基于形态聚类的频率域大地电磁数据去噪新方法研究

Research of a new algorithm on denoising frequency domain magnetotelluric data based on morphological clustering

大地电磁法是一种利用天然电磁场源进行深部地电结构勘探的地球物理方法,其信号容易受到各种噪声干扰。为了更好地识别大地电磁的信号与噪声,提取出有用的信号进行阻抗估计,本文提出一种基于频率域大地电磁信号形态进行自动聚类的算法,并使用该算法对合成数据和实测数据进行计算。研究结果表明:在频率域中,大地电磁信号形态都有一定的相似性,强干扰噪声在其对应的频段上表现出与信号不一样的特征,可以使用余弦相似度提取此类特征,并使用聚类算法将信号与噪声分离;基于合成、实测数据,采用所提出的算法计算的视电阻率曲线和相位曲线都更加平滑、连续,说明经过本方法处理后的大地电磁数据阻抗估计结果得到了明显的改善。

The magnetotelluric(MT)method is a geophysical method that uses natural electromagnetic sources for deep geoelectric structure exploration,but its signals are susceptible to various noise interferences.To better identify the MT signals and noise,a method for automatic clustering was proposed based on the MT signal patterns in the frequency domain,and the algorithm was used to calculate with synthetic and measured data.The results show that in the frequency domain,MT signal patterns have some similarities,and strong interference noise has specific characteristics different from the signal.Such characteristics can be distinguished using cosine similarity,and the signal can be separated from the noise using a clustering algorithm.The apparent resistivity and phase curves calculated by the new algorithm are smoother and more continuous,indicating a significant improvement in MT impedance estimation by using the improved method.