电磁法数据处理的奇异值分解法

本文提出用奇异值分解方法处理电磁响应数据.根据电磁响应数据的奇异值分布特征,可以指出有用信号和随机误差所对应的奇异值范围.截断随机误差所对应的奇异值,利用与有用信号对应的特征象重构数据,可以消除随机误差.对加入随机噪声的理论数据处理的结果表明,这种方法的效果明显,随机噪声的标准差可降低60％多.另外,利用不同区段的特征象重构数据,可区分不同级次的电磁响应特征.最后,给出对实测数据处理的结果.

ELECTROMAGNETIC DATA PROCESSING USING THE SINGULAR VALUE DECOMPOSITION

An essential part of the EM data processing is the rejection of the random noise. This paper presents a new approach which is based on the decomposition of the EM data matrix into eigenimages, using singular value decomposition. The distribution of the singular values in EM data indicates which singular values are related to the EM signal and which to the noise. By truncating the singular values associated with the noise, the data may be reconstructed from only a few of the first eigenimages associated with the signal, and the random noise can be rejected. The test of the method on synthetic data shows that the approach can reduce the random noise by more than sixty percent. A field example is presented.