一种用于测井油气层综合识别的支持向量机方法

On Support Vector Machines Method to Identify Oil & Gas Zone with Logging and Mudlog Information

提出了使用支持向量机分类技术在复杂储层中由常规测井、气测录井信息综合识别油气层类型的方法、步骤,重点阐述了输入特征选择、学习算法参数设置和在新疆油气田实际应用效果,并与人工神经网络、神经-模糊、最近邻分类器等3种机器学习模型识别检验结果进行了对比,检验结果表明由支持向量机所建立的油气水识别模型获得了最高的识别检验性能,其次是IncNN、FSM,而kNN相对最低,展示了支持向量机方法在复杂油气藏油气层类型识别应用中的高效性。

Support Vector Machines（SVM） algorithm is a new statistical learning technique which uses structural risk minimization inductive principle instead of empirical risk minimization principle. It has self-contained theory and its model is extended easily. We proposed Support Vector Machines Classifier to identify oil ＆ gas zone in complex reservoir with logging and mudlog data; introduced Support Vector Machines Classifier（SVMC） method and practical steps; detailed selection of input features, adjustment of SVMC learning parameters and applied cases in Xinjiang oilfield and also compared it with other 3 machine learning approaches （neural network, neural-fuzzy and nearest classifier）. The prediction results indicate that recognization capability of the SVMC method is the best, IncNN and FSM are better, kNN is bad. So, the SVMC method is very efficiency in application to recognize liquid type of complex reservoirs.