论文部分内容阅读
消除噪声干扰对大地电磁测深资料的影响是大地电磁(MT)工作中的首要问题.基于结构风险最小化原则的支持向量机能够解决小样本情况下非线性函数拟合的通用性和推广性问题,是求复杂的非线性拟合函数的一种有效技术.本文首先介绍了ε-SVR(ε不敏感损失函数—支持向量机回归)的原理及SVR相关参数的选择,在此基础上,利用该算法对大地电磁测深实测资料进行回归处理,并与当前常用的去噪方法(Robust变换结合人工筛选)进行对比,结果表明ε-SVR算法可以较好地消除MT测深曲线所受噪声影响,提高工作效率.同时给原始数据加入10%噪声,并利用该算法对加噪后的数据进行回归处理,加噪前后拟合结果的绝对误差的均方差为9.454,相对误差的均方差为1.61%,证明该模型具有良好的泛化能力和稳健性.
Eliminating the influence of noise interference on the magnetotelluric sounding data is the most important issue in the earth magnetism (MT) work.The support vector machine based on the principle of structural risk minimization can solve the generality and generalization of nonlinear function fitting in the case of small samples Problem is to find a complex non-linear fitting function of an effective technique.This paper first introduces the principle of ε-SVR (ε insensitive loss function - support vector machine regression) and SVR-related parameters of choice, on this basis, The proposed method is applied to the regression analysis of the measured data from the earth’s electromagnetic sounding system and compared with the commonly used denoising method (Robust transform and manual screening). The results show that the ε-SVR algorithm can eliminate the noise of the MT sounding curve Influence and improve work efficiency.At the same time, add 10% noise to the original data, and use this algorithm to carry on regression processing to the noise-added data. The mean square error of the absolute error before and after adding noise is 9.454, the mean square error of relative error is 1.61%, proves that the model has good generalization ability and robustness.