摘要
基于测井数据分析的岩性识别是油气地球物理勘探的基础性问题之一.支持向量机(SVM)是目前分辨能力最高的岩性识别方法,特征优化可提高SVM的识别正确率,该方法采用主成分分析(PCA)进行目标特征提取,易受特征选取的影响.针对诸如白云岩和灰岩的区分等测井响应差别不明显,岩性区分困难而又必须进行的问题,本文引入概率生成模型连续限制玻尔兹曼机(CRBM)进行测井数据岩性特征提取,然后再运用SVM在提取特征上进行岩性识别.运用所发展的CRBMSVM进行川西海相灰岩及白云岩的识别,正确率达到了81.9%.基于同样的支持向量机,CRBM提取特征识别正确率要高于PCA提取特征.
Lithology identification based on well logging data analysis is one of the basic problems for the geophysical exploration of oil and gas. Support vector machine（ SVM） is a highest currently resolution of lithology identification method.SVM is vulnerable to the influence of the feature selection when it use principal component analysis（ PCA） to extract target feature. For the problem of lithology identification difficulties such as limestone and dolomite with slight difference in well logging response identification, this paper introduced a probability generation model—continuous restricted Boltzmann machine（ CRBM） to extract new features from well logging response. Then the identifying lithology model of SVM is constructed by new features. In experiment,the model of combined CRBM with SVM applied to western Sichuan Marine limestone and dolomite identification. The coincidence rate of lithology identification and practical coring data reaches 81. 9%. Based on the same support vector machine, the recognition accuracy of CRBM feature extraction method than PCA feature extraction.
出处
《地球物理学进展》
CSCD
北大核心
2016年第2期821-828,共8页
Progress in Geophysics
基金
国家自然科学基金项目(41430323
41274128)
油气藏地质开发工程国家重点实验室自主探索课题联合资助
关键词
岩性识别
支持向量机
特征优化
连续限制玻尔兹曼机
主成分分析
lithology identification
support vector machine
feature optimization
continuous restricted Boltzmann machine
principal component analysis
