Research Of Magnetotelluric Signal-noise Separation Based On Multi-feature Parameters And Intelligent Classification

Due to the natural magnetotelluric signal is weak and easily susceptible to various types of electromagnetic noise pollution.At the same time,the complex and diverse types of magnetotelluric(MT)noise,and the energy is strong and the correlation is good.How to adaptively and accurately separate the weak and effective magnetotelluric signals from the strong interference in the ore district have become a very challenging task.In this paper,multi-feature parameters and intelligent classification methods are used to study the signal-noise separation of magnetotelluric signals.The main research is as follows:(1)Introducing approximate entropy and multi-scale entropy,and combined K-means clustering to accurately identify magnetotelluric data affected by strong interference;Noise suppression only for data segments identified as strong interference by using stagewise orthogonal matching pursuit.(2)Starting from the signal complexity of the magnetotelluric data,extract approximate entropy,fuzzy entropy,sample entropy,and LZ complexity for analysis;These four types of robust feature parameters are used as the input of support vector machine,and the sample library is trained to obtain a mathematical model for signal-noise identification;Signal-noise identification is performed on the measured magnetotelluric data,complementary ensemble empirical mode decomposition combined with wavelet threshold to perform noise suppression only on time series identified as strong interference.(3)A large number of sample libraries that conform to the characteristics of weak magnetotelluric signals and strong interference are constructed.The sample library is input to a one-dimensional convolutional neural network for training,and the corresponding training set and test set are obtained.The network and related training parameters are defined to obtain the training model;Using the obtained mathematical model for signal-noise identification of the magnetotelluric data,and noise suppression only for signal identified as strong interference by using wavelet threshold method.The above methods are processed through simulation experiments,Qinghai test point experiments,and measured data,and compared with traditional methods,the results shown that the proposed method can retain more low-frequency band information,and the apparent resistivity-phase curve is smoother and continuous,which provides a new solution for the accurate separation of strong interference in the ore district.