有导师学习的k-means方法在多支持向量机中的线性划分及其在石油储层识别中的应用

Linear Separation in Multi-Support Vector Machine with Supervised k-Means and its Applications in Oil Reservoir Recognition

提出了一种带有指导信息的k-means方法多支持向量机(SkSVM)。带有指导信息的k-means方法多支持向量机中k-means的目标是对训练数据进行划分,附加指导信息是保证k-means在对训练数据进行划分过程中确保数据的划分子集同时来自不同的2个类,即划分的子集中有正标签(+1)和负标签(-1)的数据,并且子集的中心尽量靠近不同类别的分离边界,保证提供的数据子集能够高效地为支持向量机进行学习。对每一个划分的子集采用支持向量机进行学习。选取UCI标准测试集对SkSVM和已有的FaLKSVM、SVM-KNN和CSVM算法进行对比测试。最后,用SkSVM、k-means和SVM等3种分类方法对江汉油田某区块oilsk81、oilsk83和oilsk85三口油井进行石油储层识别(油层和非油层),其中,各油井数据2/3作为训练数据,1/3作为测试数据,结果表明,在识别准确率上,SkSVM方法都优于其他两种方法。

We present a multi-Support Vector Machine with supervised k-means（SkSVM）.The main idea of SkSVM is as follows.The training dataset is divided into subsets by k-means method with additional guidance information and are trained by linear SVM independently.The additional guidance information ensures that data subsets are from two different classes at the same time,that is,apositive label（＋1）and negative label（-1）data;furthermore,the center of the subset is as close to the separation boundary as possible so that the SVM learns from each subset efficiently.We use UCI benchmark to test SkSVM and compare SkSVM with CSVM,SVM-KNN and CSVM.Then SkSVM,k-means and SVM classification methods are used to recognize oil reservoir（inferiority layer and oil layer）in oilsk81,oilsk83,oilsk85 wells in Jianghan oil field in China,where for each well,2/3data is used as training data and the rest datais used as test data.The results show that the SkSVM method gain higher recognition accuracy than the other two methods.