论文部分内容阅读
多属性分析是支持向量机的一个重要功能。它可以基于同一目标的多个属性进行综合分析并得出反映该事物属性特征的最佳判别规则,是解决识别问题的一种重要方法。相对于传统的支持向量机,近似支持向量机(PSVM)在大数据量训练集的处理效率上具有较明显的优势且准确度高,适合处理信息量丰富的测井属性组合。一般情况下,采用两种属性值做交会图并通过数据的分布差异来定性区分流体或岩性的做法较为常见,也可利用拟合等方式来实现简单的定量划分。但这些方式均在精度上有所欠缺,且不能正确划分相互重合区域的两类数据。本文讨论了基于近似支持向量机的流体识别方法。通过利用流体识别因子或各测井曲线值作为训练集,并求解基于等式约束的最优化问题得到最优决策函数,可在短时间内获取属性值分布上的学习经验。该方法在数据点相互重合区域的识别效果上有一定的改善。将该方法用于某地龙王庙组地层的气水识别,能够实现较高的准确率。
Multi-attribute analysis is an important function of SVM. It can make a comprehensive analysis based on multiple attributes of the same target and obtain the best discriminant rules that reflect the attributes of the thing. It is an important method to solve the problem of identification. Compared with the traditional support vector machines, approximate support vector machine (PSVM) has obvious advantages and high accuracy in the processing efficiency of large data volume training set, and is suitable for processing well-logging combination with abundant information. In general, it is more common to use two attribute values to make a cross-over plot and to qualitatively differentiate fluids or lithologies through the distribution of data. Simple quantitative demarcation can also be achieved by using a fitting method. However, these methods are deficient in accuracy and can not correctly classify two types of data that overlap with each other. This paper discusses the method of fluid identification based on approximate support vector machines. By using the fluid identification factor or log values as the training set and solving the optimization problem based on equal constraints, the optimal decision function can be obtained, and the learning experience on attribute value distribution can be obtained in a short time. The method has some improvement on the recognition effect of overlapped data points. This method can be used to identify gas and water in Longwangmiao formation of a certain place, which can achieve high accuracy.