支持向量机方法在低阻油层流体识别中的应用

在H油田含油层系中，低阻油层与高阻水层并存，储层的岩性和孔隙结构复杂多变，粘土矿物普遍存在，在电性上直接区分油水层比较困难，常规测井解释符合率较低。为此，在认真分析工区低阻油层地质特征的基础上，引入模式识别领域应用较好的支持向量机（SVM）方法，探索了该方法在油水层划分中的应用。首先综合常规测井资料和试油资料构建57个油水层样本（学习样本34个，预测样本23个），并进行数据归一化处理，然后根据训练样本的分布和实验结果选择核函数类型，利用网格搜索寻优法得到模型最优参数，建立起低阻油层目的层段流体识别模型。利用该模型对23个预测样本进行识别，结果正确的有21个，预测精度达91．3％，其中2个误判样本是将油水同层判识为水层。对34个建模样本进行回判，准确率达100％。对某井4号层位（1476～1481m井段）射孔试油结果产油63．7m^2／d，无水，与预测结果一致。由此表明利用支持向量机方法对未知流体属性的正确识别是可行的。

Application of support vector machine method for identifying fluid in low-resistivity oil layers

In oil-bearing formations of H oilfield, low-resistivity oil layers and high-resistivity water layers co-exist, so it is difficult to directly distinguish oil layer from water layer by logging curves, making the conventional logging interpretation a low coincidence rate. On the basis of detailed analyzing geologic characteristics of low-resistivity oil layers, the support vector machine in pattern recognition domain was introduced and its application in distinguishing oil layer from water layer probed. Firstly, the conventional logging data was used in combination with the well testing data to select and normalize 57 oil layer and water layer samples, including 34 studying samples and 23 predicting samples. Then, the kernel function was determined by referring the distribution of training samples and the experimental comparison. Further more, the optimum parameters were identified by network searching to establish the fluid recognition model for low-resistivity oil layers. The model was used to distinguish 23 predicting samples. As a result, 21 samples were correctly distinguished and the accuracy degree was 91.3%. The other two oil/water-layer samples were interpreted as water layers, without any oil layer sample missed. The 34 samples were distinguished again, with an accuracy degree of 100%. The predicted result of No.4 layer (1 476~1 481 m) in one well was proved by perforation and well testing. Oil productivity is 63.7 m3/d without water. So it is feasible to correctly distinguish the uncertain fluid property by support vector machine method.