论文详情
根据大地电磁噪声的特点和独立分量分析(ICA)中M-FastICA算法的优良性能,结合小波分析和盲源分离的相关理论,提出了一种改进的独立分量分析去噪方法。首先对观测信号进行多尺度小波分解,使信号从单道变成多道,以满足独立分量分析对观测信号的数目需求;然后采用M-FastICA算法对小波分解提取的多层高频分量进行独立分量分析以提取有效独立分量和特定独立分量;引入动态自适应因子来限制特定独立分量的权重以减小观测信号信噪比对去噪效果的影响;最后由小波低频分量和M-FastICA算法提取的两种独立分量共同构成恢复信号。模拟信号仿真实验表明,该方法的去噪性能优于传统小波阈值去噪方法。将该方法应用于实际大地电磁观测资料的去噪处理,无论是视电阻率曲线还是相位曲线,都比去噪前更加光滑和稳定,说明改进的独立分量分析算法能有效地去除大地电磁噪声。
Combined with the wavelet analysis theory and blind source separation,an improved independent component analysis algorithm for magnetotelluric data denoising is proposed,which is based on the M-FastICA algorithm,an improvement on the FastICA algorithm.First,the multi-scale wavelet decomposition is performed to convert the observed signal from single channel to multiple channels,to meet the quantity demand of the independent component analysis.Then,the M-FastICA algorithm is adopted to process the multi-layer high frequency components extracted by wavelet decomposition.This process extracts the effective independent components and specific independent components.Then,the dynamic adaptive factor is introduced to limit the weight of the specific independent components and reduce the SNR influence in the observed data on the denoising process.Finally,the low frequency components of the wavelet and two types of independent components extracted using the M-FastICA algorithm together constitute the recovery signal.Simulation test results on an analog signal show that the denoising performance of the proposed method is better than the traditional wavelet threshold denoising method.The application test on field magnetotelluric observation data shows that both the apparent resistivity and phase curves are smoother and more stable than that before denoising.These results illustrate that the proposed method can effectively remove magnetotelluric noise.
国家自然科学基金(41274082,U1562109)、长江大学长江青年基金(2015cqn76)、长江大学重磁电勘探研究中心创新基金(7011201803xm)联合资助。