基于极端随机树算法的流体识别研究

Research on fluid recognition with extremely randomized trees algorithm

对储层内流体进行识别存在较大的不确定性,多属性融合进行流体识别显得非常必要。机器学习方法已经日趋成熟,但在流体识别方面的应用还较少见。这里引入一种实现简单、具有较强普适性的方法——极端随机树方法对流体进行识别,比较了本方法较传统的机器方法的识别准确率优势,并通过均方误差及错误率的验证证实了本方法对于流体识别的准确性。最后将本方法应用于南海某油气田,良好的识别效果证实了本方法对于流体识别的有效性。

Great uncertainty occurs when we make fluid identification in the reservoir,so multi-attribute fusion for fluid identification is very necessary.Machine learning methods are becoming mature but rarely used in fluid identification.This paper introduces an Extremely randomized trees(ET) algorithm for fluid identification,which is simple to implement and has strong universality.The advantages of this method and traditional machine methods are compared,and the accuracy of the method for fluid identification is confirmed by mean square error and error rate.Finally,the method is applied to actual data in the South China Sea,whose effectiveness for fluid identification is verified by the excellent result.