基于LMD-SVD的微震信号降噪方法研究

作为一种3维、实时的监测手段，微震监测通过分析岩体破裂产生的微震信号，评估工程岩体的稳定性，为工程建设和人员安全提供预警。然而，工程现场情况复杂，采集微震信号时通常会混入一定程度的噪声，影响后续微震信号的分析工作。针对这一问题，提出一种基于局部均值分解（local mean decomposition，LMD）和奇异值分解（singular value decomposition，SVD）的LMD-SVD联合降噪法以降低噪声干扰。该方法首先使用LMD分解，获得一系列由高频到低频分布的乘积函数（product functions，PF）；通过计算原始信号与各个PF分量之间的相关系数，确定含噪信号与有效信号之间的分界位置，将分界分量之前的分量剔除，实现初步降噪。然后，针对LMD分解结果中的残留噪声，使用SVD法，以加权能量贡献率（percent of contribution to total energy，PCTE）作为奇异值阶数的确定方法，对分界PF分量进行降噪处理，实现二次滤波。通过上述处理，最终实现微震信号降噪。在仿真实验中，对于同一带噪的Ricker子波，分别使用经验模态分解（empirical mode decomposition， EMD）、LMD、LMD-SVD这3种方法进行降噪处理。其降噪前后信号的信噪比、波形图及频谱图对比结果表明LMD-SVD是一种更好的降噪方法。此外，对于白鹤滩水电站左岸地下厂房的微震监测系统所采集的信号，运用LMD-SVD对含噪微震信号进行降噪处理，表明本文方法能够有效地去除微震信号中的高频噪声，为后续微震分析工作提供帮助。

Research on Microseismic Signal Denoising Method Based on LMD-SVD

As the real and three-dimension monitoring method, microseismic monitoring can be utilized to evaluate engineering rockmass stability and offer early security warning to people. However, affected by the complicated engineering site conditions, microseismic signals are usually mixed with some noises, which would influence the subsequent microseismic analyses. To reduce noise based on local mean decomposition (LMD) and singular value decomposition (SVD), a new LMD-SVD denoising method was proposed. This new method firstly decomposed noisy microseismic signals by LMD to obtain a series of product functions (PF) from low frequency to high frequency distribution. By calculating the correlation coefficient (CC) between the original signal and each PF component, boundary PF component between the effective signal and noise was obtained. The PF component (components) before the boundary component is(are) eliminated to achieve the first step of noise reduction. Subsequently, in order to further reduce residual noise existing in the LMD decomposition results, SVD method was introduced and corresponding singular value order was determined by utilizing percent of contribution to total energy (PCTE) to complete the second step of noise reduction. Final denoised signal would be obtained through the above denoised processes. In the simulation experiments of this study, empirical mode decomposition (EMD), LMD, LMD-SVD were respectively taken to reduce noise existing in a noisy Ricker wavelet. The results of signal-to-noise ratio (SNR), waveform and spectrogram before and after noise reduction presented that LMD-SVD was a better method for noise reduction compared with other two methods. In addition, the proposed LMD-SVD method was applied to denoise noisy microseismic signals collected from microseismic monitoring system of the underground powerhouse on the left bank of Baihetan Hydropower Station. The denoised results show that LMD-SVD can effectively remove high frequency noise existing in the microseismic signals and facilitate subsequent microseismic analysis works.