模糊逻辑和神经网络及其在含油饱和度预测中的应用

对人工智能技术在油田开发中的应用做了简要回顾.介绍了模糊逻辑理论中的模糊排队算法和人工智能中的BP神经网络模型.将模糊排队算法和BP神经网络相结合,以新疆油田某井的实际测井曲线为例,用模糊曲线分析方法确定全局相关性强的输入变量(测井曲线),建立BP神经网络含油饱和度预测模型,并对含油饱和度做了预测,利用模型的预测结果和计算值相比较具有较高的吻合度,证明该方法在实际储层参数预测中具有良好的实用性.     

东海深部地层岩石可钻性预测方法研究北大核心CSCD

东海深部地层中部分井PDC钻头与地层性质匹配性不足,导致机械钻速低、钻头磨损严重等问题发生。以室内岩心微钻实验数据为依据,首先建立了测井参数预测岩石可钻性的非线性多元回归模型,同时利用多种人工神经网络方法对岩石可钻性进行了预测,结果表明非线性多元回归模型预测岩石可钻性与常规BP神经网络、级联BP神经网络、径向基RBF神经网络、BP-RBF双级联神经网络模型预测结果均具有较高可信度,但BP-RBF双级联神经网络模型预测效果最好,更适合于东海深部地层岩石可钻性预测。本文研究结果可为东海深部地层岩石可钻性预测及钻头选型提供借鉴。     

东海深部地层岩石可钻性预测方法研究

东海深部地层中部分井PDC钻头与地层性质匹配性不足,导致机械钻速低、钻头磨损严重等问题发生。以室内岩心微钻实验数据为依据,首先建立了测井参数预测岩石可钻性的非线性多元回归模型,同时利用多种人工神经网络方法对岩石可钻性进行了预测,结果表明非线性多元回归模型预测岩石可钻性与常规BP神经网络、级联BP神经网络、径向基RBF神经网络、BP-RBF双级联神经网络模型预测结果均具有较高可信度,但BP-RBF双级联神经网络模型预测效果最好,更适合于东海深部地层岩石可钻性预测。本文研究结果可为东海深部地层岩石可钻性预测及钻头选型提供借鉴。     

基于LSTM-BP神经网络的地层孔隙压力计算方法

地层孔隙压力是油气井从设计到完钻过程中重要的基础数据,准确计算地层孔隙压力是保障钻井安全、提高钻井效率的重要前提。为了克服传统地层孔隙压力计算方法精度不足、计算效率不高的问题,本文考虑到钻井与地层沉积均为序列性和非线性过程,提出了将长短期记忆神经网络(LSTM)和误差反向传播神经网络(BP)相结合计算地层孔隙压力的方法,利用LSTM层提取钻-测-录多源数据中的序列性特征信息,经过BP层构建特征信息与地层孔隙压力之间的非线性映射关系。通过对油田现场钻测录数据进行清洗并综合相关性分析和钻井经验知识优选了18种输入参数,对LSTM-BP地层孔隙压力计算模型进行训练和测试,并采用网格搜索法对LSTM-BP神经网络模型的5种模型超参数进行了优选,效果最优的单井计算模型和邻井计算模型的平均绝对误差分别为4.92 MPa和2.34 Mpa,均方根误差分别为6.65 Mpa和3.03 Mpa,平均相对误差分别为4.36%和8.31%。最后与传统BP模型、LSTM模型和支持向量机(SVM)模型的最优结果进行对比,结果显示,本文所建立的LSTM-BP神经网络模型精度均高于BP模型、LSTM模型和SVM模型...     

汽油调合优化神经网络模型的研究

采用BP型神经网络对某炼油企业汽油调合数学模型进行研究,依据汽油生产装置特点,确定了神经网络的拓扑结构,利用采集的汽油生产数据,确定了隐含层节点数和模型学习算法,并经过模型训练,得到了拟合能力和预测能力均较强的企业汽油调合神经网络。所建立的模型不需要调合机理的支持,因而具有较强的自适应能力。实际应用结果表明,该神经网络模型对调合过程中的非线性参数预测精度较高,可提供汽油调合的优化方案。     

基于BP和LSTM神经网络的顺北油田5号断裂带地层孔隙压力智能预测方法

顺北油田断裂发育,地质构造复杂,储集层埋深达8 000 m,具有高温高压、窄钻井液密度窗口等特征,地层孔隙压力的预测精度难以满足工程需求。为了提高地层孔隙压力的预测精度,利用人工智能方法在处理复杂非线性问题上的优势,采用反向传播神经网络BP和长短期记忆循环神经网络LSTM这2种人工智能算法,基于顺北油田5号断裂带上3口井的声波时差、自然电位和自然伽马等11种特征数据以及经实测校正的地层孔隙压力标签数据,建立了顺北油田5号断裂带地层孔隙压力智能预测模型,BP神经网络模型的预测误差为3.927%,LSTM神经网络模型预测误差为2.864%。测试结果表明,LSTM神经网络模型具有更好的预测效果,满足现场地层孔隙压力的预测精度,为保障顺北油田5号断裂带钻井安全提供数据参考。     

基于LM-BP神经网络的防气窜能力评价方法及应用

提出基于LM-BP神经网络进行防气窜能力评价方法,选取地层系数、静液压力系数、泥浆清除系数和水泥浆性能系数作为输入参量,将样本的防气窜能力作为输出量,在进行训练时采用LM算法对BP神经网络进行改进,提高了预测精度,得到基于LM的BP神经网络模型,利用该模型进行防气窜能力评价。研究结果表明:基于LM的BP神经网络模型的计算结果与测试样本拟合精度较高,具有广泛的应用前景。     

基于SAPSO-FNN的电动式加载系统控制研究

针对电动式加载系统中减速器带来的非线性等问题,提出了一种基于模拟退火的粒子群算法来优化模糊神经网络PID控制器初始参数的设计方法。该方法采用基于Mamdani模型的模糊神经网络,通过在线BP学习算法对PID控制器参数进行自整定,并结合基于模拟退火的粒子群算法离线调整模糊神经网络初始值。通过仿真与实验,将所设计算法与增量式PID控制器、粒子群-模糊神经网络PID控制器进行对比,证明了所设计的控制器具有良好的稳定性、有效性和快速性。     

地层可钻性级值预测新方法

对测井资料与地层可钻性级值的关系进行了分析,提出了一种基于粒子群优化支持向量机算法预测地层可钻性级值的新方法,利用测井声波时差、地层密度、泥质质量分数和地层深度进行学习训练支持向量机,并利用粒子群优化算法对支持向量机(PSO-SVM)参数进行优化,建立了预测地层可钻性级值的支持向量机模型。应用该方法对准噶尔盆地庄2井的地层可钻性级值进行了预测,并将该方法的预测结果与BP神经网络方法的预测结果进行了比较。结果表明,该方法优于BP神经网络方法,具有预测精度高、收敛速度快、推广能力强等优点。     

基于GRU神经网络的测井曲线重构

通过测井曲线解释可以获得地层岩性、电性以及孔渗饱等地层参数,然而,实际应用中时常出现部分测井数据失真或缺失的情况,而重新测井不仅价格昂贵且实现较困难。目前基于传统的线性假设和统计分析的测井曲线重构方法已不能满足储层特征的精细描述要求。门控循环单元（GRU）神经网络是一种适合于解决非线性和时序性问题的新型深度学习算法。基于深度学习的最新成果,提出使用GRU神经网络进行测井曲线重构。该方法兼顾了测井数据之间的非线性映射关系、数据随储层深度变化的趋势及历史数据之间的关联性。对实际资料进行试算,并与多元回归方法结果对比,表明GRU网络模型取得了良好的重构效果,为测井曲线重构提供了一种新的思路。     

基于神经网络的高压水射流冲蚀破碎预测模型的研究

在室内高压水射流冲蚀破岩试验结果的基础上，利用人工神经网络的基本特征．建立了高压水射流冲蚀破碎体积与射流压力、围压和喷距之间的数学模型，将其用人工神经网络的连接权值矩阵和节点阈值向量分布式表达出来，并应用人工神经网络进行了射流冲蚀破碎预测，预测结果与试验结果十分吻合，说明应用人工神经网络所建立的描述高压水射流冲蚀破碎体积与射流压力、围压、喷距之间关系的模型是可靠的。     

Hybrid Neural Network Performance Prediction Model for Gas Assisted Gravity Drainage Recovery Method Based on the Scaling Analysis

Hybrid system is a potential tool to deal with nonlinear regression problems. The authors present an efficient prediction model for gas assisted gravity drainage injection recovery process based on artificial neural network (ANN) and dimensionless groups. Ant colony optimization (ACO) is applied to determine the network parameters. Results show that ACO optimization algorithm can obtain the optimal parameters of the ANN model with very high predictive accuracy. The predicted recovery from the ACO-ANN model, in comparison with other proposed models in literature, were in good agreement with those measured from simulations, and were comparable to those estimated from the other proposed models.     

燃料油中有机硫化物在不同色谱柱上的定量结构色谱保留关系（QSRR）的研究

硫组分的含量是表征燃料油品质的重要指标。采用遗传算法-多元线性回归法(GA-MLR)、BP神经网络法、列文伯格-马夸尔特人工神经网络算法(L-M ANN)对52种有机硫化物在4种不同极性固定相上的气相色谱保留指数分别进行了定量结构-气相色谱保留关系研究。采用GA-MLR方法选取模型的输入参数,并将筛选得到的描述符：一阶分子连接性指数(1χ)、二阶分子连接性指数(2χ)、电子能(EE)、Y轴偶极(Dy)用于BP神经网络、L-M ANN人工神经网络定量结构保留（QSRR）模型的构建。结果表明：3种方法所建立的定量模型均具有较强的稳定性和良好的预测能力,其相关系数均在0.98以上,但L-M ANN模型的预测结果稍好于其它2种方法;L-M ANN算法首次被应用于燃料油中有机硫化物定量结构-气相色谱保留关系的研究中,效果十分理想,表明L-M ANN算法可以作为一种替代性的建模方法用于物质的定量结构保留关系的研究中。     

Development of artificial neural network models for predicting water saturation and fluid distribution

We have developed artificial neural network (ANN) models to predict water saturation from log data. Two Middle Eastern sandstone reservoirs were investigated. In the first case, an ANN model was tested on the Haradh formation in Oman using wireline logs and core Dean–Stark data. In the second case, the ANN was used to model the saturation–height function in a complex sandstone reservoir.In the first case study, the model is based on a three-layered neural network structure. The model was successfully tested yielding a prediction of water saturation with a root mean square error (RMSE) of around 0.025 (fraction of pore volume P.V.) and a correlation factor of 0.91 to the test data. Furthermore, the ANN model was shown to be superior to conventional statistical methods such as multiple linear regression, which gave a correlation factor of 0.41.In the second case, the model yielded a saturation–height function with an RMSE of 0.079 (fraction P.V.) in saturation when using core porosity and height above free water level. This is a considerable improvement over conventional methods. The error was also greatly reduced when permeability and a lithology indicator were introduced. A minimum error of 0.045 (fraction P.V.) was obtained when using core data such as height, porosity, permeability, lithology and a functional link. We then used gamma ray, neutron, density, resistivity wireline data and the cation exchange capacity as inputs. Our best case which gave an RMSE error of 0.046 (fraction P.V.) was obtained. The ANN was then used to predict the hydrocarbon saturation in the Gharif formation and good results were obtained. The neural network model proved the robustness of saturation prediction in another field for the same formation.     

Managed Pressure Drilling to Increase Rate of Penetration and Reduce Formation Damage

One of the advantages of managed pressure drilling (MPD) is an increase in drilling speed and a reduction of mud filtrate invasion as a result of decreasing pressure differential. Reduction of overbalanced pressure (OBP) leads to a decrease of confining pressure around the formation rock and consequently the rock is broken more easily under the bit action and therefore the rate of penetration (ROP) increases. It is also obvious that decreasing the overbalanced pressure results a reduction in mud filtrate invasion and formation damage. In the present article the effect of MPD on increasing rate of penetration and decreasing mud filtrate invasion is studied. Artificial neural networks (ANN) were implemented to develop a model for estimation of ROP by using operational inputs including overbalanced pressure. Using the ANN model, the effect of OBP was analyzed. The effect of OBP on mud filtrate invasion was studied by using developed models of the process and simple Darcy’s law. The results demonstrated that MPD leads to about 30% increase in rate of penetration and 50% decrease in mud filtrate invasion.     

垂直裂缝井开发低渗油藏非线性渗流压力分析

根据低渗透介质非线性渗流运动规律三参数连续函数模型，运用椭圆渗流的概念，建立了垂直裂缝井开发低渗油藏定常和非定常非线性渗流数学模型。运用平均质量守恒的方法，对垂直裂缝井开发低渗油藏非定常非线性渗流压力分布和压力扰动影响半径进行了数学推导。结果表明：垂直裂缝井开发低渗透油藏时，按非线性流规律与按线性渗流规律计算的结果差别明显，线性与非线性定常渗流压力分布在椭圆坐标系中分别为直线和曲线，线性定常渗流压力梯度不随椭圆坐标变化，非线性定常渗流压力梯度随椭圆坐标增大而减小；非线性渗流的影响半径发展速度比线性渗流慢；在同一无量纲时间下，非线性非定常渗流的井底压力比线性渗流的大；在对垂直裂缝井开发低渗透油气藏进行油气藏工程计算时，应考虑非线性渗流的影响。     

Development of an Artificial Neural Network Algorithm for the Prediction of Asphaltene Precipitation

Abstract

The precipitation and deposition of crude oil polar fractions such as asphaltenes in petroleum reservoirs considerably reduce rock permeability and oil recovery. Therefore, it is of great importance to determine how and how much the asphaltenes precipitate as a function of pressure, temperature, and liquid phase composition. The authors designed and applied an Artificial Neural Network (ANN) model to predict the amount of asphaltene precipitation at a given operating condition. Among this training, the back-propagation learning algorithm with different training methods was used. The most suitable algorithm with an appropriate number of neurons in the hidden layer, which provides the minimum error, was found to be the Levenberg-Marquardt (LM) algorithm. An extensive experimental data for the amount of asphaltene precipitation at various temperatures (293–343 K) was used to create the input and target data for generating the ANN model. The predicted results of asphaltene precipitation from the ANN model was also compared with the results of proposed scaling equations in the literature. The results revealed that scaling equations cannot predict the amount of asphaltene precipitation adequately. With an acceptable quantitative and qualitative agreement between experimental data and predicted amount of asphaltene precipitation for all ranges of dilution ratio, solvent molecular weight and temperature was obtained through using ANN model.     

Perforation friction modeling in limited entry fracturing using artificial neural network

The effect of perforation friction on hydraulic fracturing treating pressure is normally considered negligible. However; this is not the case in limited entry fracturing. In limited entry fracturing, perforation friction is utilized to attain large frictional pressure drop. During the treatment injection, the frictional pressure offsets the stress differences between the zones to ensure fluid injection through every perforation in each interval. The major constraint with this diversion mechanism is perforations erosion when proppant laden slurry pumped through it. The purpose of this paper is to provide a new model to assess the perforation friction after perforation erosion by proppant laden slurry. The methodology involves the application of artificial neural network (ANN) to predict the final hydraulic perforation diameter. In order to train, validate and test the proposed ANN model; real field data of limited entry fracturing treatments have been used. The paper reviews the previous research progress on the assessment of perforation friction. A comparison between the new proposed ANN model and the previous perforation friction correlations demonstrates that the results from ANN model are the most reliable estimation of perforation friction real data.     

用优化通量校正传输技术压制数值模拟的频散

用常规有限差分法求解波动方程,进行弹性波正演模拟,当单位波长内采样点数较少（粗网格）时会遇到严重的频散现象。通量校正传输(FCT)算法可有效地压制在粗网格情况下产生的数值频散。FCT校正假设所有的极值点都是由数值频散引起的，然后对所有网格点进行扩散通量校正处理，再对非局部极值点进行补偿的逆扩散通量校正。FCT方法用于高阶差分既具有较高的计算精度，又因适应采样间隔较大的情况而节省了计算量，从而具有较高的计算速度。在传统的FCT技术基础上提出的优化FCT技术只在需要压制数值频散处对波场进行平滑处理，可节省约40%的计算量。给出了应用优化FCT技术进行波动方程正演模拟的数值算例，当参数选取合适时不仅有效地压制了数值频散，完好地保存了真实波场，又因节省了计算量而提高了计算效率。     

Intelligent prediction of aliphatic and aromatic hydrocarbons in Caspian Sea sediment using a neural network based on particle swarm optimization

Abstract

In this paper an intelligent model is proposed to predict the amount of organic pollutants in Caspian Sea sediment based on a feed forward artificial neural network (ANN) optimized by particle swarm optimization (PSO) algorithm. Organic pollutants have carcinogenesis and mutagenesis properties which are derived from anthropogenic and natural sources. The PSO-ANN was developed by experimental data collected from different literature. The statistical parameters prove the satisfactory performance of the proposed PSO- ANN model. A good correlation was obtained between the predicted organic pollutants and the experimental data for test, train and validation data were 0.996, 0.997, 0.993, respectively.