摘要
作者针对目前沉积微相识别中的特征提取问题 ,提出了应用 SVM(支持向量机 )方法进行沉积微相识别的方案。该方法不是象传统方法那样首先试图将原输入空间降维 (即特征选择变换 ) ,而是设法将输入空间升维 ,以求在高维空间中问题变得线性可分 (或接近线性可分 )。因为升维后只是改变了内积运算 ,并没有使算法复杂性随着维数的增加而增加 ,因此这种方法才是可行的。所以 ,利用该方法我们可以不必将很大的精力集中于特征的提取中 ,而是借助于算法的内在特征提取能力 ,使得该方法更能胜任实际情况。实际处理表明该方法在小样本情况下 ,性能远优于神经网络 。
Aiming at the problem of pattern recognition in sedimentary facies analysis, we put forward a scheme which apply SVM to sedimentary facies recognition. Unlike traditional method try to reduce the dimension of input space(i.e. characters selection and transformation), SVM increase dimension of input space to ensure it is Linearly Separable in high dimension space. The method is feasible because it only changes inner product operation and the complexity of algorithm doesn't increase. Using SVM we needn't wasting time on character extraction but resort to the intrinsic character extraction ability which makes it more suitable in practical instance. The result of practical application indicates that the performance of SVM had superiority over RBFNN and overcome the problem of overfitting excellently.
出处
《物探化探计算技术》
CAS
CSCD
2000年第2期158-164,共7页
Computing Techniques For Geophysical and Geochemical Exploration
基金
中国石油天然气总公司"九五"攻关课题项目!(KG950 22)
关键词
统计学习理论
沉积
模式识别
沉积微相
SVM方法
Statistical Learning Theory
Support Vector Machine
Machine Learning
Pattern Recognition
RBF Neural Networks
Sedimentary Facies
