【摘要】 支持向量机 (SVM)是近年来发展起来的一种通用的机器学习方法 ,在许多分类问题和函数拟合问题上都已获得了很好的效果。对于少量样本的分类问题 ,SVM具有调节参数较少 ,运算速度快等优点。通过地震、测井等信息进行油气预测是一种典型的非线性分类器设计问题 ,它具有已知样本数较少、特征个数较少等特点 ,文章据此提出了一种基于特征扩展和特征选择的改进SVM方法。该方法将原始特征通过非线性变换到高维空间 ,然后应用线性SVM进行特征选择 ,并同时计算降维过程中各个特征子集对应的留一法错误率 ,最后选择错误率较小的特征子集来设计线性SVM分类器。在通用数据的实验中 ,这种方法仅仅用较为简单的多项式核函数就大大提高了分类器的泛化能力。与传统的模糊数学方法、神经网络方法和SVM方法相比 ,这种方法在四川观音场构造的碳酸岩盐储层数据的预测误差降低了 5 0 % ,是一种有效的油气预测方法。更多还原

【Abstract】 Support Vector Machine (SVM) is a general-purpose machine learning method developed in recent years,by which good results have been obtained in many classification and function-fitting problems. As for the classification of a small amount of samples,SVM has many advantages,such as a few adjusted parameters and fast arithmetic speed,etc. The hydrocarbon prediction by means of the seismic and log data is a typical nonlinear classificator desigh problem and it is characterized by a small amount of the number of samples and of the number of features. For this reason,an improved SVM method based on feature expansion and feature selection is proposed in the paper. This method includes to change the original features to a high-dimensional space through nonlinear transformation;to make,then,a feature selection by use of linear SVM method;to calculate simultaneously the leave-one-out error rate corresponding with each feature subset in the process of decreasing dimensions;and to design,finally,the linear SVM classificator by use of the feature subset with the smallest error rate. In the experiment of general-purpose data,the generalization ability of the classificator might be greatly raised only by use of a simple polynomial kernel function in the method. As compared with fuzzy mathematical method and neural network method,the SVM method could decrease the prediction error of the carbonate reservoir data of Guanyinchang structure in Sichuan by 50%,therefore it is an effective hydrocarbon prediction method.更多还原