常用数学物理方法在测井解释中的应用
发布时间：2018-08-11 16:14
【摘要】：本文主要研究一些常用数学物理方法在地层划分、岩性识别中的实际应用效果。地层划分是测井解释中的一个重要环节,人工分层不仅效率低下,而且易受解释人员的主观因素影响。本文利用极值方差聚类法、活度函数法、微商法并结合计算机来自动分层。这些分层方法的优点是快速、定量、自动的进行分层；还可以同时根据多条不同的测井曲线所提供的数据进行综合分层。当然这些分层法效果的好坏,很大程度上取决于所选用的数值处理方法是否恰当。 在岩性识别方法建立过程中,尝试了交会图法,但只能划分岩性大类或指出变化趋势,效果不佳。为此,本文采用贝叶斯判别法、支持向量机两种算法,结合编程实现岩性的计算机自动识别,效果明显,两种方法的平均符合率都达到了80%以上,基本满足勘探开发需要,可以作为测井岩性识别的主要方法在生产实际中应用。 本文主要研究区块及储层为呼和诺仁油田(贝301区块)南屯组二段砂砾岩储层。该区块储层总体上属于低孔、低渗、低含油饱和度,并且断层多；储层岩性主要为绿灰、灰色泥质粉砂岩、粉砂岩、细砂岩、粗砂岩、砂砾岩、砾岩,呈不等厚互层。在了解贝301区块的区域地质概况的基础上,本文从储层测井响应特征和储层岩性特征入手,结合常用的数学物理方法对研究区块的测井响应参数建立数学模型,进而进行地层分层和岩性识别。如在用活度函数法分层中,选取在泥质层和砂岩层的响应特征比较明显的自然伽马曲线,并设置了窗长为3、7、9三组活度曲线来进行比较识别,得出窗长7和9的分层效果更好。在用贝叶斯判别法识别岩性过程中,特征参数的选取至关重要,经分析岩性响应特征得出砂质砾岩,细砂岩,泥质粉砂岩的AC(声波时差测井)、ILD(深感应测井)、TLM(中感应测井)、SFL(球形聚焦电阻率测井)、GR(自然伽马测井)响应均值在不同岩性段上变化最为明显,故在此综合这五条测井曲线值作为特征参数建立贝叶斯判别模型,并用来进行岩性识别,判别结果经与录井岩性资料对比可得正判率超过80%,说明判别方法具有实用价值。通过本次研究,主要取得了以下成果与认识： 1、测井曲线的形态与变化规律受岩性控制外,还与井下仪器类型、测井速度、井眼条件等有关。可将它们归总为与岩层中心对称和不对称两大基本类型。 2、本文所采用的极值方差聚类分层法原理简单,易于编程,运算速度快,人工参与少,分层取值结果合理,效果明显。此外其适用性较强,对任何测井曲线(如SP,GR,AC,DEN,CNL,CAL,RA)都适用。 3、本文用活度极大值作为地层界面只适用于自然电位、自然伽马这类半幅点分层的测井曲线,而不适用于侧向、梯度这类不用半幅点分层的测井曲线。文中对选取了3、7、9三种活度窗长,处理结果与人工分层结果基本一致,体现了算法的有效性,能较好地达到测井自动地质分层的目的。 4、本文在进行岩性识别时需要注意对数据进行归一化处理,消除各测井参数间因量纲不同而引起的潜在问题,其次在选取测井参数时,应选用反映岩性变化能力大的测井信息。建立样本模型后,为检验岩性判别函数的可靠性,对确定岩性判别函数的样品数据进行回判,其结果如下：灰色泥质粉砂岩、细砂岩、砂质砾岩的正判率分别为85.05%,81.19%,85.7%,整个样品层的正判率为83.33%。说明贝叶斯判别法在综合运用地质、测井、岩心资料进行岩性预测时,能够得到较好的结果。 5、支持向量机方法在解决模式识别小样本、非线性及高维中表现出独特的优势和良好的应用前景。使用C语言编写处理程序,对每种岩性样本进行SVM训练,然后利用学习后的SVM模型预测储层岩性,可知,采用SVM识别贝301区块地层的岩性与实际取心资料对比,符合率为85%,特别是对泥质粉砂岩这种岩性的划分可达到90%以上,说明利用该方法可识别油藏地质中的复杂岩性,提高划分精度。
[Abstract]:This paper mainly studies the practical application effect of some commonly used mathematical and physical methods in stratigraphic division and lithologic identification.Stratigraphic division is an important link in logging interpretation.Artificial stratification is not only inefficient but also susceptible to the subjective factors of interpreters.In this paper,extreme variance clustering method,activity function method and derivative method are combined. The advantages of these stratification methods are fast, quantitative, and automatic stratification, and they can also be synthetically stratified according to the data provided by several different logging curves. Of course, the effectiveness of these stratification methods depends largely on the appropriateness of the numerical processing methods selected.
In the process of establishing lithology identification method, the intersection diagram method is tried, but the effect is not good because it can only classify lithology into big categories or point out the changing trend. Therefore, this paper adopts Bayesian discriminant method and Support Vector Machine (SVM) algorithm to realize automatic lithology identification by computer, and the effect is obvious. The average coincidence rate of the two methods is 80%. On the other hand, it can basically meet the needs of exploration and development, and can be used as the main method of lithology identification in production practice.
This paper mainly studies the sandy conglomerate reservoir of the second member of Nantun Formation in Hohhot Noren Oilfield (Bei 301 Block).The reservoir in this block is generally of low porosity, low permeability, low oil saturation and many faults.The reservoir lithology is mainly composed of green ash, gray argillaceous siltstone, siltstone, fine sandstone, coarse sandstone, sandy conglomerate and conglomerate, which are interbedded with unequal thickness. On the basis of understanding the regional geological situation of Bei 301 block, this paper starts with the logging response characteristics and reservoir lithology characteristics of the reservoir, establishes a mathematical model for the logging response parameters of the study block in combination with common mathematical and physical methods, and then carries out stratigraphic stratification and lithology identification. Natural gamma-ray curves with obvious response characteristics of rock strata are set up, and three groups of activity curves with window lengths of 3,7,9 are set for comparison and identification. It is concluded that window lengths of 7 and 9 are better for stratification. Acoustic moveout logging, ILD, TLM, SFL and GR are the most obvious changes in different lithologic sections. Therefore, Bayesian discriminant model is established by synthesizing these five logging curves as characteristic parameters and used for lithologic identification. By comparing the discriminant results with logging lithologic data, the positive rate is more than 80%, which shows that the discriminant method is of practical value.
1. The shape and variation of logging curves are controlled by lithology, and are also related to downhole tool type, logging speed and borehole conditions. They can be grouped into two basic types: symmetrical and asymmetrical with the center of the strata.
2. The extremum variance clustering stratification method adopted in this paper is simple in principle, easy to program, fast in operation, less manual participation, reasonable in stratification and obvious in effect.
3. In this paper, the maximum activity is used as the formation interface, which is only suitable for half-amplitude stratified logging curves such as spontaneous potential and natural gamma, but not for lateral and gradient logging curves without half-amplitude stratification. It can achieve the purpose of automatic geological stratification.
4. In this paper, the data should be normalized to eliminate the potential problems caused by the different dimension of each logging parameter. Secondly, when choosing logging parameters, the logging information which reflects the large lithologic variation ability should be selected. Sample data of discriminant function are judged back, and the results are as follows: the positive rate of grey argillaceous siltstone, fine sandstone and sandy conglomerate are 85.05%, 81.19%, 85.7% respectively, and the positive rate of the whole sample layer is 83.33%. It shows that Bayesian discriminant method can get better results when comprehensive application of geology, logging and core data to predict lithology.
5. Support Vector Machine (SVM) has unique advantages and good application prospects in solving small sample, non-linearity and high dimension of pattern recognition. Each lithology sample is trained by SVM in C language, and then the reservoir lithology is predicted by SVM model. It is known that the lithology of Bei 301 block is recognized by SVM. Comparing the actual coring data, the coincidence rate is 85%, especially for argillaceous siltstone, which can be divided into more than 90%, indicating that the method can identify the complex lithology in reservoir geology and improve the division accuracy.
【学位授予单位】：长江大学
【学位级别】：硕士
【学位授予年份】：2012
【分类号】：P618.13;P631.81
[Abstract]:This paper mainly studies the practical application effect of some commonly used mathematical and physical methods in stratigraphic division and lithologic identification.Stratigraphic division is an important link in logging interpretation.Artificial stratification is not only inefficient but also susceptible to the subjective factors of interpreters.In this paper,extreme variance clustering method,activity function method and derivative method are combined. The advantages of these stratification methods are fast, quantitative, and automatic stratification, and they can also be synthetically stratified according to the data provided by several different logging curves. Of course, the effectiveness of these stratification methods depends largely on the appropriateness of the numerical processing methods selected.
In the process of establishing lithology identification method, the intersection diagram method is tried, but the effect is not good because it can only classify lithology into big categories or point out the changing trend. Therefore, this paper adopts Bayesian discriminant method and Support Vector Machine (SVM) algorithm to realize automatic lithology identification by computer, and the effect is obvious. The average coincidence rate of the two methods is 80%. On the other hand, it can basically meet the needs of exploration and development, and can be used as the main method of lithology identification in production practice.
This paper mainly studies the sandy conglomerate reservoir of the second member of Nantun Formation in Hohhot Noren Oilfield (Bei 301 Block).The reservoir in this block is generally of low porosity, low permeability, low oil saturation and many faults.The reservoir lithology is mainly composed of green ash, gray argillaceous siltstone, siltstone, fine sandstone, coarse sandstone, sandy conglomerate and conglomerate, which are interbedded with unequal thickness. On the basis of understanding the regional geological situation of Bei 301 block, this paper starts with the logging response characteristics and reservoir lithology characteristics of the reservoir, establishes a mathematical model for the logging response parameters of the study block in combination with common mathematical and physical methods, and then carries out stratigraphic stratification and lithology identification. Natural gamma-ray curves with obvious response characteristics of rock strata are set up, and three groups of activity curves with window lengths of 3,7,9 are set for comparison and identification. It is concluded that window lengths of 7 and 9 are better for stratification. Acoustic moveout logging, ILD, TLM, SFL and GR are the most obvious changes in different lithologic sections. Therefore, Bayesian discriminant model is established by synthesizing these five logging curves as characteristic parameters and used for lithologic identification. By comparing the discriminant results with logging lithologic data, the positive rate is more than 80%, which shows that the discriminant method is of practical value.
1. The shape and variation of logging curves are controlled by lithology, and are also related to downhole tool type, logging speed and borehole conditions. They can be grouped into two basic types: symmetrical and asymmetrical with the center of the strata.
2. The extremum variance clustering stratification method adopted in this paper is simple in principle, easy to program, fast in operation, less manual participation, reasonable in stratification and obvious in effect.
3. In this paper, the maximum activity is used as the formation interface, which is only suitable for half-amplitude stratified logging curves such as spontaneous potential and natural gamma, but not for lateral and gradient logging curves without half-amplitude stratification. It can achieve the purpose of automatic geological stratification.
4. In this paper, the data should be normalized to eliminate the potential problems caused by the different dimension of each logging parameter. Secondly, when choosing logging parameters, the logging information which reflects the large lithologic variation ability should be selected. Sample data of discriminant function are judged back, and the results are as follows: the positive rate of grey argillaceous siltstone, fine sandstone and sandy conglomerate are 85.05%, 81.19%, 85.7% respectively, and the positive rate of the whole sample layer is 83.33%. It shows that Bayesian discriminant method can get better results when comprehensive application of geology, logging and core data to predict lithology.
5. Support Vector Machine (SVM) has unique advantages and good application prospects in solving small sample, non-linearity and high dimension of pattern recognition. Each lithology sample is trained by SVM in C language, and then the reservoir lithology is predicted by SVM model. It is known that the lithology of Bei 301 block is recognized by SVM. Comparing the actual coring data, the coincidence rate is 85%, especially for argillaceous siltstone, which can be divided into more than 90%, indicating that the method can identify the complex lithology in reservoir geology and improve the division accuracy.
【学位授予单位】：长江大学
【学位级别】：硕士
【学位授予年份】：2012
【分类号】：P618.13;P631.81
【参考文献】
相关期刊论文 前10条
1 冯敬英;肖慈s，
本文编号：2177520
本文链接：https://www.wllwen.com/falvlunwen/sflw/2177520.html
