核Fisher判别分析在火山岩岩性识别中的应用  

Identification of Volcanic Rock Based on Kernel Fisher Discriminant Analysis

作　　者：王鹏[1] 王志章[1] 纪友亮[1] 段文浩 潘潞[1] 

机构地区：[1]中国石油大学(北京)油气资源与探测国家重点实验室,北京102249 [2]中国石油煤层气有限责任公司,北京100028

出　　处：《测井技术》2015年第3期390-394,共5页Well Logging Technology

摘　　要：针对火山岩储层岩性识别困难的问题,将核Fisher(KFDA)判别分析方法用于火山岩储层的岩性识别中。在详细介绍核Fisher(KFDA)判别分析方法的理论基础上,选取能够反映火山岩岩性变化的常规测井曲线作为特征变量,从8口取心井中随机选取70%的数据样本点参与核Fisher(KFDA)的模型建立,剩余30%数据样本点作为测试集进行岩性正确率的识别。通过对克拉美丽气田石炭系349块火山岩薄片的数据分析表明,利用核Fisher判别分析方法可以有效实现火山岩岩性的分离,与支持向量机(SVM)、BP神经网络和Fisher(FDA)方法比较,该方法识别准确率高,图形化显示方面直观性强,可作为利用常规测井资料识别火山岩岩性的有效手段。In view of the difficult lithology identification of volcanic reservoir, Kernel Fisher （KFDA） discriminant analysis method is used in this article to identify the reservoir lithology. Based on the theory of Kernel Fisher （KFDA） discriminant analysis method, we selected conventional logging curves which can reflect lithology changes in volcanic rocks in the studied area as characteristic variables, and used 70% of the data sample points randomly selected in model building of Fisher （KFDA） from 8 coring wells, the remaining 30% as a test set to identify the correct rate of lithology. Thin flake data analysis of 345 blocks of volcanic rocks of carboniferous from Kelameili gas field showed that the use of kernel Fisher method can effectively achieve the separation of the lithology of volcanic rocks. Compared with methods of Support Vector Machine （SVM）, BP neural network and Fisher （FDA）, this method has higher recognition accuracy and stronger intuitiveness in terms of graphical display, thus can be used as an effective mean using conventional log data to identify lithology of volcanic rocks.

关 键 词：测井解释 火山岩储层 核FISHER判别 岩性识别 测井资料 克拉美丽气田 

分 类 号：P631.84[天文地球—地质矿产勘探]