论文部分内容阅读
作为经验模态分解(EMD)的改进型算法,完备总体经验模态分解(CEEMD)不但有效解决了EMD的模态混叠问题,同时也保留了EMD处理非平稳信号的优势,如自适应性、二进滤波特性等.CEEMD能自适应地将一个复杂信号分解为一系列本征模态函数(IMF)分量,且IMF分量满足从高频到低频系列分布,随机噪声往往分布在第一个或前几个高频IMF分量.考虑到地震信号的非平稳性和去噪方法对非平稳信号的适应性,针对CEEMD直接舍弃高频IMF分量去噪容易造成高频有效信息损失以及小波阈值去噪方法存在的不足,本文提出了一种基于CEEMD的小波阈值去噪方法.该方法首先引入自相关曲线判别出含噪较多的高频IMF分量,然后对CEEMD直接去噪要舍弃的这些含噪高频分量进行小波阈值降噪,以保留这些分量中的高频有效信息,最后与不含噪声的其他IMF分量一起重构原信号.模型和实际地震数据试算结果表明,该方法在显著提高地震数据信噪比的同时,能有效地保留原信号中的高频有效成分和弱信号信息,是一种相对保幅的有效去噪方法.
As an improved algorithm of empirical mode decomposition (EMD), complete empirical mode decomposition (CEEMD) not only effectively solves the EMD modal aliasing problem, but also preserves the advantages of EMD in dealing with non-stationary signals such as adaptive , Binary filtering characteristics, etc. CEEMD can adaptively decompose a complex signal into a series of IMF components, and the IMF components satisfy the distribution from high frequency to low frequency series, random noise is often distributed in the first Or the first few high-frequency IMF components.Considering the non-stationary seismic signal and the adaptability of denoising methods to non-stationary signals, eliminating the high-frequency IMF component directly for CEEMD is likely to cause loss of high-frequency effective information and wavelet threshold This paper proposes a wavelet threshold denoising method based on CEEMD.This method firstly introduces the autocorrelation curve to identify the high frequency IMF component with more noises and then rejects the CEEMD denoising The noise and high frequency components are denoised by the wavelet threshold to preserve the high frequency effective information of these components and finally reconstruct the original signal with other IMF components without noise.The model and actual seismic data trial calculation That this method significantly improve the signal to noise ratio of seismic data while effectively retain the active ingredient and a weak high-frequency signal in the original signal information, a web relative guarantee effective denoising.