支持向量机组合方法在砂泥岩储层岩性识别中的应用

岩性是表征储油物性、建立各类地质模型的重要参数,其与测井参数的函数关系很复杂。基于支持向量机一类对一类法对岩性识别分类精度不高的现状,结合二叉树法初始分类精度较高、分类速度快等优点,提出了一种新的组合方法——一类对一类法与二叉树法的结合应用。该方法对样本数据较少的类别设置权重系数,减少样本不平衡的影响,并利用几何平均准确率作为评价岩心识别效果的指标,其对岩性的分类效果远优于单一方法。具体步骤为:首先对不均衡的样本设置相应的权重系数,然后利用二叉树法将易于与砂泥岩区分的灰岩区分开来,再利用一对一分类法将剩下的砂泥岩样本进行分类。运用此方法对某油田测井数据进行岩性分类,分类的整体准确率以及几何平均准确率均有很大的提高。

Lithology recognition by SVM in sand-shale reservoir

Rock lithology is the important parameter to characterize physical properties of reservoir and establish various geological models. Function relation between the lithology and the well-log parameter is complex. Based on low accuracy of one versus one (OVO) method and high accuracy and speed of binary tree (BT) method this paper proposes a new method to combine the two methods. The new method sets weight coefficients for lithologies with less specimen data to reduce influence of sample imbalance. The geometric mean accuracy is used as the index of evaluation of core lithology identification. Identification effect of the combined one is much better than that of the any single method. The operation steps are: 1 setting weight coefficient for the imbalanced samples; 2 identification of limestone from sandstone and mudstone by BT; 3 classification of the rest sandstone and mudstone by OVO. The combined method is applied to lithology classification of a oil field and the whole classification accuracy and geometric mean accuracy are highly improved.