【作者】 陆万雨；

【导师】 管志宁； 宋子齐；

【作者基本信息】 中国地质大学（北京） ， 地球探测与信息技术， 2003， 博士

【摘要】 测井储层评价研究是剩余油研究的核心内容之一，本文结合“新疆克拉玛依油田八区克上组砾岩油藏中高含水期剩余油研究”(地质模型部分)课题，开展了测井储层评价新技术的应用研究，研究内容包括测井解释方法研究、储层评价方法研究、储层非均质性研究、储层流动单元研究、储层三维建模等。 测井解释方法研究主要侧重于神经网络测井解释方法的研究，通过对BP算法的分析，结合测井解释的特点，提出了改进方案：采用变学习率、增加“动量”项、S函数输出限幅、增加陡度因子，并在学习样本选取方面采取了剔除异常点、减少具有相同特征的样本的数量、加入反算样本等措施，通过区块测井数据分析、岩心资料深度归位及饱和度校正，提取出有代表性的学习样本，建立了能够适应研究区砾岩油藏克上组S1-3、S4-5层组测井储层参数提取计算的统计分析解释模型和神经网络解释模型，改进后的BP神经网络用于测井解释时达到了项目测井储层参数计算精度的要求； 测井储层评价研究是本文研究的重点，这一部分的方法研究主要有两部分：灰色系统理论方法研究和CP神经网络方法研究。利用灰色系统理论新的思维方式，建立测井地质关系数据库，选择、匹配、拟合和提取砾岩油藏岩性、物性和含油气性评价的特征性参数，建立评价砾岩油藏储层的解释标准及权系数，利用矩阵作数据列处理并采用最大隶属原则得出灰色评价预测结论；CP神经网络方法为首次引入，针对测井储层评价的特点，对CP神经网络进行了改进研究，采取了打乱学习样本输入顺序、增加误差处理过程、增设网络测试功能、提出中间层神经元的经验取值等措施，开发了相应的应用软件，在实用中取得了很好的效果；最后，根据研究区的具体情况采用灰色系统理论和CP神经网络相结合的方法开展了测井储层评价研究，经与试油、试采情况进行对比，符合率达到了85％； 本研究区的一大特点是储层非均质性十分严重，为此展开了储层非均质性研究，研究重点是采用分形理论研究本区的平面非均质性，应用分形克里金方法进行井间插值，制作了各小层的孔隙度、渗透率、含油饱和度、有效厚度平面分布图件和隔、夹层厚度平面分布图，这些研究有效地了解了储层参数在储层空间上的分布情况和储层各层组的连通情况以及隔夹层的分布情况，为该区进一步开发调整提供了较可靠的依据； 为了进一步了解研究区储层及其内部流体流动特征，本文还开展了储层流动单元研究；从储层孔喉几何学角度出发，采用储层流动带指标(FZI)来划分流动单元，采用聚类分析的方法并结合岩性、物性等资料划分出了五类流动单元，并通过对取心井流动单元对应的测井参数的统计分析，建立起未取心井的流动单元预测模型，计算了各井的流动单元指标值；最后结合项目的沉积微相研究成果，制作了各小层流动单元分布图并进行了描述，这些研究对储层岩性、物性及非均质性、小层连通性等均有了进一步的认识，为油田的开发及调整挖潜提供了依据，也为后面的储层地质建模研究奠定了基础； 最后，在前面各项研究工作的基础上开展了储层三维地质模型研究；采用“二次建模”的方法进行储层参数建模，采用地质统计学克里金方法进行井间插值，建模软件采用的是Gridstat，按照软件运行的要求组织了各种数据，分别建立了储层孔隙度、渗透率、含油饱和度、泥质含量的三维地质模型，并分东北—西南向的8215—151井和东南—西北方向的8289—8237井制作了参数分布剖面图，这些图件形象直观的展示了研究区的储层属性；利用网络积分法，结合克里金估值技术和储层三维地质建模的成果，进行了研究区油藏储量计算研究，分两次进行了储量计算，与射孔方案计算的储量(1266.0×10~4t)相比，第 0 一次储量计算 （123．gX10‘t）偏低，第二次储量计算 （135刀X104 t）偏高，第二次储量 计算结果与当前实际生产状况趋于吻合；最后对研究区各小层油藏储量及采出情况进行了 分析研究，指出本区真正具有挖潜潜力的主要对象已不在S4‘－S。‘层组，而应在上部的S；’－ S。‘层组（占整个剩余可采储量的74．5％），其次在S。’－S。’层组（它的剩余可采储量占整个 S。’－S。’层组剩余开采储量的85．3％）。 本文直接针对油田生产中的一些实际问题，较系统地将神经网络、灰色理论、分形理 论、流动单元、三维建模等新技术方法综合应用于测井储层评价研究，为油田生产提供了 相应的技术理论和方法并促进了这些新技术方法的推广和应用，本文的研究也将为其他研 究者提供借鉴和参考。更多还原

【Abstract】 The logging reservoir evaluation is one of the main contents of the remaining oil research. This dissertation develops the new technique applied research in logging reservoir evaluation based on the research project "The middle-high containing water oil field’s remaining oil research of conglomerate petroleum deposit in Keshangzu, No.8 district, Xinjian Karamay oil field", the research includes the logging interpretation, the reservoir evaluation, inhomogeneity, flow unit, 3D modeling, and so on.The neural network application research is the main method used in the logging interpretation. Based on analyzing the BP neural network’s algorithm and combing the feature of the logging, a scheme was put forward. It adopted variable learning rate, added a dynamic term and a steep factor, limited the S-function’s output amplitude. In taking the learning samples, we eliminated the anomaly point, reduced the number of the samples with similar feature, and used the reconstruction samples. Through logging data analysis, rock depth immigration, saturation correction, some representative learning samples were selected, the model of the statistical analysis and the BP-neural network were established, and this model is applicable to the studied area.The reservoir evaluation is the major part of this dissertation. It is composed of two parts, the gray-theory and the CP-neural network. Using the new concept of gray-theory, the logging-geology database was built. The evaluation feature parameter, including the conglomerate petroleum deposit’s lithology, geophysical properties and oiliness, was selected, matched, modeled and extracted, and used to establish the interpretation standard and weight coefficients. Then the evaluation result was got by using the matrix transformation and the max subjection principle. The CP-neural network is used firstly in the logging reservoir evaluation. For the features of logging reservoir evaluation, the CP-neural network is modified by confusing the learning-samples, adding the error processing and test phase, and supposing the experienced cell number of the middle layer in the CP neural network. Based upon the ideal above, the CP-neural network’s software was developed. It works well in the research. Combining the gray-theory with the CP-neural network, the logging reservoir evaluation in the studied area was complished. The coincidence between the evaluation and the oil testing is about 85%.One of the obvious features of the studied area is the reservoir’s inhomogeneity. The fractal theory was used to deal with the problem. The plot in plane view of each thin layer’s porosity, perm, oil saturation, net thickness and the separate/inter layer’s thickness was obtained. This will help to understand the space distribution of the parameters, and provide evidences for the oil field development.In order to get more information about the reservoir’s inner flow liquid, someresearch were done on the reservoir flow unit. Through the reservoir pore geometry, the flow zone index (FZI) was adopted to identify the flow unit. The flow unit is divided into 5 classes with the cluster analysis combined with its lithology and geophysical properties. The unlifting core well’s FZI was forecast by statistical analyzed the lifting core well’s FZI. Combined with the result of the micro-Sedimentary facies, each thin layer’s flow unit distribution was obtained and descripted. This will help to get more information on the reservoir lithology, geophysical properties, inhomogeneities and the thin layer’s connectivity.Finally, based on the research above, the 3D model was built. Using the Gridstat’s software to build the 3D model of the reservoir’s porosity, perm, oil saturation and mud content. Two cross-sections were acquired, one from Well-151 to Well-8215, another from Well-8237 to Well-8239. Combined with the Kriging techniques and the 3D modeling result, the petroleum deposit reserve was calculated two times by the grid-integral. Compared with the calculated result of the perforating technique (1266.0#10更多还原