数预测模型，推广到新工区会存在较大预测误差.常规地质方法是在岩心与测井响应特征分析基础上建模，利用测井资料计算储层参数，流程复杂.该方法需要岩心校准模型，同样难以快速推广到新的工区.考虑地层纵横向非均质性，本文设计了一种深度Transformer迁移学习网络，通过已有工区的测井与岩心资料构建预测模型，实现未取心新工区储层参数快速准确预测.首先利用无监督学习算法-孤立森林剔除测井数据中存在的异常噪声数据.然后设计Transformer特征提取网络，提高网络特征提取能力，以此深入挖掘测井数据与储层参数的内在联系.最后设计深度迁移学习网络，构建网络损失函数，利用随机梯度下降算法优化网络参数，实现储层参数准确预测.本方案应用于四川南部地区五峰组—龙马溪组页岩储层参数孔隙度、总有机碳含量和总含气量预测.实验结果与工区校正后计算结果、主流机器学习模型预测结果对比，本方案结果与岩心数据具有更高的一致性.应用结果表明：本文方案具有实用性、有效性和可推广性.     

基于孔隙结构储层分类的中低孔特低渗储层渗透率确定——以B区块S油层为例

针对B区块S油层含泥含钙中低孔特低渗储层渗透率计算精度低的难题，分析岩性、物性、孔隙结构对储层渗透率的影响，明确了孔隙度、泥质含量、钙质含量、孔隙结构是影响B区块S油层特低渗储层渗透率的主要因素，其中，孔隙结构是影响特低渗储层渗透率的关键因素.综合运用压汞曲线、孔喉半径分布特征以及流动单元指数反映特低渗储层孔隙结构变化，将特低渗储层按不同孔隙结构划分成3种类型，建立了特低渗储层类型的判别标准.利用中子测井、密度测井、声波测井、微球形聚焦测井、深浅侧向电阻率测井差值的绝对值等5个储层类型识别的敏感测井响应及参数，使用决策树法、最邻近结点法、BP神经网络法和支持向量机法建立了4种基于机器学习的储层判别方法，储层类型判别准确率依次提高，其中，基于支持向量机的储层类型判别方法判别准确率最高92.2%,且对3种类储层判别效果均很好.针对3类储层分别建立了渗透率计算公式．实际井解释结果表明，基于机器学习储层分类的渗透率模型计算B区块S油层特低渗储层渗透率精度明显高于储层分类前渗透率计算精度，其中，基于支持向量机储层分类计算的渗透率精度最高．     

油水饱和泥质砂岩流动电位的频散特性

油水饱和泥质砂岩中流动电位的研究对于揭示含油储层震电勘探和动电测井的机理有着重要的意义.本文首先从岩石孔隙的微观结构出发,构造了描述水润湿条件下油水饱和泥质砂岩储层的毛管模型.在模型中依据油水流动遵守的Navier-Stokes方程和电化学传质动力学理论,建立了描述油水饱和泥质砂岩流动电位的数学方程,并数学模拟了岩石储渗参数对流动电位频散特性的影响规律.研究结果表明:储层孔隙内流体受到的粘滞力与惯性力控制着水相和油相的流动,从而决定了流动电位的频散特性.随着孔隙度的增大,油水两相各自的有效渗透率均增大;而含水饱和度的升高使得水相有效渗透率增大,油相有效渗透率减小.在水润湿条件下,流动电位耦合系数随含水饱和度升高而增大,随束缚水饱和度的升高而减小.另外,流动电位相对耦合系数也随含水饱和度的升高而增大,但无频散现象.     

致密砂岩气储层的岩石物理模型研究

根据鄂尔多斯盆地苏里格气田以往实测和新测的共17口井51块岩样超声波实验数据,得到304组不同孔隙度和不同含水饱和度下对应的纵横波速度、泊松比等弹性参数.重新优选计算体积模量和泊松比与含气饱和度的关系,表明苏里格气田上古生界二叠系石盒子组盒8致密砂岩储层的模型与Brie模型(e=2)相似度最高.由此建立的苏里格气田储层岩石物理模型,更好的表征了致密岩石储层物理参数随含气饱和度变化规律,为该区储层预测提供了理论依据.致密储层岩石物理模型研究成果应用于苏里格气田多波地震资料气水预测中,实际例子表明该模型适用于该区的储层和含气性预测,并取得了较好的效果.     

基于BP神经网络的储层渗透率预测及质量评价方法

储层渗透率预测和评价是油气藏勘探与开发急需突破的瓶颈技术之一，BP神经网络预测储层渗透率的研究在行业中已有一定的应用，但受限于数据规模、参数调整及模型评价方法，该方法预测结果不稳定，且不能准确给出全井段储层的连续渗透率的预测质量，在油田现场并未大规模推广应用.本文针对传统BP神经网络预测储层渗透率方法中存在的问题，在对机器学习的数据处理、参数选择系统考察的基础上，定量分析了不同输入曲线、网络结构、样本大小对渗透率预测模型精度的影响，总结了BP神经网络预测渗透率模型的参数优选方案；并提出了一种基于模型森林的预测曲线质量逐点评价方法，实现了对全井段渗透率预测的质量评价.实际应用表明，本研究提出的储层渗透率预测及质量评价方法与实际岩心渗透率吻合度高，推广应用前景良好.     

机器学习在地震预测中的应用进展

机器学习(Machine Learning, ML), 特别是深度学习(Deep Learning, DL), 在最近几年发展迅速, 在数据挖掘、 计算机视觉、 自然语言处理、 数据特征提取和预测等方面的应用中取得了令人振奋的进展。 地震预测是复杂、 涉及面广、 不成熟而且充满争议的科学问题; 其发展受到尚不清楚的地震机理和孕震结构、 不完备的观测数据与真伪不清的地震现象等方面的限制。 但是, 机器学习有可能改善复杂地震数据的挖掘和发现, 推动地震预测科学的发展。 本文回顾了机器学习在地震预测的应用, 包括强震、 强余震和岩石破裂失稳等方面的预测, 并展望了机器学习在地震预测方面的研究趋势。     

SediNet: a configurable deep learning model for mixed qualitative and quantitative optical granulometry

I describe a configurable machine-learning framework to estimate a suite of continuous and categorical sedimentological properties from photographic imagery of sediment, and to exemplify how machine learning can be a powerful and flexible tool for automated quantitative and qualitative measurements from remotely sensed imagery. The model is tested on a dataset consisting of 409 images and associated detailed label data. The data are from a much wider sedimentological spectrum than previous optical granulometry studies, consisting of both well- and poorly sorted sediment, terrigenous, carbonate, and volcaniclastic sands and gravels and their mixtures, and grain sizes spanning over two orders of magnitude. I demonstrate the model framework by configuring it in several ways, to estimate two categories (describing grain shape and population, respectively) and nine numeric grain size percentiles in pixels from a single input image. Grain size is then recovered using the physical size of a pixel. Finally, I demonstrate that the model can be configured and trained to estimate equivalent sieve diameters directly from image features, without the need for area-to-mass conversion formulas and without even knowing the scale of one pixel. Thus it is the only optical granulometry method proposed to date that does not necessarily require image scaling. The flexibility of the model framework should facilitate numerous application in the spatiotemporal monitoring of the grain size distribution, shape, mineralogy and other quantities of interest of sedimentary deposits as they evolve, as well as other texture-based proxies extracted from remotely sensed imagery. © 2019 John Wiley & Sons, Ltd.     

机器学习在地震学中的应用进展

理解并预测多尺度、高维度和非线性的地震学现象是一个极具挑战性的科学任务.与日俱增的海量观测数据降低了信息收集和信息解读之间的耦合程度,增加了信息解读的抽象性和不确定性.然而,伴随大数据一同来临的还有人工智能计算机技术——机器学习.机器学习突出的隐式关系提取和复杂任务处理能力推动着研究学者们不断将机器学习的应用推向更广阔的领域.本文介绍了地震学中常用的机器学习算法及其应用范围,讨论了人工智能与地震数据相结合的发展方向.     

机器学习在地震紧急预警系统震级预估中的应用

地震预警是地震减灾工作的重要途径,而震级预估是整个地震紧急预警系统中重要且较为困难的一个环节.目前,世界上多个国家和地区都已建立了各自的地震预警系统,并且形成了特征频率(τ_p和τ_c等)相关和特征振幅(Pd等)相关的两类震级紧急预警的方法,但各有局限性.本文在已有的方法和理论基础上,运用机器学习算法,将日本KIK和KNET台网从2015年至2017年所记录到的843条地震目录,55426条记录作为全数据集,设计、训练出一套用于常见震级范围的机器学习震级预估模型.与已有方法的预估结果相比,机器学习方法不仅使预估的整体误差和方差下降,同时多台联合评估单一地震事件的截面方差也更低.本研究的结果显示了机器学习算法在震级紧急预估问题上具有较广阔的应用前景,同时也为较为复杂的深度学习类算法框架下端到端模型提供了实践基础和研究可能.     

基于机器学习的水—气界面CO2、CH4扩散通量预测及影响因素分析——以三峡水库为例

传统的水-气界面温室气体通量的监测方法具有诸多局限,对其影响因素的分析也大多基于数学统计层面。对此,本研究提供了一种较为新颖的研究和分析方法——基于机器学习的数据预测和分析。本研究采用2种经典机器学习算法——随机森林(RF)和支持向量机(SVM)和2种深度学习算法——卷积神经网络(CNN)和长短时记忆神经网络(LSTM),通过环境因素预测水库水-气界面CO₂和CH₄扩散通量。此外,采用RF中的特征重要性评估和经典算法决策树(DT),对环境因素和水库温室气体扩散通量的关系进行了全新角度的数据挖掘和分析。结果表明:深度学习算法的预测效果均较好,经典机器学习算法中RF预测效果显著优于SVM。LSTM和RF分别产生了最优的CO₂扩散通量和CH₄扩散通量的预测精度,均方根误差(RMSE)分别为0.424 mmol/(m²·h)和0.140μmol/(m²·h),预测值与实测值的R²分别为0.960和0.758。RF的特征重要性评估表明沉积物因子...     

Comparative study of multilayer perceptron-stochastic gradient descent and gradient boosted trees for predicting daily suspended sediment load: The case study of the Mississippi River,U.S.

Monitoring sediment transport is essential for managing and maintaining rivers.Estimation of the sediment load in rivers is fundamental for the study of sediment movement,erosion,and flood control.In the current study,three machine learning models-multi-layer perceptron(MLP),multi-layer perceptron-stochastic gradient descent(MLP-SGD),and gradient boosted tree(GBT)-were utilized to estimate the suspended sediment load(SSL)at the St.Louis(SL)and Chester(CH)stations on the Mississippi River,U.S.Four evaluation criteria including the Correlation Coefficient(CC),Nash Sutcliffe Efficiency(NSE),Scatter Index(SI),and Willmott’s Index(WI)were utilized to evaluate the performance of the used models.A sensitivity analysis of the models to the input variables revealed that the current day discharge variable had the most effect on the SSL at both stations,but in the absence of current-day discharge data(Qt),a combination of input parameters including SSL_t-3,SSL_t-2,SSL_t-1,Q_t-3,Q_t-2,Q_t-1 can be used to estimate the SSL.The comparative outcomes indicated the high accuracy of MLP-SGD-5 model with a CC of 0.983,SI of 0.254,WI of 0.991,and NSE of 0.967 at station CH and the MLP-SGD-6 model with a CC of 0.933,SI of 0.576,WI of 0.961,and NSE of 0.867,respectively,at station SL.The results of MLP models were improved by SGD optimization.Therefore,the MLP-SGD method is recommended as the most accurate model for SSL estimation.     

Deep learning approaches for improving prediction of daily stream temperature in data-scarce,unmonitored, and dammed basins

Basin-centric long short-term memory (LSTM) network models have recently been shown to be an exceptionally powerful tool for stream temperature (T_s) temporal prediction (training in one period and predicting in another period at the same sites). However, spatial extrapolation is a well-known challenge to modelling T_s and it is uncertain how an LSTM-based daily T_s model will perform in unmonitored or dammed basins. Here we compiled a new benchmark dataset consisting of >400 basins across the contiguous United States in different data availability groups (DAG, meaning the daily sampling frequency) with and without major dams, and studied how to assemble suitable training datasets for predictions in basins with or without temperature monitoring. For prediction in unmonitored basins (PUB), LSTM produced a root-mean-square error (RMSE) of 1.129°C and an R² of 0.983. While these metrics declined from LSTM's temporal prediction performance, they far surpassed traditional models' PUB values, and were competitive with traditional models' temporal prediction on calibrated sites. Even for unmonitored basins with major reservoirs, we obtained a median RMSE of 1.202°C and an R² of 0.984. For temporal prediction, the most suitable training set was the matching DAG that the basin could be grouped into (for example, the 60% DAG was most suitable for a basin with 61% data availability). However, for PUB, a training dataset including all basins with data was consistently preferred. An input-selection ensemble moderately mitigated attribute overfitting. Our results indicate there are influential latent processes not sufficiently described by the inputs (e.g., geology, wetland covers), but temporal fluctuations can still be predicted well, and LSTM appears to be a highly accurate T_s modelling tool even for spatial extrapolation.     

Adjusting wavelet‐based multiresolution analysis boundary conditions for long‐term streamflow forecasting

We propose a novel technique for improving a long‐term multi‐step‐ahead streamflow forecast. A model based on wavelet decomposition and a multivariate Bayesian machine learning approach is developed for forecasting the streamflow 3, 6, 9, and 12 months ahead simultaneously. The inputs of the model utilize only the past monthly streamflow records. They are decomposed into components formulated in terms of wavelet multiresolution analysis. It is shown that the model accuracy can be increased by using the wavelet boundary rule introduced in this study. A simulation study is performed to evaluate the effects of different wavelet boundary rules using synthetic and real streamflow data from the Yellowstone River in the Uinta Basin in Utah. The model based on the combination of wavelet and Bayesian machine learning regression techniques is compared with that of the wavelet and artificial neural networks‐based model. The robustness of the models is evaluated. Copyright © 2015 John Wiley & Sons, Ltd.     

Common-offset domain reflection–diffraction separation with deep learning

Separation of diffracted from reflected events in seismic data is still challenging due to the relatively low amplitude of the diffracted wavefield compared to the reflected wavefield as well as the overlap in the kinematics of reflection and diffraction events. A workflow based on deep learning can be a simple and fast alternative, but using training data made by physics-based modelling is expensive and lacks diversity in terms of noise, amplitude, frequency content and wavelet. This results in poor generalization beyond the training data without retraining and transfer learning. In this paper, we demonstrate successful applications of reflection–diffraction separation using a conventional U-net architecture. The novelty of our approach is that we do not use synthetic data created from physics-based modelling, but instead use only synthetic data built from basic geometric shapes. Our domain of application is the pre-migration common-offset domain where reflected events resemble local geology and the diffracted wavefield consists of downward convex hyperbolic diffraction patterns. Both patterns were randomly perturbed in many ways while maintaining their intrinsic features. This approach is inspired by the common practice of data augmentation in deep learning for machine vision applications. Since many of the standard data augmentation techniques lack a geophysical motivation, we have instead perturbed our synthetic training data in ways to make more sense from a signal processing perspective or given our ‘domain knowledge’ of the problem at hand. We did not have to retrain the network to show good results on the field data set. The large variety and diversity in examples enabled to trained neural networks to show encouraging results on synthetic and field data sets that were not used in training.     

USTC-Pickers: a Unified Set of seismic phase pickers Transfer learned for China

Current popular deep learning seismic phase pickers like PhaseNet and EQTransformer suffer from performance drop in China. To mitigate this problem, we build a unified set of customized seismic phase pickers for different levels of use in China. We first train a base picker with the recently released DiTing dataset using the same U-Net architecture as PhaseNet. This base picker significantly outperforms the original PhaseNet and is generally suitable for entire China. Then, using different subsets of the DiTing data, we fine-tune the base picker to better adapt to different regions. In total, we provide 5 pickers for major tectonic blocks in China, 33 pickers for provincial-level administrative regions, and 2 special pickers for the Capital area and the China Seismic Experimental Site. These pickers show improved performance in respective regions which they are customized for. They can be either directly integrated into national or regional seismic network operation or used as base models for further refinement for specific datasets. We anticipate that this picker set will facilitate earthquake monitoring in China.