|
| 支持向量机在砂泥岩储层岩性识别中的应用 |
| Lithology recognition by SVM in sand-shale |
| 投稿时间：2014-04-15 修订日期：2014-04-15 |
| DOI： |
| 中文关键词: SVM改进 岩性识别 多分类不平衡样本 |
| 英文关键词:SVM lithology recognition multiple classification unbalanced sample |
| 基金项目: |
|
| 摘要点击次数: 870 |
| 全文下载次数: 0 |
| 中文摘要: |
| 岩性是表征储油物性、建立各类地质模型的重要参数。因其与测井参数函数关系很复杂,本文提出了一种新的组合属性,对岩性的分类效果远优于单一属性。并利用支持向量方法(SVM)对岩性进行识别,针对岩性识别这一不平衡多分类问题提出了利用几何平均准确率作为评价岩心识别效果的指标。基于常规SVM一类对一类法(OVO)分类精度不高的现状,结合二叉树法的初始分类精度较高、分类速度快等优点,提出了对一类对一类法与二叉树法的结合应用,并对样本数据较少的类别设置适当的权值,减少样本不平衡的影响。首先对不均衡的样本设置相应的权重系数,然后使用二叉树法将易于与砂泥岩区分的灰岩区分开来,再使用一对一分类法对剩下砂泥岩样本进行分类。实例证明,此方法预测结果不仅是在分类的整体准确率还是几何平均准确率都有相应的提高,是可行的。 |
| 英文摘要: |
| Rock lithology are some properties which could reflect the characteristics of rocks, they are also important parameters which could establish geological model and describe reservoir properties. Due to the function with the logging parameters is complex, this paper proposes a new combination of attributes, lithology classification results far superior to a single attribute. The samples are identified by SVM, this paper proposes use the geometric mean recognition accuracy to evaluate the effect of core indicators. The accuracy based on one versus one method is not high status, we proposed a combined method to forecast them. Examples show that this method of prediction accuracy both in the overall classification accuracy rate and the geometric mean accuracy rate is raised, it is feasible. |
| 查看/发表评论 下载PDF阅读器 |
| 关闭 |
|
|
|
