【作者】 郭海湘；

【导师】 诸克军；

【作者基本信息】 中国地质大学 ， 资源管理工程， 2008， 博士

【摘要】 本论文的选题及研究内容获得诸克军教授主持的国家自然科学基金项目“石油勘探管理中软计算集成的理论和方法研究(NO．70573101)”和笔者主持的高等学校博士学科点专项科研基金“石油储层纵向预测软硬计算融合的理论与方法研究(NO．20070491011)”的联合资助。1996年我国产石油15729×10~4t，石油消费量为17307×10~4t，石油对外依存度为9．12％；2004年产石油17499×10~4t，石油消费量为31873×10~4t，石油对外依存度为45．10％，该年的石油消费量和石油进口量分别居世界第二位和第三位；2007年产原油18665．7×10~4t，石油消费量34593．7×10~4t，石油对外依存度为46．04％。以上数据显示：我国原油产量虽有大幅度增长，但是我们面对的主要情况是我国国民经济发展速度快，而石油产量的年增长率相对滞后，供需矛盾进一步凸显，并且随着石油勘探领域的不断扩大，储层预测面临的研究对象越来越复杂，已有的预测技术与不断提高的解释要求之间的矛盾也越来越突出。如何有效利用新的技术手段和思路，科学地预测储层的岩性和含油性，对于指导石油勘探和开发具有十分重要的实际意义，并普遍受到人们的关注。本研究以江汉油田某区块为石油储层纵向预测的案例，基于将该区块储层的测井数据转化成信息，再将信息转化成认识的动态过程，把软计算和硬计算进行融合，提出软计算和硬计算融合的模式，并且在这些融合模式下进行算法设计，对石油储层进行纵向预测：(1)在分析了软计算和硬计算的基本理论和原理的基础上，提出了软计算和硬计算的融合模式(分离模式、并行连接模式、串联模式和嵌套模式)，并分析了各个模式的特点。(2)创建三维表(三维包括搜索策略、评价标准和识别任务类别。其中，搜索策略包括：穷举法、启发方法和随机方法；评价标准模式包括：过滤模式、包裹模式和杂交模式；识别任务类别包括分类和聚类。)对属性优化方法(特征选择)进行了归纳，并且在归纳的基础上提出了算法选择平台，依据算法选择平台可以从归纳的特征选择方法中选择匹配的算法进行属性优化。(3)在储层岩性识别中通过硬计算方法对数据进行整理，然后通过算法选择平台得出匹配的算法和各种软计算和硬计算方法中的分类器进行组合，最后得出识别该区块岩性最优的属性子集为孔隙度(POR)，然后运用人工神经网络(ANN)对数据进行学习，从中提取对应各个类别的识别函数(硬计算函数)，把该识别函数作为目标函数，再用遗传算法(GA)从中进行规则提取：①如果POR是中，则该层为砂岩；②如果POR是低，则该层为砂岩；③如果POR是高，则该层为泥岩。(4)在储层含油性识别中，通过算法选择平台得出匹配的算法和各种软计算和硬计算方法中的分类器进行组合，最后得出识别该区块含油性最优的属性子集为声波时差(AC)和含油饱和度(So)，接着提出了数据驱动的灰色关联预测方法来识别储层的含油性，然后进行样本约简，最后运用人工神经网络(ANN)对数据进行学习，从中提取对应各个类别的识别函数(硬计算函数)，把该识别函数作为目标函数，再用遗传算法(GA)从中进行规则提取①如果AC是低和So是低，则该层为干层；②如果AC是中和So是低，则该层为干层；③如果AC是高和So是低，则该层为水层；④如果AC是中和So是中，则该层为差油层；⑤如果AC是低和So是中，则该层为油层；⑥如果AC是高和So是中，则该层为油层；⑦如果So是高，则该层为油层。(5)在岩性识别的最优属性子集POR和含油性最优属性子集AC和So的基础上，如果在测井数据集中没有或者缺少这些关键属性时，需要解决的问题就是怎样在不需要额外的开支下(钻井、化验等)构建预测POR，AC和So的模型。首先通过硬计算方法中的回归模型分别建立预测POR，AC和So的回归方程；然后再用软计算方法中的遗传算法和BP神经网络的嵌套(GA-BP，其中遗传算法用来优化确定BP神经网络的输入属性组合和隐含层神经元的个数)得到满意的BP神经网络预测模型来预测和识别POR，AC和So；最后对硬计算和软计算得到的模型进行比较和分析，得出GA-BP模型优于多元回归预测模型。(6)该研究区块的储层分为四个类别：①如果该层岩性为砂岩，含油性为油层，那么该储层为类别Ⅰ；②如果该层岩性为泥岩，含油性为油层，那么该储层为类别Ⅱ；③如果该层岩性为砂岩，含油性为差油层，那么该储层为类别Ⅲ；④如果该层岩性为泥岩，含油性为差油层，那么该储层为类别Ⅳ。石油储层纵向预测可以为减少石油勘探风险、准确评估石油储量、明确合理开发方案、提高石油采收率等提供极为重要的决策依据。更多还原

【Abstract】 This paper is supported by both National Natural Science Foundations of China (NSFC)funded project“A Study on Integration of Soft Computing Theories and Methods in Oilexploration Management (NO.70573101)”and the Specialized Research Fund for the DoctoralProgram of Higher Education of China“A Study on Fusion of Soft Computing and HardComputing Theories and Methods of Oil Reservoir Forecast (NO.20070491011)”.In 1996,in China oil production and consumption was 15729×10~4t and 17307×10~4trespectively,and external dependence rate was 9.12%.In 2004,oil production was 17499×10~4t,oil consumption was 31873×10~4t,and external dependence rate was 45.10%.In 2004 oilconsumption ranked the second place and oil import volume ranked the third place in the world;In 2007,oil production was 18665.7×10~4t,oil consumption was 34593.7×10~4t,and externaldependence rate was 46.04%.These data shows that although Chinese oil production increasingsharply,the annual rate of growth of oil production is relatively backward because of the rapiddevelopment of national economy,and the problems existing between supply and demand of oilare more serious.As the area of oil exploration continuously enlarge,the research objects of theforecast of reservoir are more and more complicated,the contradictions between the forecasttechnique existing and the explanation demand constantly increasing are prominent.How toeffectively utilize new technology and idea to forecast lithology and oiliness of all kinds of oilreservoirs has important practical significance for directing oil exploration,and causes extensiveconcern.This research takes a certain block in Jianghan oil field as the case of oil reservoir forecast,based on the dynamic process of transforming well logging data into information,thentransforming information into acknowledge,and fusing soft computing with hard computing,putforward the patterns of fusion of soft computing and hard computing,and design algorithm underthe fusion patterns to forecast oil reservoir:(1)On basis of analyzing the fundamental theory and principle of soft computing and hardcomputing,put forward the patterns of fusion of soft computing and hard computing (isolatedpattern、parallel pattern、cascaded pattern and nested pattern),and analyze the feature of allpatterns. (2)Developing a three dimensional categorizing framework.One dimension is searchstrategy including complete search,sequential search and random search.Another dimension isevaluation criteria including filter model,the wrapper model and the hybrid model.The thirddimension is task including clustering with unlabeled data (unsupervised feature selection)andclassification with labeled data (supervised feature selection)and put forward algorithm selectionplatform,and we can select matched algorithm to optimize attributes from the induced featureselection methods according to algorithm selection platform.(3)Analyzing data with hard computing methods in the recognition of reservoir lithology,then combine the matching algorithm which was selected from categorizing framework based onalgorithm selection platform with classifiers of various soft computing methods and hardcomputing methods,and get the optimal attribute set of recognition lithology in this block is POR.Training the data with ANN and extract the corresponding recognition functions (hard computingfunctions).Then use recognition function as objective function for rule extraction with GA.①If POR is at middle level,this layer is sandstone;②If POR is at low level,this layer issandstone;③If POR is at high level,this layer is mudstone.(4)Combining the matching algorithm which was selected from categorizing frameworkbased on algorithm selection platform with classifiers of various soft computing methods andhard computing methods,and get the optimal attribute set of oiliness in this block is AC and So,then propose a method of data driven gray relational analysis for recognizing oil-bearingcharacteristics in reservoir and reduce samples,at last training the data with ANN and extract thecorresponding recognition functions (hard computing functions).Use recognition function asobjective function for rule extraction with GA.①If both of AC and So are at low level,this isdry layer;②If AC is at middle level and So is at low level,this is dry layer;③If AC is athigh level and So is at low level,this is water layer;④If both of AC and So are at middle level,this layer is inferior oil layer;⑤If AC is at low level and So is at middle level,this layer is oillayer;⑥If AC is high level and So is at middle level,this layer is oil layer;⑦if So is at highlevel,this layer is oil layer.(5)Based on the optimal attribute set of recognition lithology is POR and the optimalattribute set of oiliness is AC and So in this block,the problem to be solved is how to establishand predict the models of POR,AC and So without additional expenditure (well drilling,chemical examination etc.)if there is no or not enough key attributes in the data set of welllogging.Firstly,establish the regressive equations for the prediction of POR,AC and So withregression model of hard computing methods.Then,obtain satisfactory neural network model ofBP to predict and recognize POR,AC and So by nested genetic algorithm of soft computingmethods and BP neural network (The genetic algorithm is used to optimize the input attributescombination of BP neural network and determine the number of neurons in the hidden layer).Lastly,compare the models which are obtained from hard computing methods and softcomputing methods and conclude that GA-BP model is better than multiple regression model.(6)This research block are divided into four classes:①If the lithology of this ayer issandstone and the oiliness is oil layer,the reservoir belongs to classⅠ;②If the lithology of this layer is mudstone and the oiliness is oil layer,the reservoir belongs to classⅡ;③If thelithology of this layer is sandstone and the oiliness is inferior oil layer,the reservoir belongs toclassⅢ;④If the lithology of this layer is mudstone and the oiliness is inferior oil layer,thereservoir belongs to classⅣ.The prediction of oil reservoir provides important decision basis to reduce oil explorationrisks,evaluate oil reserves correctly,clear scientific development plan and increase the oilrecovery rate.更多还原