一种加权的支持向量机及其在储层识别中的应用

A Weighted Support Vector Machine and Its Application in Reservoir Recognition

标准支持向量机(SVM)抗噪声能力不强,当训练样本中存在有噪声或者野点时,会影响最优分类面的产生,最终导致分类结果出现偏差。针对这一问题,提出了一种考虑最小包围球的加权支持向量机(WSVM),给每个样本点赋予不同的权值,以此来降低噪声或野点对分类结果的影响。对江汉油田某区块的oilsk81,oilsk83和oilsk85三口油井的测井数据进行交叉验证,其中核函数采用了线性、指数和RBF这3种不同的核函数。测试结果显示,无论是在SVM还是在WSVM中,核函数选择RBF识别率都是最高的,同时提出的WSVM不受核函数的影响,识别稳定性好,且在交叉验证中识别率都能够达到100%。

Support vector machine（SVM） is sensitive to noises and it will influence the output of optimal classification face when the training sample contain noise or outliers,ultimately lead to the deviation of the classification results.In order to deal with these problem,this paper proposes a weighted support vector machine（WSVM） for considering the minimum surrounded ball.The WSVM reduces the effects of noise or outliers on classification by assigning different weights to each sample.This paper executes three different tests with linear,exponential and RBF kernel function.Cross validation simulation on log data of three wells（oilsk81, oilsk83 and oilsk85 in Jianghan Oilfield） shows that,either in SVM or in WSVM, the recognition rate will be the highest when selecting the RBF kernel function. Meanwhile, the WSVM in this paper works stably and will not be affected by the kernel function, in addition, its recognition rate can reach 100% in all cross validation.