核主成分分析与支持向量机模型在储层识别中的应用  

THE APPLICATION OF A KPCA-AVM MODEL TO RESERVOIR IDENTIFICATION

作　　者：庞河清[1] 匡建超[1,2] 王众[1] 刘海松[1] 蔡左花[3] 黄耀综 

机构地区：[1]成都理工大学能源学院,四川成都610059 [2]成都理工大学管理科学学院,四川成都610059 [3]中石化西南油气分公司勘探开发研究院贵阳所,贵州贵阳550004 [4]中石化胜利油田孤东采油4矿29队,山东东营257237

出　　处：《物探与化探》2012年第6期1001-1005,1013,共6页Geophysical and Geochemical Exploration

基　　金：教育部规划基金(11YJAZH043);四川石油天然气研究中心(川油气科SKA09-01)项目联合资助

摘　　要：针对低孔、低渗致密储层识别较常规储层难这一问题,首次应用核主成分分析与支持向量机(KPCA-SVM)模型进行储层识别。该模型先通过核主成分分析(KPCA)进行非线性特征参数提取,然后将提取的特征参数作为支持向量机(SVM)的输入变量,最终实现储层识别。由于KPCA-SVM模型集成了核函数、主成分和支持向量分类机的优点,较好地解决非线性小样本的问题,能消除数据之间的噪音,降低维数,而又不缺失有效信息,达到准确快速预测的功能。将该模型应用到新场须二气藏新856井区储层预测中,预测结果验证了本模型的优越性,可作为致密储层预测的可选方法。It is more difficult to predict the low porosity and low permeability tight reservoir than to predict the regular reservoir. The authors therefore tentatively applied kernel principal component analysis and support vector machine, called KPCA-SVM model, to solve this problem. Through the polynomial kernel function of the KPCA, the model can obtain the nonlinear feature extraction. Then the Gaussian kernel function in the SVM is chosen to perform optimization again. Finally, reservoir identification is implemented in the SVM. As the model incorporates the advantages of kernel function, principal component analysis and support vector classification, it can better solve the problem of nonlinear small sample, eliminate the noise of the data and reduce the dimension without missing valid information. In addition, it can achieve the prediction function quickly and accurately. The model was employed to predict the reservoir in x856 well block, which belongs to Xu2 member gas reservoir of the Xinchang gas field. The prediction results show the superiority of this model, which can be used as an optional method in tight reservoir prediction.

关 键 词：核主成分 支持向量机 KPCA-SVM模型 储层判别 新场须二气藏 

分 类 号：P631.4[天文地球—地质矿产勘探]