地球物理学进展 ›› 2015, Vol. 30 ›› Issue (3): 1257-1263.doi: 10.6038/pg20150335

• 应用地球物理学Ⅰ • 上一篇    下一篇

利用BP神经网络法对致密砂岩气藏储集层复杂岩性的识别

单敬福1, 陈欣欣1, 赵忠军2,3, 葛雪1, 张芸1   

  1. 1. 长江大学油气资源与勘探技术教育部重点实验室, 武汉 430100;
    2. 中国石油长庆油田分公司苏里格气田研究中心, 西安 710018;
    3. 低渗透油气田勘探开发国家工程实验室, 西安 710018
  • 收稿日期:2014-07-09 修回日期:2014-09-25 出版日期:2015-06-20 发布日期:2015-06-02
  • 作者简介:单敬福,男,1977年生,博士后,副教授,主要从事沉积、储层与开发地质学研究工作.(E-mail:shanjingfu2003@163.com)
  • 基金资助:

    国家自然基金(41372125)、湖北省教育厅基金(Q20121210)和中国地质大学(武汉)构造与油气资源教育部重点实验室开放基金(TPR-2012-23)联合资助.

Identification of complex lithology for tight sandstone gas reservoirs sase on BP neural net

SHAN Jing-fu1, CHEN Xin-xin1, ZHAO Zhong-jun2,3, GE Xue1, ZHANG Yun1   

  1. 1. Key Laboratory of Exploration Technologies for Oil and Gas Resources, MOE, YangtzeUniversity, Wuhan 430100, China;
    2. Research Center of Sulige Gas Field, Changqing Oilfield Company, PetroChina, Xi'an 710018, China;
    3. National Engineering Laboratory for Exploration and Development of Low-Permeability Oil & Gas Fields, Xi'an 710018, China
  • Received:2014-07-09 Revised:2014-09-25 Online:2015-06-20 Published:2015-06-02

摘要:

在岩心和录井资料较少,又非常依赖测井资料进行地质综合解释的研究区域,利用测井资料进行岩性识别是一项基础而又重要的工作.测井资料的数据种类虽然较多,但对岩性敏感的曲线较少,因此,如何优选对岩性敏感的测井曲线,然后进行网络学预测岩性,则显得尤为关键.在进行BP神经网络学习前,利用已知岩心资料,优选了本研究区对岩性较为敏感的自然伽玛和光电吸收截面指数这两种测井曲线,并做标准化与归一化处理,以消除测井系列、型号和测井曲线度量单位的不同引起的刻度和数量级误差,从而提高网络收敛速度,建立准确岩性识别模型,识别了未取芯井的岩性.研究结果表明,利用优选输入向量的BP神经网络法对苏里格气田复杂岩性进行识别,识别准确率较高,平均符合率达到了近90%.因此,通过采用该方法对岩性的识别,也为后续基础性研究工作提供了宝贵的一手资料.

Abstract:

Identification of Complex Lithology is a basic and important work in the condition of less core and logging data and very dependent on geological interpretation of the study area. Although the kinds of logging data are very much, but the sensitive curveto lithology is less, so, it is particularly critical how to choose sensitive logs to lithology in order to finish the network study.Before BP neural network learning, natural gamma ray and photoelectric absorption cross-section index the two kinds of logging curves whose standardization and normalization were finishedin order to eliminate the errors from logging series, models and magnitude differences, thereby, the network convergence rate was improved and an accurate model of lithology identification was established to identify the lithology of no coring wells.The results show that using the BP neural network to identify the objective interval complex lithology in Sulige gasfield, and the recognition results are highand the average compliance rate reached nearly 90%. Therefore, the valuable first-hand data for basis geology research work was provided by using this method to identify lithology.

中图分类号: