WELL TEST MODEL IDENTIFICATION BY ARTIFICIAL NEURAL NETWORKS

The aim of this research is to investigate the performance of artificial neural networks computing technology, to identify preliminary well test interpretation model based on derivative plot. The approach is based on training the neural network simulator uses back-propagation as the learning algorithm for a predefined range of analytically generated well test response. The trained network is then requested to identify the well test identification model for test data, which is not used during training sessions. For creation of training examples, an analytical response generator is implemented which is capable of producing responses of various models. Both the neural network simulator and the analytical response generator is enfolded into a single package which is capable of producing diagnosis plots, transferring data and filtering the input pattern. Unlike the ones presented in literature the package utilises a distributed modular structure, by which saturation possibility of the neural network is reduced considerably. Moreover, the distributed structure allows the training sequence to be initiated on different computers, thus reducing the training time up to sixteen folds. The package is verified with several examples either analytically generated or taken from literature.     

A New Approach to Sand Production Onset Prediction Using Artificial Neural Networks

Abstract

Sand production prediction has always been an important issue when dealing with production phenomena. Knowing all significant consequences of precise sand production prediction, different methods were developed using a variety of criteria and material models were implemented to obtain more accurate results. Although sand rate prediction has become a prevalent challenge nowadays, it does not reduce sanding onset prediction. Dealing with different methods and knowing the disadvantages of each one will clarify the necessity of developing a technique having the exactness and accuracy of numerical and experimental methods and simplicity of analytical ones. There was an endeavor in this article to apply powerful tools of an artificial neural network to predict critical bottomhole flowing pressure inhibiting sand production. Comprehensive well data gathered from 38 wells distributed in three oilfields producing from the same source rock were investigated to find the main parameters causing sand production. After verifying the proposed model with test wells, it was evaluated against well-accepted analytical models. The final results illustrate a reliable and more exact method that can predict sand initiation with a high degree of accuracy.     

NEURAL NETWORK PREDICTION OF PERMEABILITY IN THE EL GARIA FORMATION, ASHTART OILFIELD, OFFSHORE TUNISIA

The Lower Eocene El Garia Formation forms the reservoir rock at the Ashtart oilfield, offshore Tunisia. It comprises a thick package of mainly nummulitic packstones and grainstones with variable reservoir quality. Although porosity is moderate to high, permeability is often poor to fair with some high permeability streaks. The aim of this study was to establish relationships between log‐derived data and core data, and to apply these relationships in a predictive sense to uncored intervals. An initial objective was to predict from measured logs and core data the limestone depositional texture (as indicated by the Dunham classification), as well as porosity and permeability. A total of nine wells with complete logging suites, multiple cored intervals with core plug measurements together with detailed core interpretations were available. We used a fully‐connected Multi‐Layer‐Perceptron network (a type of neural network) to establish possible non‐linear relationships. Detailed analyses revealed that no relationship exists between log response and limestone texture (Dunham class). The initial idea to predict Dunham class, and subsequently to use the classification results to predict permeability, could not therefore be pursued. However, further analyses revealed that it was feasible to predict permeability without using the depositional fabric, but using a combination of wireline logs and measured core porosity. Careful preparation of the training set for the neural network proved to be very important. Early experiments showed that low to fair permeability (1–35 mD) could be predicted with confidence, but that the network failed to predict the high permeability streaks. “Balancing ” the data set solved this problem. Balancing is a technique in which the training set is increased by adding more examples to the under‐sampled part of the data space. Examples are created by random selection from the training set and white noise is added. After balancing, the neural network's performance improved significantly. Testing the neural network on two wells indicated that this method is capable of predicting the entire range of permeability with confidence.     

连续重整—加氢裂化联合装置集散仿真培训器的研制

开发了ＰＣＷｉｎｄｏｗｓ任务和Ａｌｐｈａ服务器任务间的内存对内存直接数据通讯的技术，应用于石油化工过程ＤＣＳ仿真培训器，通讯速度大幅度提高，使一套仿真培训器能同时运行多套生产装置的动态模型；采用面向对象思想和开发手段，研制出商品化程度较高的通用Ｉ／Ａ′Ｓ集散系统仿真软件平台。编制了比真正的ＤＣＳ系统更开放、实用的组态工具软件和操作员界面软件；应用新机理分析方法建立生产装置的动态数学模型；实现了６００ｋｔ／ａ连续重整、１Ｍｔ／ａ加氢裂化联合装置压缩机组等多套炼油装置的同时培训要求。     

PREDICTING THE EFFECT OF FEEDSTOCK ON PRODUCT YIELDS AND PROPERTIES OF THE FCC PROCESS

ABSTRACT

The mechanism of petroleum refining processes are too complex, and no thorough model has yet been developed. Neural networks represent an effective alternative to mathematical modeling of refinery operations if a sufficient amount of input-output data is available. In this paper, a feed forward neural network that models the Fluid Catalytic Cracking (FCC) process will be presented. The FCC process is the workhorse of the petroleum refining industry, making small and medium sized molecules out of big ones (gasoline and distillate out of gas oils). The input-output data to the neural network was collected from the literature on pilot and commercial plant operations and were obtained from actual refineries. Several network architectures were tried and the network that best simulates the FCC process was retained. This network is able to predict yields of products of the FCC unit as well as their properties. The network consists of one hidden layer of twenty neurons, an input layer of four neurons, and an output layer of twelve neurons. The predictions of the neural network model were compared to those of a commercial simulator of the FCC process, to non-linear regression models, and to published charts. The results show that the neural network model consistently gives better predictions.     

The Challenge of Model Identification in Well Test Interpretation—A Unique Build Up Analysis Case Study

Reservoir engineers are frequently faced with a very difficult question, thought to be a simple one, when attempting to analyze pressure transient test data. Correct model identification that can be used to analyze test data is ought to be a very simple exercise due to numerous analytical models, which are published in the literature. However, dealing with almost a black box situation “Underground Reservoir,” it is still difficult today to identify a reservoir model that represents a particular test. In fact, the pressure derivative technique in identifying flow regimes that pressure behavior undergoes during the well testing has introduced another dimension and is considered to be an important step in analyzing pressure transient tests. However, the non-uniqueness in model responses, which produces similar pressure derivatives, hinders the engineers, and most often causes in taking a wrong decision. This study looks into non-unique model response, introduces a simplistic method using numerical simulator and any reservoir information available in eliminating non-realistic solutions. Analysis of actual well test data, to obtain the very basic properties, and integrating any other available data, one can build different model realizations. An actual model was build by history matching the pressure derivative that helped reducing the problem of non-uniqueness response.     

应用卷积神经网络识别测井相

选用东海F气田的砂质辫状河三角洲的自然伽马数据作为训练数据构建深度卷积神经网络,并首次用于测井相识别。选用四种自然伽马曲线形态作为特征,将数值转变为图像形式,首先对图像做标准化、添加噪声、旋转和转灰度等处理,再对数据增强与扩充,建立训练和测试数据集;然后,训练卷积神经网络建立测井相识别模型,并在训练过程中加入了Dropout、局部响应归一化和L₂正则化等策略限制了模型的复杂程度,提高了模型泛化能力;针对测井信息中不同级次沉积单元响应叠加带来的自动识别难题,使用不同尺度的小波基函数及极值分割处理和切分测井数据,最终有效划分了不同尺度沉积单元。通过与其他分类算法对比,验证了所提方法具有较好的测井相识别效果。     

清洗技术在输油站设备运行中的应用

石油石化设备在使用过程中结垢严重会引起效率低下，腐蚀穿孔，不能正常使用，甚至引起安全事故，清洗技术能较好地解决这些问题。以热煤炉、原油换热器为例，介绍了装置的结垢状况，分析了结垢原因以及产生的影响；以东营输油站的热煤炉、原油换热器结垢清洗为例介绍清洗技术的工业应用；详细介绍了清洗的方法、步骤，化学清洗过程中用药量的控制，采取的必要措施，以及清洗后达到的效果；热煤炉通过采用化学清洗取得良好的效果，原油换热器通过高压水清洗取得良好的效果。清洗技术在降低设备腐蚀，提高设备使用效率，保障设备安全运行，延长设备使用寿命上起到了积极的作用。     

基于BP神经网络的腐蚀管道可靠性分析

利用BP人工神经网络算法建立基于BP神经网络腐蚀管道失效预测模型。通过BP神经网络拟合极限状态方程,借助神经网络的函数映射关系产生大量的极限状态函数值,作为下一步的分析数据。采用蒙特卡洛法随机抽样的思路,对大范围的数据进行概率分析,通过概率分析得到极限状态函数值的均值和标准差,求得腐蚀管道可靠性指标,解决了腐蚀管道的可靠性分析问题。     

DS-100钻井模拟器的研制与应用

DS-100钻井模拟器可在室内实现野外钻井作业过程及操作技能,达到逼真的声、像、作业过程效果,为培训技术工人和技术干部提供了极好的手段。它包括计算机控制系统、仪表检测显示系统、图形生成系统、数字音响系统、数据采集系统、通讯控制系统、电源系统等七大系统;正常钻进、起下钻、井控、教学演示等四部分共170个数学模型;还具有钻井设备基本操作训练、正常钻进训练、起下钻训练、井控训练、钻井工程故障判断、仪表系统故障自检、教学演示等功能。     

连续油管直缝高频焊热影响区最薄弱区硬度的神经网络预测

通过试验得出了连续油管HFW焊接接头最薄弱区域的力学性能,采用BP神经网络对该区域工艺性能进行仿真预测,研究了不同训练函数对网络性能的影响。对比分析不同训练函数下的网络性能,得出连续油管HFW焊接接头最薄弱区线能量一硬度预测模型,最终选取LM算法、SCG算法和动量BP算法对网络进行训练,采用这3种算法建立起的线能量一硬度模型精度较高,测试数据预测值与实测值平均相对误差分别为0．12％,0．095％和O．11％,表明神经网络模型能够很好地对“未知”硬度进行预测。     

基于振动信号分析的电潜柱塞泵故障诊断方法研究

振动信号能够全面反应电潜柱塞泵的运行工况,在电潜柱塞泵的故障诊断中,针对不同工况下的振动信号进行分析尤为重要。提出了一种基于振动信号分析的电潜柱塞泵故障诊断方法,采用改进的隐层神经元数变动的神经网络系统,根据不同情况选择不同节点数,使其达到提高诊断精度及缩短诊断周期的双重要求。对电潜柱塞泵正常运行以及动子不平衡、动子机械磨损运行时的振动信号进行分析研究,利用小波包提取振动信号的能量特征,利用神经网络识别故障。现场试验结果表明,网络训练后实际输出达到允许误差范围,网络测试结果也能与实际状态相对应。该方法能够对电潜柱塞泵进行有效、准确的故障诊断。     

改进人工神经网络原理对储层渗透率的预测——以北部湾盆地涠西南凹陷为例

人工神经网络的计算方法是一种非线性处理系统,是根据测井数据进行储层物性参数预测的方法。以往在利用遗传算法预测渗透率的时候,因为只考虑了单一的数据点,没有把临近层位的数据加入学习过程中来,故影响了预测模型的精度和可信度。为弥补这一不足,利用相临多个层位的数据点进行学习,进而建立储层渗透率的预测模型,并在岩心分析化验数据和相关测井曲线数据归一化的基础上,利用改进的开窗技术,借助反馈的神经网络方法对地层的渗透率进行逐点计算。通过北部湾盆地涠西南凹陷的实例实践表明,用该方法预测的渗透率与实测的渗透率的值符合较好。     

构建三维深度监督网络的断层检测方法

地震资料人工解释断层往往具不确定性。随着计算机和人工智能的发展,深度学习技术越来越多地应用于地球物理领域,多种基于卷积神经网络的算法也广泛地应用于断层识别。为此,结合三维U-Net和深度残差网络,引入多层深度监督的机制,构建了一种基于三维深度监督网络的断层检测方法。残差模块的引入能够简化网络的学习目标,降低训练难度,而多层的深度监督能够为网络提供更多的反馈,减轻训练过程中潜在的梯度消失,使解码器子网络能够学习到不同尺度的断层语义信息,可进一步提高断层识别的准确性。理论模型测试和实际地震资料的应用表明,该方法可以有效识别断层位置;与常规U-Net网络相比,减少了小断层的漏识别和错误识别;识别的大断层连续性好,断层细节更丰富,明显提高了断层识别的准确性。     

斜井多井系统中一口井压力降落曲线试井分析

本文提出一套斜井多井系统中单井测试曲线的试井分析方法,其中包括任意倾斜角度、任意变化产量、任意排列形式的任意多口斜井的多井系统压力响应的算法和适合于斜井多井系统多解特征的求解试井参数的非线性规划算法。两种算法都通过算例得到了验证该计算方法简单明了,计算率和计算精度可靠。     

基于改进人工神经网络的气水层识别技术

在改进的神经网络训练算法的基础上．提出了利用神经网络快速识别气、水层的方法。为了迅速、准确地判断储层性质,选用了Kohonen自组织网络和BP神经网络,利用测井参数,建立了长庆气田气、水层识别模型。仿真计算与测井综合解释相对比,样本符合率高达81．3%。分析表明,该方法所需参数少、适用范围广,能定量识别出气水层,从而为制定有水气井改造措施提供较可靠的依据。     

基于人工神经网络的石油化工工程建设项目管理绩效评价

针对非线性多输入多输出的石油化工工程建设项目管理绩效评价问题,应用人工神经网络构建评价模型。使用50个项目的287个学习案例数据,以10个影响因素为输入,6个指标为输出,对BP神经网络、基于遗传算法的BP神经网络、径向基函数神经网络与广义回归神经网络4类网络模型进行训练和测试。通过均方误差的比较,发现基于遗传算法的BP神经网络优于一般的BP神经网络,广义回归神经网络的测试结果优于BP神经网络,径向基函数神经网络具有最好的误差精度。2个应用示例表明,人工神经网络应用于石油化工工程建设项目管理绩效的评价是可行和有效的。     

A New Approach for Computing Pressure Derivative in Well Test Analysis

Abstract

Analysts commonly use the pressure derivative to identify flow regimes and well test interpretation models in pressure-transient test analysis. This article developed a novel adaptive smoothing algorithm by recursive differentiation–integration to improve the pressure derivative calculation. Our method can efficiently suppress measurement errors and produce smooth pressure derivatives from well test data. Equally important, it can prevent oversmoothing of the data by avoiding inappropriate use of large window size, and it can preserve the characteristic behavior of the pressure derivative. We validate our approach with a synthetic case and demonstrate its applicability to actual field examples.     

A New Approach for the Development of Fast-analysis Proxies for Petroleum Reservoir Simulation

Abstract

The bottleneck of all processes that are using field-scale numerical simulators is the computationally expensive objective function evaluation. Hence, always a gap exists between simulation runs and real-time processing. In this study, a new approach is presented that uses online-adaptive artificial neural networks to develop proxies that mimic the behavior of the actual reservoir simulator. In this approach, initially Latin hypercube sampling is used and then an intelligent sample selection algorithm is developed to improve the online network prediction. The cited approach improves the surrogate model development in two directions. First, proxies can be used while they are developing and, second, samples are selected intelligently and this reduces computational cost.