一种基于相空间高斯混合模型的测井相识别方法

A method of well-log facies recognition based on a Gaussian mixture model in phase space

油层测井相识别是油田开发的重要依据.油层的测井相类别如果发生误判,将会影响它和邻井对应油层之间的连通对比关系的判断,造成开采方案的失败.测井相的类间差异小,领域知识要求高,属于难度较大的模式识别问题.在研究了混沌建模的基本方法后,给出了一种基于相空间高斯混合模型的测井相识别方法.对每类训练样本形成一个重构相空间,用EM算法学习得到一个高斯混合模型,待识序列按其属于不同类别的高斯混合模型的条件概率进行分类.该方法可以对不规则样本的形态参数进行估计,从而提高测井相的识别精度.实验结果证明了该方法的有效性.

Proper identification of facies of oil bearing layers in well-logs is crucial to oilfield development. Misjudged facies impair judgment of a layer＇s connective relationship with adjacent oil-layers, and can cause the failure of an oil field development plan. Well-log facies recognition is quite difficult because of the small differences among their classes and requires knowledge of multiple disciplines. This paper presents a method for well-log facies recognition based on a Gaussian mixture model in phase space after chaotic modeling methods are studied. For each class, a phase space is reconstructed with respect to each class and a Gaussian mixture model is learned by an EM algorithm. The sequence to be recognized can be classified according to the conditional probabilities belonging to the different Gaussian mixture models. This method can estimate the shape parameters of irregular samples so that the recognition rate of well-log facies increases. The effectiveness of this method has been experimentally demonstrated.