石油学报 ›› 2013, Vol. 34 ›› Issue (2): 312-322.DOI: 10.7623/syxb201302013

• 地质勘探 • 上一篇    下一篇

火山岩岩性的支持向量机识别

朱怡翔   石广仁   

  • 收稿日期:2012-07-04 修回日期:2012-10-28 出版日期:2013-03-25 发布日期:2013-01-31
  • 通讯作者: 朱怡翔
  • 作者简介:朱怡翔,男,1960年2月生,1982年获合肥工业大学学士学位,2004年获中国地质大学(北京)博士学位,现为中国石油勘探开发研究院教授级高级工程师、博士生导师,主要从事储层地质、地震和测井的综合表征和预测、复杂岩性储层的测井评价等研究工作。
  • 基金资助:

    国家重大基础研究发展计划(973)项目“高效天然气藏形成分布与凝析、低效气藏经济开发的基础研究”(2008CB209100)资助。

Identification of lithologic characteristics of volcanic rocks by support vector machine

ZHU Yixiang  SHI Guangren   

  • Received:2012-07-04 Revised:2012-10-28 Online:2013-03-25 Published:2013-01-31

摘要:

提出了用9种火山岩的岩石类型描述火山岩储层岩性的模型,表达岩性对优质储层的控制作用。基于该模型,选取了对火山岩的岩性、组构、成因和孔隙结构反应灵敏的15种岩石物理测井参数,分别采用多元回归分析(MRA)、人工神经网络(ANN)和支持向量机(SVM)3种机器学习算法,尝试火山岩岩性的识别。在三塘湖盆地马朗凹陷牛东油田的实例中,使用了3口井的数据,其中N9-10井和N9-19井的火山岩储层为学习样本,N8-10井的火山岩储层为预测样本。利用N9-10井1361个样本和N9-19井881个样本(每个样本含15种测井参数及岩性),通过这3种机器学习算法分别获得预测火山岩岩性的知识;然后,利用N8-10井961个样本(每个样本仅含15种测井参数),根据上述学习获得的知识,得到这961个样本的岩性。研究发现:对于学习样本,MRA、ANN和SVM的计算与实际的平均相对误差绝对值分别为51.84%,48.66%和0;对于预测样本,则分别为52.44%,46.31%和6.30%。实例分析表明,只有SVM适用于本实例,这是由于火山岩岩性与15种岩石物理测井参数的非线性关系十分强烈。

关键词: 火山岩, 岩性划分, 储层, 岩石物理测井, 多元回归分析, 人工神经网络, 支持向量机

Abstract:

A model to characterize lithology of volcanic rock reservoirs by using nine types of volcanic rocks was presented to indicate lithologic control over high-quality reservoirs. Based on this model, we selected 15 petrogeophysical logging parameters sensitive to lithology, fabrics, genesis and pore structures of volcanic rocks, and adopted three machine learning algorithms, i.e. multiple regression analysis (MRA), artificial neural network (ANN) and support vector machine (SVM), respectively, to identify lithologic characteristics of volcanic rocks. Taking the Niudong oilfield of the Malang sag in the Santanghu Basin as an example, we employed data from three wells where volcanic rock reservoirs in Well N9-10 and Well N9-19 served as learning samples while that in Well N8-10 as a prediction sample. In particular, 1361 samples from Well N9-10 and 881 samples from Well N9-19 were employed and each of them contained 15 parameters of logging and lithology, the knowledge to predict lithologic characteristics of volcanic rocks could be obtained, respectively, by these algorithms. Then, 961 samples from Well N8-10 were used and each sample only had 15 logging parameters, while their lithologic characteristics were gained based on the aforementioned knowledge obtained by learning. The result shows that as for the learning samples, the absolute value of mean relative errors (%) between calculated and field-measured results for MRA, ANN and SVM are 51.84%, 48.66% and 0, respectively; and as for the prediction samples, these errors are 52.44%, 46.31% and 6.30%, respectively. Therefore, in this case only SVM is applicable because a nonlinear relationship between lithologic characteristics of volcanic rocks and 15 petrogeophysical logging parameters is very strong.

Key words: volcanic rock, lithologic division, reservoir, petrogeophysical logging, multiple regression analysis (MRA), artificial neural network (ANN), support vector machine (SVM)