基于支持向量机的交互检验储层预测

基于统计学习理论的支持向量机(support vector machine,SVM)是有限样本情况下的机器学习方法,具有严格的理论基础,能较好地解决小样本、非线性、高维数和局部极小点等问题。在地震储层预测中,影响支持向量机应用效果的主要因素在于惩罚因子及核函数参数的设置,其值设置过小或过大,都会使估计函数的泛化能力变差,降低储层预测精度。为提高支持向量机在储层预测中的应用效果,将已知样本随机划分为若干组,依次选其中的一组作为检验样本,其余样本作为学习样本,交互检验惩罚因子及核函数参数对储层预测精度的影响;优选惩罚因子及核函数参数,提高支持向量机储层预测精度。通过实际资料应用,验证了方法的有效性。

 A support vector machine (SVM) based on statistical learning theory is a machine learning method for a limited sample size.It has a strict theoretical basis and could also solve problems related to small samples,nonlinearity,high dimensions,and local minima.In seismic reservoir prediction,the main factor affecting the excellent performance of the SVM is the setting of the penalty factor and the kernel function parameter.If the value is set too small or too large,the generalization capability of the estimation function will degrade and the accuracy of reservoir prediction will be reduced.Therefore,the known samples were randomly divided into several groups,and one of them was selected as the test sample,and the remaining samples were used as the learning sample,to cross-validate the effect of the penalty factor and the kernel function parameter on the accuracy of reservoir prediction.Next,optimizing the selected penalty factor and kernel function parameters was carried out to improve the accuracy of reservoir prediction based on SVM.An application example is given to verify the effectiveness of the method.