自组织竞争神经网络在砂岩型铀矿测井数据解释中的应用研究

发布时间：2018-01-20 03:37

本文关键词： 砂岩型铀矿测井解释岩性识别 SOM神经网络分层定量解释　出处：《东华理工大学》2015年硕士论文　论文类型：学位论文

【摘要】：随着科学技术的进步与发展,测井的解释方法与技术也在日益提高。测井资料解释处理方法的发展倾向于综合化、人机交互以及全自动化解释等。当前,更高效、快速、准确的测井数据综合解释的方法成为了岩性识别研究的方向。岩性识别作为测井曲线解释中的关键部分,然而传统人工识别岩性方法,解释周期长、精度低,导致主观因素对解释结果的影响很大。因此,本文利用自组织竞争型神经网络识别方法,对测井数据进行综合岩性识别,以实现挖掘测井数据潜力的目的。本文研究了某盆地铀矿床的区域地质概况。在结合砂岩型铀矿床的特征及其放射性平衡的规律的基础上,总结了地层岩性与各测井曲线幅值的关系。结合前人研究成果,并分析了两种标准学习样本指标选择的应用效果,决定了样本的各项指标。根据实际可用测井资料,以SP、LL3、CAL、DEN测井数据作为学习样本指标,建立了钻井资料中6种常见岩性的标准样本,即泥岩、粉砂岩、细砂岩、中砂岩、粗砂岩和煤层的标准学习样本,并对需要解释钻孔的测井数据进行标准化。通过利用自组织竞争型神经网络方法在砂岩型铀矿床钻井资料中的应用,划分了地层及其岩性。研究表明,自组织竞争型神经网络方法在岩性的分类识别方面有着有较快的收敛速度和较高识别率的优势。与专家编录的岩性剖面相比较,自动识别的岩性符合地质规律,其岩性分类的结果满足工作的要求。根据核测井资料定量解释原理采用分层反褶积解释的方法,编写了求取相关参数的函数代码,对井中异常矿化段进行了伽马定量解释。结合生产实践中的所用的给定含量法,对比讨论了给定含量法与三点式反褶积法、五点式反褶积法计算的含量。结果表明,与生产中所有的含量解释方法相比,五点式反褶积比给定含量法的解释结果精度更高。
[Abstract]:With the progress and development of science and technology, logging interpretation methods and techniques are also increasing. The development of logging interpretation and processing methods tend to be integrated, man-machine interaction and fully automated interpretation. More efficient, fast and accurate comprehensive interpretation of logging data has become the research direction of lithology recognition. Lithology recognition is a key part of log curve interpretation, but traditional manual lithology recognition method. The interpretation period is long and the precision is low, which leads to the subjective factors have great influence on the interpretation results. Therefore, the self-organizing competitive neural network recognition method is used to identify the lithology of logging data. In order to excavate the potential of logging data, this paper studies the regional geological situation of uranium deposits in a certain basin, on the basis of combining the characteristics of sandstone-type uranium deposits and the regularity of radioactive balance. The relationship between formation lithology and the amplitude of logging curves is summarized. Combined with the previous research results, the application effect of selecting two standard learning sample indexes is analyzed. According to the available logging data, the SPLL3CALDEN logging data is taken as the learning sample index, and six common lithologic standard samples in drilling data are established. Standard learning samples of mudstone siltstone fine sandstone medium sandstone coarse sandstone and coal seam. Through the application of self-organizing competitive neural network method to the drilling data of sandstone-type uranium deposits, the strata and their lithology are divided. The self-organizing competitive neural network method has the advantages of faster convergence speed and higher recognition rate in the classification and recognition of lithology. Compared with the lithologic profile recorded by experts, the lithology of automatic recognition accords with geological law. The results of lithologic classification meet the requirements of the work. According to the principle of quantitative interpretation of nuclear logging data, the functional codes for calculating the relevant parameters are compiled by using the method of stratified deconvolution interpretation. The gamma quantitative interpretation of abnormal mineralized section in wells is carried out. Combined with the method of given content used in production practice, the method of given content and the method of three-point deconvolution are compared and discussed. The results show that the interpretation accuracy of the five-point deconvolution method is higher than that of the given content deconvolution method compared with all the content interpretation methods in production.
【学位授予单位】：东华理工大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：P619.14;P631.81

【相似文献】