基于CEEMDAN和SWT的地震信号随机噪声压制

Seismic random noise attenuation based on CEEMDAN and SWT

完备集合经验模态分解(CEEMD)通过添加正负成对辅助噪声可较好的解决集合经验模态分解(EEMD)中信号被噪声污染的问题,但CEEMD方法分解后的单个本征模态函数(IMF)分量中仍存在部分随机噪声信息.通过转变辅助噪声形式和分解流程提出自适应噪声完备集合经验模态分解(CEEMDAN)方法,该方法在较少集总次数和筛选迭代次数的情况下,即可实现优良的信噪分离功能,大大缩减处理耗时,具备分解精度高、具有完备性的特征.同时,针对传统经验模态分解(EMD)类方法去噪时直接舍弃第1~2阶高频IMF分量,导致其内高波数有效能量损失的问题,通过计算相邻IMF分量互信息熵获取高频噪声和低频有效信号的最优能量分界,对分界点前的各阶IMF分量进行同步压缩小波变换(SWT)处理,分离有效高频信息,最后与低频IMF分量重构达到噪声压制的目的.合成及实际地震资料处理结果表明,本文联合多步骤地震随机噪声压制策略具有较好的去噪效果和能量保持能力,在运算耗时指标上优于传统的EMD噪声辅助类方法.

Complete Ensemble Empirical Mode Decomposition(CEEMD)method solved the problem of noise pollution in Ensemble Empirical Mode Decomposition(EEMD)by adding positive and negative paired auxiliary noise.However,there is still some random noise in the Intrinsic Mode Function(IMF)component decomposed by the CEEMD method.By transforming the auxiliary noise form and decomposition process,we propose a Complete Ensemble Empirical Mode Decomposition with Adaptive Noise(CEEMDAN)method.The proposed method can achieve excellent signal-to-noise separation function in the case of a smaller number of realizations and sifting iterations,greatly reduce the computational cost,with high decomposition accuracy and complete features.At the same time,aiming at the problem that the first or sencond order IMF components are directly discarded by the traditional EMD method,which leads to the loss of the effective high-wavenumber energy.The optimal energy boundary of high-frequency noise and low-frequency effective signal is obtained by calculating the Mutual Information Entropy(MIE)of adjacent IMF components.The IMF components before the boundary point are processed by Synchrosqueezed Wavelet Transforms(SWT),the effective high-frequency information is separated,and finally the low-frequency IMF components are reconstructed to obtain the noise attenuation signal.The synthetic and field seismic records testing results indicate that the improved joint noise attenuation strategy has strong denoising effect and energy holding ability,and is superior to the traditional EMD noise-assisted methods in computational time-consuming indexes.