基于贝叶斯证据框架下SVM的油层识别模型研究

Oil layer recognition model based on SVM within Bayesian evidence framework

支持向量机(support vector machine,SVM)方法在石油测井领域的油层识别中取得了很好的应用效果,但SVM方法的识别效果受到惩罚参数和核参数的影响,不同的参数组合直接影响识别精度的优劣。为了在油层识别中获得更好的识别效果,提出一种基于贝叶斯证据框架下SVM的油层识别模型,即根据测井数据的训练样本信息,采用贝叶斯证据框架的理论求解惩罚参数以及核参数,再通过所求得的决策函数对测井数据的测试样本进行识别。实际测井数据实验表明,基于贝叶斯证据框架下SVM的油层识别模型的油层识别效果得到提高,优于传统SVM方法和基于粒子群优化算法(particle swarm optimization,PSO)的SVM方法。

Support Vector Machine（ SVM） method is successfully applied in the petroleum logging field to recognize oil layer. But the recognition effect is influenced by penalty parameter and kernel parameter,and the recognition accuracy is affected directly by different combinations of parameters. In order to get better results in oil layer recognition,an oil layer recognition model based on SVM in Bayesian evidence framework is proposed,which is according to sample information in logging training. The penalty parameter and kernel parameter can be solved by Bayesian evidence framework theory firstly and then the test sample is recognized through decision function. The experiment results of actual logging datum show that the recognition effect is improved by Bayesian-SVM and the recognition effect is superior to that of the traditional SVM and PSO-SVM.